10 Shattuck St, Boston MA 02115 | (617) 432-2136
| |
Copyright © 2020 President and Fellows of Harvard College. All rights reserved.
Office: 919.785.1304 | Fax: 919.785.1306
Biomedical research is the broad area of science that looks for ways to prevent and treat diseases that cause illness and death in people and in animals. This general field of research includes many areas of both the life and physical sciences.
Utilizing biotechnology techniques, biomedical researchers study biological processes and diseases with the ultimate goal of developing effective treatments and cures. Biomedical research is an evolutionary process requiring careful experimentation by many scientists, including biologists and chemists. Discovery of new medicines and therapies requires careful scientific experimentation, development, and evaluation.
Why are Animals Used in Biomedical Research?
The use of animals in some types of research is essential to the development of new and more effective methods for diagnosing and treating diseases that affect both humans and animals. Scientists use animals to learn more about health problems, and to assure the safety of new medical treatments. Medical researchers need to understand health problems before they can develop ways to treat them. Some diseases and health problems involve processes that can only be studied in living organisms. Animals are necessary to medical research because it is impractical or unethical to use humans.
Animals make good research subjects for a variety of reasons. Animals are biologically similar to humans. They are susceptible to many of the same health problems, and they have short life-cycles so they can easily be studied throughout their whole life-span or across several generations. In addition, scientists can easily control the environment around animals (diet, temperature, lighting), which would be difficult to do with people. Finally, a primary reason why animals are used is that most people feel it would be wrong to deliberately expose human beings to health risks in order to observe the course of a disease.
Animals are used in research to develop drugs and medical procedures to treat diseases. Scientists may discover such drugs and procedures using alternative research methods that do not involve animals. If the new therapy seems promising, it is tested in animals to see whether it seems to be safe and effective. If the results of the animal studies are good, then human volunteers are asked to participate in a clinical trial. The animal studies are conducted first to give medical researchers a better idea of what benefits and complications they are likely to see in humans.
A variety of animals provide very useful models for the study of diseases afflicting both animals and humans. However, approximately 95 percent of research animals in the United States are rats, mice, and other rodents bred specifically for laboratory research. Dogs, cats, and primates account for less than one percent of all the animals used in research.
Those working in the field of biomedical research have a duty to conduct research in a manner that is humane, appropriate, and judicious. CBRA supports adherence to standards of care developed by scientific and professional organizations, and compliance with governmental regulations for the use of animals in research.
Scientists continue to look for ways to reduce the numbers of animals needed to obtain valid results, refine experimental techniques, and replace animals with other research methods whenever feasible.
© California Biomedical Research Association
Clinical research is the study of health and illness in people. It is the way we learn how to prevent, diagnose and treat illness. Clinical research describes many different elements of scientific investigation. Simply put, it involves human participants and helps translate basic research (done in labs) into new treatments and information to benefit patients. Clinical trials as well as research in epidemiology, physiology and pathophysiology, health services, education, outcomes and mental health can all fall under the clinical research umbrella.
A clinical trial is a type of clinical research study. A clinical trial is an experiment designed to answer specific questions about possible new treatments or new ways of using existing (known) treatments. Clinical trials are done to determine whether new drugs or treatments are safe and effective. Clinical trials are part of a long, careful process which may take many years to complete. First, doctors study a new treatment in the lab. Then they often study the treatment in animals. If a new treatment shows promise, doctors then test the treatment in people via a clinical trial.
People often confuse a clinical research or clinical trials with medical care. This topic can be especially confusing if your doctor is also the researcher. When you receive medical care from your own doctor, he or she develops a plan of care just for you. When you take part in a clinical research study, you and the researcher must follow a set plan called the “study protocol.” The researcher usually can’t adjust the plan for you – but the plan includes steps to follow if you aren’t doing well. It’s important to understand that a clinical trial is an experiment. By its nature, that means the answer to the research question is still unknown. You might or might not benefit directly by participating in a clinical research study. It is important to talk about this topic with your doctor/the researcher.
Medical research has become an important part of the health care industry, and advances in technology have made it possible for much of it to be done on an outpatient basis, meaning that investigators sometimes don’t need to do extensive studies in a research facility. The field of medical research is one of the most interesting fields in all of science because of its focus on science and medicine in conjunction with a great deal of research.
Medical research covers a wide variety of studies, stretching from ‘baseline’ investigation, through systematic reviews and to the cutting-edge of medical science. It involves the study of all human diseases or may have only disease-specific research. For example, AIDS research includes both studying patients who have AIDS and those who do not, as well as studying children with AIDS and children without AIDS. Similarly, researchers may be investigating the causes of Parkinson’s disease in old age and Parkinson’s disease in young adulthood.
4 Phases of Medical Research Studies
The four phases of Medical Research Studies are experimental, comparative/expository, understudy, and last, analysis and validation/regression.
There are many reasons why medical research is so valuable. Whether you aim to start a career in this field or to gain more knowledge about health conditions and their treatments, it’s important to understand the benefits that medical research provides. You can learn about medical research at http://hrmdresearch.com/ .
The Importance Of Medical Research
The breakthroughs that people enjoy today are virtually unimaginable without the knowledge gained through medical research. Here are some of the most important reasons medical research is important:
Medical research helps people learn more about themselves and their health. The knowledge gained by medical research is constantly improving.
2. Development Of New Drugs
Medical research must be done to find a cure for diseases and illnesses. Without medical research, medicine and other medical innovations as we know it could not exist. Sometimes called pharmaceutical research, medical research encompasses a broad spectrum of scientific studies. It starts with the research and development of drugs, followed by treatments and procedures used in clinical practice.
The process of new drug development may involve the following steps:
3. Improve The Quality Of Life
Medical researchers don’t just look for ways to manage the symptoms of diseases, they try their best to find a cure for a specific illness or a group of diseases. There are several ways drug research and testing improve the quality of life:
Why is medical research so important? Medical research saves lives every day. Scientists and researchers work day and night to develop new treatments, drugs, and procedures. Without the help of dedicated scientists, doctors, and other medical professionals, the advances made would be slow and limited.
When a patient participates in a trial, they must undergo several physical tests and provide some information about their lifestyle and diet. The findings and insights derived from these studies and trials are invaluable in coming up with treatments and cures for various health conditions.
Nih clinical research trials and you.
The NIH Clinical Trials and You website is a resource for people who want to learn more about clinical trials. By expanding the below questions, you can read answers to common questions about taking part in a clinical trial.
Clinical research is medical research that involves people like you. When you volunteer to take part in clinical research, you help doctors and researchers learn more about disease and improve health care for people in the future. Clinical research includes all research that involves people. Types of clinical research include:
Clinical trials are part of clinical research and at the heart of all medical advances. Clinical trials look at new ways to prevent, detect, or treat disease. Clinical trials can study:
The goal of clinical trials is to determine if these treatment, prevention, and behavior approaches are safe and effective. People take part in clinical trials for many reasons. Healthy volunteers say they take part to help others and to contribute to moving science forward. People with an illness or disease also take part to help others, but also to possibly receive the newest treatment and to have added (or extra) care and attention from the clinical trial staff. Clinical trials offer hope for many people and a chance to help researchers find better treatments for others in the future
People may experience the same disease differently. It’s essential that clinical trials include people with a variety of lived experiences and living conditions, as well as characteristics like race and ethnicity, age, sex, and sexual orientation, so that all communities benefit from scientific advances.
See Diversity & Inclusion in Clinical Trials for more information.
The idea for a clinical trial often starts in the lab. After researchers test new treatments or procedures in the lab and in animals, the most promising treatments are moved into clinical trials. As new treatments move through a series of steps called phases, more information is gained about the treatment, its risks, and its effectiveness.
Clinical trials follow a plan known as a protocol. The protocol is carefully designed to balance the potential benefits and risks to participants, and answer specific research questions. A protocol describes the following:
A clinical trial is led by a principal investigator (PI). Members of the research team regularly monitor the participants’ health to determine the study’s safety and effectiveness.
Most, but not all, clinical trials in the United States are approved and monitored by an Institutional Review Board (IRB) to ensure that the risks are reduced and are outweighed by potential benefits. IRBs are committees that are responsible for reviewing research in order to protect the rights and safety of people who take part in research, both before the research starts and as it proceeds. You should ask the sponsor or research coordinator whether the research you are thinking about joining was reviewed by an IRB.
Clinical trial sponsors may be people, institutions, companies, government agencies, or other organizations that are responsible for initiating, managing or financing the clinical trial, but do not conduct the research.
Informed consent is the process of providing you with key information about a research study before you decide whether to accept the offer to take part. The process of informed consent continues throughout the study. To help you decide whether to take part, members of the research team explain the details of the study. If you do not understand English, a translator or interpreter may be provided. The research team provides an informed consent document that includes details about the study, such as its purpose, how long it’s expected to last, tests or procedures that will be done as part of the research, and who to contact for further information. The informed consent document also explains risks and potential benefits. You can then decide whether to sign the document. Taking part in a clinical trial is voluntary and you can leave the study at any time.
There are different types of clinical trials.
Clinical trials are conducted in a series of steps called “phases.” Each phase has a different purpose and helps researchers answer different questions.
In clinical trials that compare a new product or therapy with another that already exists, researchers try to determine if the new one is as good, or better than, the existing one. In some studies, you may be assigned to receive a placebo (an inactive product that resembles the test product, but without its treatment value).
Comparing a new product with a placebo can be the fastest and most reliable way to show the new product’s effectiveness. However, placebos are not used if you would be put at risk — particularly in the study of treatments for serious illnesses — by not having effective therapy. You will be told if placebos are used in the study before entering a trial.
Randomization is the process by which treatments are assigned to participants by chance rather than by choice. This is done to avoid any bias in assigning volunteers to get one treatment or another. The effects of each treatment are compared at specific points during a trial. If one treatment is found superior, the trial is stopped so that the most volunteers receive the more beneficial treatment. This video helps explain randomization for all clinical trials .
" Blinded " (or " masked ") studies are designed to prevent members of the research team and study participants from influencing the results. Blinding allows the collection of scientifically accurate data. In single-blind (" single-masked ") studies, you are not told what is being given, but the research team knows. In a double-blind study, neither you nor the research team are told what you are given; only the pharmacist knows. Members of the research team are not told which participants are receiving which treatment, in order to reduce bias. If medically necessary, however, it is always possible to find out which treatment you are receiving.
Many different types of people take part in clinical trials. Some are healthy, while others may have illnesses. Research procedures with healthy volunteers are designed to develop new knowledge, not to provide direct benefit to those taking part. Healthy volunteers have always played an important role in research.
Healthy volunteers are needed for several reasons. When developing a new technique, such as a blood test or imaging device, healthy volunteers help define the limits of "normal." These volunteers are the baseline against which patient groups are compared and are often matched to patients on factors such as age, gender, or family relationship. They receive the same tests, procedures, or drugs the patient group receives. Researchers learn about the disease process by comparing the patient group to the healthy volunteers.
Factors like how much of your time is needed, discomfort you may feel, or risk involved depends on the trial. While some require minimal amounts of time and effort, other studies may require a major commitment of your time and effort, and may involve some discomfort. The research procedure(s) may also carry some risk. The informed consent process for healthy volunteers includes a detailed discussion of the study's procedures and tests and their risks.
A patient volunteer has a known health problem and takes part in research to better understand, diagnose, or treat that disease or condition. Research with a patient volunteer helps develop new knowledge. Depending on the stage of knowledge about the disease or condition, these procedures may or may not benefit the study participants.
Patients may volunteer for studies similar to those in which healthy volunteers take part. These studies involve drugs, devices, or treatments designed to prevent,or treat disease. Although these studies may provide direct benefit to patient volunteers, the main aim is to prove, by scientific means, the effects and limitations of the experimental treatment. Therefore, some patient groups may serve as a baseline for comparison by not taking the test drug, or by receiving test doses of the drug large enough only to show that it is present, but not at a level that can treat the condition.
Researchers follow clinical trials guidelines when deciding who can participate, in a study. These guidelines are called Inclusion/Exclusion Criteria . Factors that allow you to take part in a clinical trial are called "inclusion criteria." Those that exclude or prevent participation are "exclusion criteria." These criteria are based on factors such as age, gender, the type and stage of a disease, treatment history, and other medical conditions. Before joining a clinical trial, you must provide information that allows the research team to determine whether or not you can take part in the study safely. Some research studies seek participants with illnesses or conditions to be studied in the clinical trial, while others need healthy volunteers. Inclusion and exclusion criteria are not used to reject people personally. Instead, the criteria are used to identify appropriate participants and keep them safe, and to help ensure that researchers can find new information they need.
Clinical trials may involve risk, as can routine medical care and the activities of daily living. When weighing the risks of research, you can think about these important factors:
Most clinical trials pose the risk of minor discomfort, which lasts only a short time. However, some study participants experience complications that require medical attention. In rare cases, participants have been seriously injured or have died of complications resulting from their participation in trials of experimental treatments. The specific risks associated with a research protocol are described in detail in the informed consent document, which participants are asked to consider and sign before participating in research. Also, a member of the research team will explain the study and answer any questions about the study. Before deciding to participate, carefully consider risks and possible benefits.
Well-designed and well-executed clinical trials provide the best approach for you to:
Risks to taking part in clinical trials include the following:
If you are thinking about taking part in a clinical trial, you should feel free to ask any questions or bring up any issues concerning the trial at any time. The following suggestions may give you some ideas as you think about your own questions.
This information courtesy of Cancer.gov.
The goal of clinical research is to develop knowledge that improves human health or increases understanding of human biology. People who take part in clinical research make it possible for this to occur. The path to finding out if a new drug is safe or effective is to test it on patients in clinical trials. The purpose of ethical guidelines is both to protect patients and healthy volunteers, and to preserve the integrity of the science.
Informed consent is the process of learning the key facts about a clinical trial before deciding whether to participate. The process of providing information to participants continues throughout the study. To help you decide whether to take part, members of the research team explain the study. The research team provides an informed consent document, which includes such details about the study as its purpose, duration, required procedures, and who to contact for various purposes. The informed consent document also explains risks and potential benefits.
If you decide to enroll in the trial, you will need to sign the informed consent document. You are free to withdraw from the study at any time.
Most, but not all, clinical trials in the United States are approved and monitored by an Institutional Review Board (IRB) to ensure that the risks are minimal when compared with potential benefits. An IRB is an independent committee that consists of physicians, statisticians, and members of the community who ensure that clinical trials are ethical and that the rights of participants are protected. You should ask the sponsor or research coordinator whether the research you are considering participating in was reviewed by an IRB.
For more information about research protections, see:
For more information on participants’ privacy and confidentiality, see:
For more information about research protections, see: About Research Participation
After a clinical trial is completed, the researchers carefully examine information collected during the study before making decisions about the meaning of the findings and about the need for further testing. After a phase I or II trial, the researchers decide whether to move on to the next phase or to stop testing the treatment or procedure because it was unsafe or not effective. When a phase III trial is completed, the researchers examine the information and decide whether the results have medical importance.
Results from clinical trials are often published in peer-reviewed scientific journals. Peer review is a process by which experts review the report before it is published to ensure that the analysis and conclusions are sound. If the results are particularly important, they may be featured in the news, and discussed at scientific meetings and by patient advocacy groups before or after they are published in a scientific journal. Once a new approach has been proven safe and effective in a clinical trial, it may become a new standard of medical practice.
Ask the research team members if the study results have been or will be published. Published study results are also available by searching for the study's official name or Protocol ID number in the National Library of Medicine's PubMed® database .
Only through clinical research can we gain insights and answers about the safety and effectiveness of treatments and procedures. Groundbreaking scientific advances in the present and the past were possible only because of participation of volunteers, both healthy and those with an illness, in clinical research. Clinical research requires complex and rigorous testing in collaboration with communities that are affected by the disease. As research opens new doors to finding ways to diagnose, prevent, treat, or cure disease and disability, clinical trial participation is essential to help us find the answers.
This page last reviewed on October 3, 2022
Patient discussion about research.
Q. Where can I get updates about new researches on fibromyalgia? I have fibromyalgia and I would like to know all there is to know and see if they found new breakthroughs on the subject. A. I use this site: http://www.nfra.net/ the main issue of the site is Fibromyalgia research…
Q. I need to do an interview with someone with knowledge on lupus for a research paper any takers? a couple of questions should do it. it doesn't have to be extensive. A. I HAVE SLE AND A FUW MORE THANS THAT ARE KNOW TO BE KNOW TO COME FROM HAVEING SLE LUPUS I AM NOT 100% OF ALL THAT COMES WITH SLE BUT I AM WILLING TO TELL U ALL I KNOW THANK YOU
Q. i have heard that number of scientists found out in one of there researches that breasts Cancer is capable to just disappear with out a treatment , have any one read this article/research ? or maybe just heard about it ? because it is interesting why and how this result happens ... A. hi pinkofdestiny - try also these links, i know and read a lot about the books of Phillip Day and recommend them to everybody. cancer can be healed and there are also ways to make with success prevention: http://www.credencegroup.co.uk/Eclub/ses/sessearch.php?q=breast+cancer&pvdc=0 before a woman should loose her breast, she should make a therapy with vitamine B17 - the vitamin which can eliminate cancer in any form, but you should not know about it! it is terrible, but it is the way how politicians and industry-trust treat us.
Published 01 Jun 2022 - Updated 17 Mar 2023
Every treatment, intervention, medication, way of care, and aftercare in the medical field or health care system came from discoveries. This high quality of care we can experience today was not discovered overnight, but rather through years of effort by medical professionals who investigated the risk factors, causes, preventions, and treatments of diseases. This type of investigation is known as medical/health research.
The general definition of research is, 'an investigation that is intentionally designed to help develop or contribute to knowledge'. When you add a medical purpose to 'research', the general definition stays the same, but the goal becomes more specific. Ultimately, the goal shifts to a focus on increasing medical knowledge, improving patient care, developing new medicines or procedures, and enhancing the already existing medicines and procedures.
There are several forms of medical research being conducted today. Here are 3 common forms:
Why is research important in medicine? The simple answer is that medical research has led to many medical breakthroughs and developments. It would also strongly contribute to shaping the future of medicine.
Here's how:
Medical research has led to the development of diagnostic tools and technologies that allow for earlier and more accurate diagnoses of diseases.
For instance, breast cancer is one of the most common cancers worldwide. Medical research led to the development of an effective screening method known as mammography which has resulted in earlier detection and a 20% fall in mortality rates.
Another example is the development of pap smears for the early diagnosis of cervical cancer. This as well as caused a significant decrease in late presentation and mortality rates due to cervical cancer.
A host of other effective screening methods have been developed as a result of medical research such as genetic testing, imaging techniques, and so on.
Medical research has led to the development of new treatments for a wide range of diseases, such as cancer, allergies, HIV/AIDS, heart disease, and so on.
Research is essential to find out what treatments work best, and more specifically what treatments work best for what patient. It can provide important information about how effective a medical intervention is and its possible adverse effects. These interventions include drugs, vaccines, medical devices, and others.
By being specific with participant requirements, medical professionals can study how certain groups of people react to certain treatments . An example of this can be seen here at Infiuss Health. As a CRO in Africa, we at Infiuss Health focus on the demographics of the continent to ensure people of African ancestry receive effective care.
Medical research would lead to newer developments in medicine such as personalized medicine and targeted therapies, that would ensure that each individual would have treatment options unique to them. Increasing research in this area is the only way to make this a reality in the future of medicine.
Medical research has contributed to the prevention of diseases such as polio, smallpox, and measles which caused the deaths of millions of people in the past.
Recently, following the Covid-19 pandemic, medical research led to the development of vaccines that gradually slowed down the progress of the disease.
Medical research has contributed to our understanding of public health issues and how to address them.
A typical example was in 1854 when there was an outbreak of cholera in the Golden Square Area in London. An Anaesthesiologist known as John Snow conducted an epidemiological study and found that the source of contamination was a public pump. When the contaminated pump was closed from public access, the outbreak of cholera ended.
Research provides important information about disease trends and risk factors, outcomes of treatment or public health interventions, functional abilities, patterns of care, and health care costs and use.
Economists have found that medical research can have an enormous impact on the quality of healthcare which in turn affects human health and longevity.
Healthy individuals tend to be more productive and that contributes greatly to the national economy. If the research enterprise is impeded, or if it is less robust, important societal interests are affected.
Covid-19 vaccine development, for example, contributed to the lifting of the lockdown in many countries and allowed individuals to resume work.
Compared to treatment, current research on disease prevention shows that preventive services are able to significantly reduce deaths and illnesses at reasonable costs. All of these findings have informed and influenced national budget planning and policy decisions.
The simple fact is that clinical research improves our lives. It leads to significant discoveries, improves health care, and ensures that patients receive the best care possible. It is what makes the development of new medicines and treatments possible, without it we would not be able to move forward in the development of medicine.
Infiuss Health, as a CRO in Africa, aims to make it easier to do more clinical trials/ medical research in Africa by use of technology and other means.
When you support, participate in, or conduct medical research, you are helping to continue to build the future of medicine.
Infiuss Health insights contains inspiring thought leadership on health issues and the future of health data management and new research.
Warning: The NCBI web site requires JavaScript to function. more...
An official website of the United States government
The .gov means it's official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you're on a federal government site.
The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Institute of Medicine (US) Clinical Research Roundtable; Tunis S, Korn A, Ommaya A, editors. The Role of Purchasers and Payers in the Clinical Research Enterprise: Workshop Summary. Washington (DC): National Academies Press (US); 2002.
(Clinical Research: A National Call to Action, November 1999) Clinical research is a component of medical and health research intended to produce knowledge valuable for understanding human disease, preventing and treating illness, and promoting health. Clinical Research embraces a continuum of studies involving interactions with patients, diagnostic clinical materials or data, or populations in any of the following categories: (1) disease mechanisms (etiopathogenesis); (2) bi-directional integrative (translational) research; (3) clinical knowledge, detection, diagnosis and natural history of disease; (4) therapeutic interventions including development and clinical trials of drugs, biologics, devices, and instruments; (5) prevention (primary and secondary) and health promotion; (6) behavioral research; (7) health services research, including outcomes, and cost-effectiveness; (8) epidemiology; and (9) community-based and managed care-based trials.
Sponsors include private and public sector funding organizations such as the National Institutes of Health, pharmaceutical companies, medical device manufacturers, biotechnology firms, universities, private foundations, and national societies. Within the public sector the National Institutes of Health (NIH) is the largest clinical research sponsor, followed by the Department of Defense (DOD), the Department of Veterans Affairs (VA), Agency for Healthcare Research and Quality (AHRQ), and the Centers for Disease Control (CDC).
Research organizations include academic health centers, private research institutes, survey research organizations, federal government intramural research programs, and contract research organizations.
Investigators are the scientists performing clinical research from varied disciplines with a range of academic qualifications (e.g., MD, Ph.D., RN, DDS, PharmD).
Participants are the human volunteers, medical information and biological materials of human origin, or data derived from volunteers. Participants may have particular health conditions or may be healthy volunteers or populations at large.
Oversight entities include Institutional Review Boards, Food and Drug Administration, Department of Health and Human Services, Veterans Affairs, National Committee for Quality Assurance, and other national regulatory agencies.
Stakeholders/Consumers include health insurers, managed care organizations, health care systems, organized medicine, voluntary health agencies, patient advocacy groups, purchasers of health care, and providers of health care, public health systems, and individual consumers.
Recent activity.
Your browsing activity is empty.
Activity recording is turned off.
Turn recording back on
Connect with NLM
National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894
Web Policies FOIA HHS Vulnerability Disclosure
Help Accessibility Careers
Advarra Clinical Research Network
Onsemble Community
Diversity, Equity, & Inclusion
Participants & Advocacy
Advarra Partner Network
Community Advocacy
Institutional Review Board (IRB)
Institutional Biosafety Committee (IBC)
Data Monitoring Committee (DMC)
Endpoint Adjudication (EAC)
GxP Services
Research Compliance & Site Operations
Coverage Analysis
Budget Negotiation
Calendar Build
Technology Staffing
Clinical Research Site Training
Research-Ready Training
Custom eLearning
Sponsors & CROs
Sites & Institutions
BioPharma & Medical Tech
Oncology Research
Decentralized Clinical Trials
Early Phase Research
Research Staffing
Cell and Gene Therapy
Solutions for need.
Clinical Trial Management
Clinical Data Management
Research Administration
Study Startup
Site Support and Engagement
Secure Document Exchange
Solutions for Sites
Enterprise Institution CTMS
Health System/Network/Site CTMS
Electronic Consenting System
eSource and Electronic Data Capture
eRegulatory Management System
Research ROI Reporting
Automated Participant Payments
Solutions for Sponsors/CROs
Clinical Research Experience Technology
Center for IRB Intelligence
Resource library.
White Papers
Case Studies
Ask the Experts
Frequently Asked Questions
COVID-19 Support
Consulting Opportunities
Leadership Team
Our Experts
Accreditation & Compliance
Learn more about our company team, careers, and values. Join Advarra’s Talented team to take on engaging work in a dynamic environment.
We all know there are numerous acronyms and abbreviations used in clinical research. While some can be easily deciphered, others may take some searching to find their meaning. Particularly with the recent surge in electronic systems and regulations, it can be hard to keep track of necessary abbreviations and terms.
Whether you’re new to the clinical research world or need a refresher, here’s a condensed list of common acronyms and abbreviations you may come across.
AACI: Association of American Cancer Institutes
AAHRPP: Association for the Accreditation of Human Research Protection Programs
ABSA: Association of Biosafety and Biosecurity
ACRP: Association of Clinical Research Professionals
ACTS: Association for Clinical and Translational Science
ADME: Absorption, Distribution, Metabolism, and Elimination
ADR: Adverse Drug Reaction
AE: Adverse Event
ALCOA: Attributable, Legible, Contemporaneous, Original, Accurate
AMC: Academic Medical Center
API: Active Pharmaceutical Ingredient
API: Application Program Interface
ARO: Academic Research Organization
ASCO: American Society of Clinical Oncology
ASCPT: American Society for Clinical Pharmacology and Therapeutics
ASGCT: American Society of Gene & Cell Therapy
BA/BE: Bioavailability/Bioequivalance
BLA: Biological Licensing Application
BSM: Biospecimen Management
caBIG: Cancer Biomedical
CAPA: Corrective and Preventive Action
CAR-T: Chimeric Antigen Receptors and T cells
CBER: Center for Biologics Evaluation and Research
CBRN: California Board of Registered Nursing
CCEA: Complete, Consistent, Enduring, Available
CCOP: Community Clinical Oncology Program
CCR: Center for Cancer Research
CCSG: Cancer Center Support Grant
CCTO or CTO: Centralized Clinical Trials Office or Clinical Trials Office
CDASH: Clinical Data Acquisition Standards Harmonization
CDER: Center for Drug Evaluation and Research
CDM: Clinical Data Management
Related article: “Improve Data Quality with 5 Fundamentals of Clinical Data Management”
CDP: Clinical Development Plan
CDRH: Center for Devices and Radiological Health
CDS: Clinical Data System
CDUS: Clinical Data Update System
CFR: Code of Federal Regulations
CMO: Contract Manufacturing Organization
CMS: Centers for Medicare & Medicaid Services
CRA: Clinical Research Associate
CRC: Clinical Research Coordinator
Related article: “Deciphering the CRC Career Path: Key Skills and Responsibilities”
CRF: Case Report Form
CRMS: Clinical Research Management System
CRO: Contract Research Organization
CRPC: Clinical Research Process Content
CSO: Contract Safety Organization
CSR: Clinical Study Report
CTA: Clinical Trial Authorization
CTCAE: Common Terminology Criteria for Adverse Events
CTMS: Clinical Trial Management System
CTRP: Clinical Trials Reporting Program
CTSA: Clinical and Translational Science Award
DDI: Drug-Drug Interaction
DHHS: Department of Health and Human Services
DM: Data Manager
DMC: Data Monitoring Committee
DSMB: Data and Safety Monitoring Board
EC: Ethics Committee
eCOA: Electronic Clinical Outcome Assessment
eCRF: Electronic Case Report Form
EDC: Electronic Data Capture
Learn more about Advarra’s electronic data capture system, Advarra EDC .
EHR: Electronic Health Record
EMR: Electronic Medical Record
ePRO: Electronic Patient-Reported Outcomes
eTMF: Electronic Trial Master File
FAIR: Findable, Accessible, Interoperable, Reusable
FDA: Food and Drug Administration
FE: Food Effect
FIH: First In Human
FWA: Federalwide Assurance
GCP: Good Clinical Practice
GCRC: General Clinical Research Center
GDP: Good Documentation Practice
GLP: Good Laboratory Practice
GMP: Good Manufacturing Practice
GVP: Good Pharmacovigilance Practice
HIPAA: Health Insurance Portability and Accountability Act
HRPP: Human Research Protection Program
IBC: Institutional Biosafety Committee
ICF: Informed Consent Form
ICH: International Council for Harmonization
IDE: Investigational Device Exemptions
IEC: Independent Ethics Committee
IHCRA: In House Clinical Research Associate
IIT: Investigator Initiated Trial
IND: Investigational New Drug (Application)
IP: Investigational Product
IRB: Institutional Review Board
ITT: Intent to Treat
IVRS: Interactive Voice Response System
IWRS: Interactive Web Response System
LTFU: Long Term Follow Up
LRAA: Local Regulatory Affairs Associate
MAC: Medicare Administrative Contractor
MAD: Multiple Ascending Dose
MCA: Medicare Coverage Analysis
Related webinar: Build a Better Budget: Using Medicare Coverage Analysis to Streamline Study Startup .
MRN: Medical Record Number
NCI: National Cancer Institute
NDA: New Drug Application
NHV: Normal Healthy Volunteer
NIH: National Institutes of Health
NLM: National Library of Medicine
OCT: Office of Clinical Trials
OHRP: Office for Human Research Protections
OSR: Outside Safety Report
PAC: Post Approval Commitments
PC: Protocol Coordinator
PD: Protocol Director
PHI: Protected Health Information
PI: Principal Investigator
PK/PD: Pharmacokinetic/Pharmacodynamic
PRE: Prompt Reporting Event
PRMC: Protocol Review and Monitoring Committee
PRMS: Protocol Review and Monitoring System
QC: Quality Control
QCT: Qualifying Clinical Trial
QMS: Quality Management System
SAD: Single Ascending Dose
SAE: Serious Adverse Event
SC: Study Coordinator
SDR: Source Document Review (Also Source Data Review)
SDTM: Study Data Tabulation Model
SDV: Source Document Verification
SIF: Site Investigator File
SMO: Site Management Organization
SOC: Standard of Care
SOE: Schedule of Events
SOP: Standard Operating Procedure
Related article: “Data Collection in Clinical Trials: 4 Steps for Creating an SOP”
SPOREs: Specialized Programs for Research Excellence
SRB: Scientific Review Board
SRC: Scientific Review Committee
SUSAR: Suspected Unexpected Serious Adverse Reaction
SVT: Subject Visit Template
TMF: Trial Master File (also eTMF)
TMO: Trial Management Organization
UADE: Unanticipated Adverse Device Effect
UADR: Unexpected Adverse Drug Reaction
UAP: Unanticipated Problem
Is there an abbreviation or acronym you see regularly in clinical research that isn’t listed here? Let us know .
To learn more about the fundamentals of clinical research, check out these free resources:
Tagged in: CRO , institutions , sites , sponsors
Meghan Hosely creates educational content for Advarra, such as blogs, eBooks, white papers, and more.
Back to Resources
Subscribe to our monthly email.
Receive updates monthly about webinars for CEUs, white papers, podcasts, and more.
Advancing Clinical Research: Safer, Smarter, Faster ™
Copyright © 2024 Advarra. All Rights Reserved.
You have full access to this open access article
27 Accesses
Explore all metrics
The success of current deep learning models depends on a large number of precise labels. However, in the field of medical image segmentation, acquiring precise labels is labor-intensive and time-consuming. Hence, the challenge of achieving a high-performance model via datasets containing noisy labels has attracted significant research interest. Some existing methods are unable to exclude samples containing noisy labels and some methods still have high requirements on datasets. To solve this problem, we propose a noisy label learning method for medical image segmentation using a mixture of high and low quality labels based on the architecture of mean teacher. Firstly, considering the teacher model’s capacity to aggregate all previously learned information following each training step, we propose to leverage a teacher model to correct noisy label adaptively during the training phase. Secondly, to enhance the model’s robustness, we propose to infuse feature perturbations into the student model. This strategy aims to bolster the model’s ability to handle variations in input data and improve its resilience to noisy labels. Finally, we simulate noisy labels by destroying labels in two medical image datasets: the Automated Cardiac Diagnosis Challenge (ACDC) dataset and the 3D Left Atrium (LA) dataset. Experiments show that the proposed method demonstrates considerable effectiveness. With a noisy ratio of 0.8, compared with other methods, the mean Dice score of our proposed method is improved by 2.58% and 0.31% on ACDC and LA datasets, respectively.
Avoid common mistakes on your manuscript.
Background. Currently, the technology based on machine learning and deep learning is constantly advancing and evolving, and has been successfully applied to many fields [ 1 , 2 ], including medical image analysis [ 3 ]. These applications include the assessment of diseased and normal images, as well as the detection and segmentation of diseased organs. However, the effectiveness of convolutional neural networks (CNNs) based on supervised learning relies heavily on precise annotation. Unlike natural images, the features of different types of medical images have different meanings. That is, it is necessary to collect their corresponding accurate pixel-level labels for medical image segmentation [ 4 ]. However, due to the poor interpretability, pixel-level labeling of medical images often necessitates expert annotation and consumes a substantial amount of time. In contrast, rough labeling of the target not only does not require a large number of professionals, but also takes significantly less time, thereby reducing labeling costs.
Challenges. Based on the above, we naturally expect the segmentation model could achieve excellent performance based on rough annotations. However, the learning process of traditional supervised learning methods completely trusts the given labels [ 5 ]. The model learns noisy labels and clean labels indiscriminately, which makes it difficult to learn accurate information. Previously, noisy label learning has been well studied in classification tasks [ 6 , 7 ]. However, in the segmentation task, each pixel has its corresponding label. The huge number of labels makes it more difficult to the study of noisy label learning.
Especially for medical image segmentation, the accuracy of target edge segmentation is higher. And target edges are also the most likely to be mislabeled. Therefore, it is worth studying how to obtain good segmentation performance in the case of rough labeling.
For the noisy label learning in segmentation task, researchers have done some research on the loss function and the structure of the model. Zhang et al. [ 8 ] proposed a noise-aware loss (NAL), which reduces the impact of noisy labels on the network performance. In addition to studying the robust loss function, Wang et al. [ 9 ] proposed a method called meta corrupted pixels mining based on a simple network. And to improve the tolerance of the model to noisy labels, some studies employ multiple networks for alternate learning and selectively choose clean samples for training [ 10 , 11 ]. Fang et al. [ 12 ] not only leverage the cooperation between two models but also incorporate an enhanced consistency constraint. This constraint facilitates the extraction of specific reliable knowledge from each model and integrates it into the other model. Such methods mitigate the impact of noisy labeled data by selecting clean data to learn. While these methods pay significant attention to clean data, there remains underutilization of data containing noisy labels.
Interestingly, the researchers found that during the learning process, CNNs tend to learn clean samples in the early stage, and fit wrong samples in the later stage. So in order to use all the samples, Liu et al. [ 13 ] made use of this learning feature and proposed the model named ADELE to correct pixel-level noisy labels while training. However, the learning mode that relies on a single network to correct noisy labels while training is easy to cause the accumulation of errors. And the study found that for the same dataset containing noisy labels, it can achieve better results by treating low quality (LQ) and high quality (HQ) labels separately [ 14 ]. Inspired by semi-supervised learning, some work applies the branch architecture to noisy label learning. Samples with LQ and HQ labels enter different branches, and the branches guide each other to learn [ 15 ]. The methods of differentiating between HQ and LQ labels have yielded promising results. These methods not only maximize the utilization of information within HQ labels but also effectively captures pertinent information from LQ labels. However, in practical scenarios, acquiring a dataset capable of distinguishing between HQ and LQ labels proves challenging. Consequently, these methods put additional constraints on the datasets.
Our contributions. In order to solve the above problems, we propose a noisy label learning method based on mixed learning of HQ and LQ samples. According to reference [ 13 ], we know that models tend to fit clean labels first during the learning process. Using this property, the model can correct the noisy label while training. However, relying on a single network is too easy to accumulate errors.
Inspired by mean teacher [ 16 ], the weight of the teacher model is obtained from the exponential moving average (EMA) of the student model. Based on this, we believe that the teacher model is closer to the “early-learning" stage than the student model. Therefore, to avoid error accumulation, we add a teacher model to guide the correction of noisy labels during training.
In computer vision, image-level perturbation of input images is the most common way of data enhancement [ 18 ]. The disturbed image is similar to the original image but not exactly the same. Common methods of image perturbation include Gaussian blur, reduction, rotation, inversion and grayscale. However, these methods are completely limited to the image level and only increase the diversity of the training set through small geometric transformations or color transformations. With the development of semi-supervised learning, some work has also explored the positive effects of feature level perturbations on models [ 19 ]. Compared with image perturbation, feature perturbation provides a wider disturbance space and can enhance the stability of the model. Noisy label learning is similar to semi-supervised learning in that they both carry imperfect labels. So we introduce feature perturbations into our work.
Therefore, our proposed method is an architecture based on early-learning and MT for mixing high-low quality samples on noisy label learning tasks. Our main contributions are as follows.
We propose a new noisy label learning method based on early-learning and mean teacher. We do not distinguish between HQ and LQ samples, which makes it more compatible with various datasets with noisy labels. According to the early learning characteristics of CNNs, we propose to use teacher model to correct the noisy labels in datasets when conditions are met.
Inspired by the unified dual-stream perturbations method in the semi-supervised field, the feature perturbation is added to the student model to enhance the robustness of the framework, so that it is not easy to fit the noisy labels.
We conducted experiments on the Automated Cardiac Diagnosis Challenge (ACDC) dataset and the 3D Left Atrium (LA) dataset. The results show that our method outperforms the methods that do not distinguish between high and low quality samples.
Medical image segmentation plays a crucial role in the analysis of medical imagery and serves as a vital technology for assisting diagnostic procedures. It accurately extracts regions of interest from the background, such as organs, tumors, and blood vessels.
The success of CNNs has also driven the development of medical image segmentation. U-Net [ 20 ], one of the most classical model, was proposed in 2015. Characterized by its distinctive U-shaped architecture, it can extract both deep and superficial features from images with remarkable effectiveness. As a result, U-Net [ 20 ] and its variants [ 21 , 22 , 23 ] have undergone extensive research and have been widely implemented in the domain of medical image segmentation. Building on the Transformer’s triumph in natural language processing, some studies [ 24 , 25 , 26 ] have integrated it with U-Net to enhance the precision of target area segmentation. Yuan et al. [ 27 ] fully exploited the characteristics of Transformer and CNNs, thereby proposed a network combining the complementary features of the two frameworks. By fusing features from the distinct domains of CNNs and Transformers, the method enhances feature representation capabilities and demonstrates robust performance across various medical image segmentation datasets. Nevertheless, the impressive performance of models based on supervised learning is contingent upon the availability of precise pixel-level annotations.
Given that medical image labels are harder to acquire compared to those of natural images, there is an increased likelihood of encountering incomplete or incorrect annotations. To address the issue of missing labels, researchers have developed solutions that involve designing semi-supervised learning architectures [ 14 , 28 , 29 ]. Because the target of medical image is not clear, the pixels in the edge area of the segmentation target are more likely to be classified incorrectly. Due to the often ambiguous nature of targets in medical imaging, pixels located in the border regions of the segmentation target are more susceptible to misclassification. To solve this problem, noisy label learning for medical image segmentation has attracted much attention from researchers [ 12 , 30 , 31 ]. In our study, we propose an effective noisy label learning method for medical image segmentation.
With the rapid development of deep neural networks, models trained with precise label supervision have attained remarkable performance. Generating a large number of precise labels is a time-consuming and labor-intensive process, particularly in specialized domains like medical imaging, where expert professionals are required to create these annotations. Therefore, noisy label learning has gradually been paid attention by researchers [ 32 , 33 ]. At present, the related work on noisy label learning for classification tasks is relatively comprehensive.
Initially, researchers aimed to accommodate noisy labels by modifying the architecture and loss functions of CNNs. Goldberger et al. [ 34 ] proposed to add an extra softmax layer to the end of their model to characterize the noisy labels. Likewise, a noise layer, whose parameters are initially unknown, is integrated into the design, allowing these parameters to be learned concurrently with the neural network’s parameters during training [ 35 ]. In order to accommodate to noisy label learning, a generalized cross entropy (GCE) loss function [ 36 ] was proposed. Ma et al. [ 37 ] also conducted an examination into the robustness of loss functions that are widely employed. Furthermore, the active passive loss was proposed, which consists of two robust loss functions that can promote each other. While these methods can suppress simple noisy labels, such as symmetric ones, their capacity to withstand intricate and irregular noisy labels remains comparatively limited.
To minimize the impact of samples with noisy labels on the training process, researchers aim to refine sample selection methods that encourage the model to predominantly learn from clean labels. The Co-teaching [ 10 ] involves the collaborative training of two neural networks simultaneously. Each of the two networks selects the samples with lower loss with respect to the other network in order to effectively address the issue of noisy labels. Song et al. [ 38 ] proposed a two-step training process for their model to prevent overfitting on data that contains noisy labels. In the initial phase, all samples are utilized to update the network and a subset of clean samples are identified. Subsequently, in the later phase, the model exclusively employs the samples deemed clean for continued training. These methods effectively leverage clean labeled data to enhance model performance. However, they overlook the valuable information embedded within noisy labeled data.
To acquire a larger, cleaner dataset for training in scenarios where datasets have noisy labels, the LongReMix [ 39 ] was introduced. This method first obtains a clean sample set, and then samples and oversamples it to increase the size of the clean sample set. In order to make full use of samples, Zheng et al. [ 40 ] proposed a method of meta-label correction. This approach treats the rectification of noisy labels as a meta-learning task, wherein a meta-model is employed to generate refined labels for the primary model. Research has indicated that the application of contrastive learning and hybrid attention mechanisms [ 41 ] can effectively mitigate the impact of label noise and subsequently enhance the overall performance of the model.
Image segmentation is indeed a more complex task compared to image classification, it requires labels for all pixels in an image. The large number of labels also makes it more prone to producing incorrect annotations. Noisy label learning in image segmentation is a critical area of research because noisy, inaccurate, or mislabeled data can significantly degrade the performance of segmentation models.
Intuitively, similar to classification tasks, researchers aim for models to discern clean labels within noisy datasets. Hence, references [ 10 , 38 ] and etc. are also applicable to image segmentation. Taking inspiration from the Co-teaching [ 10 ], Zhang et al. [ 11 ] introduced an innovative “Tri-network" learning framework to better deal with pixel-level noisy labels. This framework encompasses three networks that cyclically assume the role of a teacher, guiding the learning process of the other networks. Inspired by Co-learning [ 42 ], Fang et al. [ 12 ] proposed an algorithm that enables two models to engage in complementary co-training, thereby improving their learning efficiency and accuracy. In this method, a sophisticated distillation module is designed to purify superior-quality knowledge and enhance the use of reliable knowledge. This kind of sample selection method mitigates the influence of inaccurate labels on the model. Nevertheless, Liu et al. [ 43 ] explored the influence of noisy labels at different layers in supervised model training and affirmed that noisy labels contain valuable information. Therefore, the sample selection methods concurrently necessitate the sacrifice of valuable insights that might be embedded within noisy labels.
Given the significant correlation between pixel-level labels in image segmentation, researchers extensively exploit this information. To leverage the inter-pixel information for the purpose of mitigating the effect of noisy labels, Guo et al. proposed a joint class-affinity segmentation (JCAS) framework [ 44 ]. This method rigorously exploits the inherent affinity relationships that exist intra-class and inter-class to enhance the accuracy and robustness of the model. According to Li et al., it is posited that in the context of image segmentation, noisy labels are predominantly encountered proximal to the boundaries of objects [ 45 ]. Therefore, a robust learning strategy guided by superpixels is proposed. Guo et al. [ 46 ] propose a noise-label tolerant SAC-Net and use this network for two-stage training. While designing a structure resilient to noisy labels can enhance their tolerance, their adverse impact on the training process persists.
For noisy label learning in medical image segmentation, researchers have also done related studies. For the fetal brain tissue segmentation task, Karimi et al. [ 47 ] developed a new label smoothing procedure and loss function to train a model. This method properly considers the uncertainty of the target boundary. To improve the robustness of noisy label learning models for retinal blood vessel segmentation, Zhou et al. [ 48 ] developed the study group learning (SGL) method. In datasets annotated by various non-expert individuals, a method using two coupled networks was proposed. The first network is devised to estimate the true segmentation probability and the other to model the annotation propensity by estimating a pixel-by-pixel confusion matrix. Similarly, Liao et al. [ 49 ] proposed preference-involved annotation distribution learning distribution learning (PADL) framework. In addition to the above methods, Liu et al. [ 13 ] used the learning characteristics of deep learning to propose the method to gradually correct pixel-level noisy labels during the training process. To improve the accuracy of correcting noisy labels, Liu et al. [ 50 ] proposed contour information as an auxiliary tool to correct inaccuracies in pixel-level labels. Depending solely on a single model to correct noisy labels during training can lead to error accumulation. We introduce a teacher model to provide guidance in correcting noisy labels, thereby reducing the likelihood of error accumulation.
Inspired by previous works, researchers have proposed work on noisy label learning based on teacher-student structure. Zhang et al. [ 51 ] developed a framework for medical image segmentation that effectively mitigates the detrimental effects of noisy labels by engaging a novel teacher-student network architecture. The teacher network is initially trained using the original dataset, notwithstanding the presence of noisy labels. Subsequently, it leverages confidence learning coupled with spatial label smoothing regularization to refine the noisy annotations. The student network is then trained on these purified labels. This approach systematically diminishes the influence of noisy labels on the model’s performance. Nevertheless, the model needs to go through two stages of training, which increases the training time. In addition, in references [ 14 , 15 , 52 ], strong annotations and weak annotations within the dataset are discretely processed, ensuring that the information contained within the strongly annotated segment is exhaustively extracted and employed. However, these methods increase the requirements of the datasets. In practice, acquiring a dataset capable of effectively discerning between HQ and LQ labels poses a significant challenge. So, our proposed method adopts a teacher-student architecture for mixing HQ and LQ samples to approximate real-world datasets with noisy labels.
The framework of our proposed method. The mixture of HQ and LQ samples is simultaneously fed into the student model and teacher model. The student model generates two outputs through feature perturbation. FP denotes the feature perturbation. The prediction of teacher model and the original label are processed by confidence learning and label refinement to yield the denoised label. When certain conditions are met, part of the pixel-level annotations in the original label are corrected
To adapt to a wider range of datasets with noisy labels, we propose a noisy label learning method with mixed learning of HQ and LQ samples, as shown in Fig. 1 .
Unlike some methods [ 14 , 15 ] for noisy label learning, our proposed method is designed to operate even when the HQ samples are unknown. Here, we assume that the entire training set consists of M samples, of which N samples belong to the samples labeled with HQ, and the remaining \((M-N)\) samples is labeled with LQ. For the sake of clarity and convenience, we will denote the samples labeled with high quality as \(\phi _h = {(\textbf{X}_{(i)}, \textbf{Y}_{h(i)})}^N_{i=1}\) , where \(\textbf{X}_i\) represents an image and \(\textbf{Y}_{h(i)}\) represents the HQ label corresponding to \(\textbf{X}_i\) . Similarly, samples labeled with LQ are represented as \(\phi _l = {(\textbf{X}_{(i)}, \textbf{Y}_{l(i)})}^{M}_{i=N+1}\) , where \(\textbf{Y}_{l(i)}\) is the label with LQ. In addition, \(\textbf{X}_{(i)} \in \mathbb {R}^{\Omega _i}\) , and \(\textbf{Y}_{h(i)}, \textbf{Y}_{l(i)} \in \{0,1\}^{\Omega _i}\) .
In a realistic scenario, it becomes challenging to distinguish between samples with HQ and LQ labeling if no specific treatment is applied. Therefore, We mix \(\phi _h\) and \(\phi _l\) as \(\phi _{mix} = {(\textbf{X}_{(i)}, \textbf{Y}_{mix(i)})}^M_{i=1}\) in random order during training. In this way, the samples of HQ and LQ are not distinguished, and are respectively sent into the teacher and student models. As with the MT, a small perturbation noise \(\zeta \) is added to the image when it is fed into the teacher model.
Previous research [ 13 , 53 ] has demonstrated that leveraging the learning characteristics of deep learning can aid in correcting noisy labels within the dataset during training, thereby maximizing the utilization of all the samples. However, relying solely on a single network to simultaneously train and correct may lead to error accumulation. Inspired by multi-model learning methods such as Co-teaching [ 10 ], we hope to make use of the early-stage learning characteristics of the model and make another model guide the noisy label learning process. In the field of semi-supervised learning, mean teacher [ 17 ] has proven to be an effective architecture. It consists of two models: a student model and a teacher model.
The weight of the teacher model is determined through the EMA of the student model, and the specific update strategy is introduced in Sect. Teacher-guided correction with early learning . The teacher model aggregates all the previously learned information as soon as each step is completed. In this way, the output quality of all layers is improved, and the model can better represent the middle and even high-level semantic information. Based on the update strategy of the teacher model, we can infer that the update of teacher model will be slower than that of student model.
From the view of early learning, the model first memorizes clean labels and then memorizes noisy labels [ 13 , 53 ]. Therefore, after a period of training, the teacher model remembers more clean labels than the student model for the same epoch. Therefore, we use the teacher model to generated the prediction and correct noisy labels in the dataset. To be specific, the prediction of teacher model and original labels are input into the confident learning module, and the original noisy labels are corrected by label refinement.
To enhance the learning capability of the student model, feature perturbation is incorporated. This involves intentionally adding noise or variations to the input features, which encourages the student model to generate two sets of predictions based on the perturbed features. The consistency loss is then computed between these two sets of predictions.
Liu et al. [ 13 ] propose to use the inherent characteristics of neural networks, namely “early-learning", to address the issue of noisy label correction during the learning process. However, this method employs a single network for training and correcting the noisy labels. Therefore, inspired on the early-learning, we propose a novel medical image segmentation method that utilizes teacher-guided correction for learning with mixed HQ and LQ data.
(1) Early-learning based on teacher model
The weight of teacher model is obtained by the EMA of the student model. Assuming that the current training epoch is t , the weight of the student model is expressed as \(\theta _t\) , and the weight of the previous epoch is expressed as \(\theta _{t-1}\) . Then the weight of teacher model at the t epoch \(\theta '_{t}\) is calculated as shown in Eq. 1 :
where \(\beta \) is the decay rate and set to 0.99 according to [ 17 ]. \(\theta _{t-1}^{\prime }\) is the weight of teacher model at \(t-1\) epoch.
As illustrated in Fig. 1 , we aim to leverage the teacher model as the “third party" [ 15 ] to guide the identification the noisy labels and generate the refined labels. Therefore, the original labels and the refined labels jointly guide the learning of the student model.
Previous works [ 13 , 53 ] have indicated that models prioritize fitting clean labels during the learning process with noisy labels. From Eq. 1 , it can be seen that the weights of the teacher model aggregate the previously learned information. Therefore, the teacher model is less likely to overfit to noisy labels compared to the student model. Based on this, we propose that teacher model to guide the correction of noisy labels. It is worth noting that the output of the teacher model only serves as a guide to correct noisy labels and does not participate in the updating process of the student model.
In order to guide the correction of noisy labels more effectively, we use confident learning (CL) [ 54 ] to identify error labels at the pixel level and correct the original labels. In this way, we accomplish the objective of reducing noisy labels during the training process.
(2) Identification of pixel-level noisy labels
To identify errors in labels, Northcutt et al. [ 54 ] proposed a belief learning method. Through the estimation of the joint distribution matrix between the noisy labels and the ground truth, the noise in the labels is robustly represented.
Based on the prediction \({P}_t\) of the teacher model and the original noisy label \(\textbf{Y}\) , the error map that can represent the error in the original label is obtained through CL. The following steps are required to obtain the error map \(Map_{err}\) corresponding to image \(\textbf{X}\) .
First, assume that in mask \(\textbf{Y}\) , the pixel x is labeled \(y = u\) . In the prediction result \({P}_t\) of the teacher model, the probability \(\hat{p}_u(x)\) denotes the self-confidence that pixel x belongs to class u . However, if the probability of pixel x being classified into class v is higher than the threshold \(t_v\) in \(P_t\) , it suggests that the pixel is more likely to belong to the class v rather than u . This latent ground truth is represented as \(y^*\) . The calculation of threshold for class v , i.e. \(t_v\) , is shown in the Eq. 2 :
where \(\textbf{X}_{y=v}\) represents the pixels classified as v in the image \(\textbf{X}\) .
According to y and \(y^*\) , the counting matrix \(\textbf{C}\) of confidence is obtained as shown in Eq. 3 :
where \(\mathcal {C}_m \in \{1, 2,..., m\}\) is a collection of label classes, and m is the number of classes.
Secondly, the matrix \(\textbf{C}\) is calibrated to obtain \(\textbf{C}'\) , as shown in the Eq. 4 :
Then, we obtain the joint distribution matrix \(\hat{\textbf{Q}}\) that estimates y and \(y^*\) , as shown in Eq. 5 :
According to the obtained joint distribution matrix \(\hat{\textbf{Q}}\) and the strategy provided in [ 54 ], we employ the “Prune-by-class" approach to identify noisy labels in \(\textbf{Y}\) . Finally, we get the estimate error map \(Map_{err}\) corresponding to the image \(\textbf{X}\) . In this error map, “1" means that the corresponding pixel is incorrectly labeled, and “0" is the opposite.
(3) Adaptive label correction
Based on the estimated error map \(Map_{err}\) obtained through CL, we employ a computationally simple strategy known as hard refinement to process the mask \(\textbf{Y}\) that contains noisy labels. The mask refined by the hard label is represented as \(\textbf{Y}'\) , which is defined as the Eq. 6 :
Due to the nature of deep learning models, it is easier to learn clean labels during the early stages of training. Therefore, we adaptively correct the noisy labels during training process. The refined label \(\textbf{Y}'\) not only serves as a guide for training the student model, but also acts as a reference for correcting the original label \(\textbf{Y}\) under certain conditions. It is necessary to first evaluate whether to start the correction during training. According to [ 13 ], the \(IoU_{train}\) of the outputs \(P_t\) and the original labels \(\textbf{Y}\) increases rapidly during early stage of training. Therefore, the least square method is employed to fit the \(IoU_{train}\) change during training, as shown in the Eq. 7 :
where t represents the iterations of training. \(0< a \le 1\) , \(b \ge 0\) and \(c \ge 0\) are the parameters that need to be fitted. Thus, the \(IoU_{train}\) change curve is obtained. To determine the appropriate time to initiate the correction of the noisy label, we can utilize the derivative of the function f ( t ), denoted as \(f'(t)\) . When \(\frac{\mid f^{\prime }(1)-f^{\prime }(t)\mid }{\mid f^{\prime }(1)\mid }>g\) is satisfied, we begin to correct \(\textbf{Y}\) according to \(\textbf{Y}'\) , where g takes 0.8.
It is essential to note that when dealing with multiple classes in the label, the time to correct each class is calculated individually. The following is the pseudo-code for the aforementioned correction procedure, as presented in Algorithm 1 .
Teacher-guided noisy label correction with early-learning
Student model. The the input image is denoted by X , and its features are extracted via the Encoder. This feature is directly fed into the Decoder to yield one prediction. Additionally, the feature enters the Decoder through the feature perturbation to obtain another prediction. FP in the figure represents feature perturbation
For ACDC dataset, the distribution of the IoU between the ground truth and the noisy label for different noisy ratios
We focus more on the robustness of the student model as the teacher model only participates in the correction process of the noise labels. To enhance the model’s learning ability, a method of expanding the perturbation space using auxiliary feature perturbation flow is proposed [ 19 ]. This method does not require additional effort to explore the optimal combinations of perturbations and hyperparameters, and has emerged as a prominent method within the field of semi-supervised learning, offering notable advantages over other methods. Inspired by this, we propose a student model with feature perturbation, as shown in Fig. 2 .
Specifically, the original student model consists of one feedforward stream that yields \(P_s\) . Now, we introduce an additional feedforward stream incorporating a feature perturbation between the Encoder e and the Decoder d in the student model. Through this feedforward stream, we obtain \(P_s^{fp}\) which is defined as follows:
where t refers to the intermediate features extracted by the Encoder from the image \(\textbf{X}\) . F is the feature perturbation. According to the reference [ 19 ], feature perturbation can be effectively achieved by utilizing a simple channel dropout. That means we introduce feature perturbations in the student model without incurring excessive computational overhead. Incorporating the consistency loss between \(P_s\) and \(P_s^{fp}\) to enhance the learning capacity of the student model.
The loss function for the method is comprised of three distinct components, denoted as \(\mathcal {L}_c\) , \(\mathcal {L}_1\) and \(\mathcal {L}_2\) , respectively. \(\mathcal {L}_c\) denotes the consistency loss:
where N denotes the number of samples. And \(P_s\) and \(P_s^{fp}\) is two prediction of student model.
\(\mathcal {L}_1\) represents the loss calculation between the output \(P_s\) of the student model and the denoised label as shown in Fig. 1 . And \(\mathcal {L}_2\) represents the loss calculation between \(P_s\) and the label \(\textbf{Y}\) or the corrected label \(\dot{\textbf{Y}}\) . \(\mathcal {L}_1\) and \(\mathcal {L}_2\) consist of cross entropy (CE) loss, dice loss, focal loss and boundary loss. The definition is as follows:
where \(\lambda \) is set to 0.5. The items in Eq. 11 represent the loss between the predicted results of the student model and the labels in the current dataset. And the items in Eq. 12 represent the loss between the predicted results and the denoised labels. According to Eqs. 10 , 11 and 12 , the total loss \(\mathcal {L}\) is expressed as
where \(\mu \) is also set to 0.5. \(\gamma \) is set to a weight that gradually increases with the training period. Specifically, \(\gamma \) uses a Gaussian ramp-up curve [ 55 ]:
where \(\varphi _{max}\) is typically set to 0.8, and \(t_{max}\) represents the maximum number of training iterations.
To validate the efficacy of our proposed method, we executed experiments on two distinct medical datasets. To assess the robustness of our method against datasets with imprecise annotations, we introduced noise to the labels within both datasets.
Automated Cardiac Diagnosis Challenge (ACDC) dataset. ACDC leverages dynamic magnetic resonance imaging (MRI) to delineate the anatomical structures of the heart, focusing on the left ventricle (LV), right ventricle (RV), and myocardium (Myo). This dataset encompasses a total of 150 patient cases. Following [ 12 ], we apply the same data processing to the dataset. Specifically, patches with dimensions of 128 \(\times \) 128 pixels are extracted, centering on the heart region. Within this dataset, a subset of 1312 patches is allocated for training, while 380 patches are dedicated to the validation set.
Noisy label generation of ACDC. Since there are three classes in the ACDC dataset, the incidence of labeling inaccuracies is substantially higher. Such discrepancies not only manifest as annotation errors contrasting the targeted segmentation zones against the background but also amongst different target classes themselves. To simulate noisy labels in the segmentation task, we randomly employ a series of image manipulation operations: resizing, deformation, rotation, translation, and morphological dilation of the mask. In the training set of ACDC, we randomly select the samples with proportion r as the samples with noisy labels. In Fig. 3 , we show the distribution of the intersection-over-union (IoU) ratio between the ground truth and the noisy label for different noisy ratios. Figure 4 shows the examples of the original images, the original labels and the noisy labels.
ACDC dataset with noisy labels. a and d are medical images. b and e are the ground truth (clean labels). c and f are the noisy labels. In masks, LV is labeled yellow, RV is labeled green, and Myo is labeled red
For LA dataset, the distribution of the IoU between the ground truth and the noisy label for different noisy ratios
LA dataset with noisy labels. a and d are medical images. b and e are the ground truth (clean labels). c and f are the noisy labels. The parts labeled in red are the segmented organs
3D Left atrium (LA) dataset. The LA dataset contains 100 subjects, which are composed of 3D gadolinium enhanced magnetic resonance images and their corresponding precise labels. The spatial resolution of the 3D image is 0.625 \(\times \) 0.625 \(\times \) 0.625 mm \(^{3}\) . Consistent with [ 56 ], we use 80 samples as the training set and 20 samples as the test set.
Noisy label generation of LA. The LA dataset contains labels for only one object class in addition to the background. To mimic realistic labeling inaccuracies within this dataset, we emulated the methodology delineated in [ 15 ], whereby the binary masks undergo stochastic morphological transformations such as erosion and dilation within a pixel range spectrum of 3 to 15, thusly engendering synthetic label noise. Since the LA dataset contains few objects that need to be segmented, we deliberately increase the ratio of samples with noisy labels. Consequently, the distribution of IoU between the noisy labels and clean labels is shown in Fig. 5 . The examples of original images, original labels and noisy labels in LA dataset is shown in Fig. 6 .
Implementation and evaluation metrics. Our experimental framework was executed using a single Nvidia 3090 GPU operating under Ubuntu 22.04 LTS. The development environment was structured utilizing Python 3.6, complemented by PyTorch 1.11.0 for constructing our deep learning models. For ACDC and LA datasets, we employed 2D Unet and 3D Unet architectures as our backbone networks respectively. For the ACDC dataset, we conducted training for 100 epochs, while for the LA dataset, the number of training epochs was set as 400, and we set batch size as 4. We initialized the learning rate at 0.01 and engaged the Stochastic Gradient Descent (SGD) optimizer for parameter updates.
In our study, we have employed standard evaluation metrics that are widely recognized in the domain of medical image segmentation. These include the Dice similarity coefficient, the Jaccard (JAC), and the 95% Hausdorff distance (95HD), providing a comprehensive assessment of segmentation accuracy and similarity to ground truth.
Compared method. To validate the robustness and effectiveness of our proposed method of noisy label learning, we benchmarked its performance against state-of-the-art methods:
GCE [ 36 ], a generalized cross entropy loss function is proposed for learning noisy labels.
Co-teaching [ 10 ], the two networks undergo simultaneous training, operating under a cooperative learning paradigm within mini-batch processing. Specifically, each network is responsible for identifying and selecting clean samples. These clean samples are then presented to the other network for learning.
TriNet [ 11 ], to select clean samples and learn more accurately, an additional network is added to the architecture of Co-teaching in this method. This configuration allows the networks to alternate roles, where each takes turns functioning as a ‘teacher’ to guide the learning process of the other networks.
2SRnT [ 51 ], There are also two models in this method. The teacher model is trained on the original dataset with noisy labels. The trained teacher model corrects the noise in the original dataset. The student model is trained on the corrected labels for the next step.
ADELE [ 13 ], according to the characteristics of network learning, this method makes the network correct the noise labels while being trained.
RMD [ 12 ], an algorithm in which two models perform complementary Co-training. A distillation module is designed to refine high-quality knowledge and enhance the use of reliable knowledge in this method.
Comparison of segmentation results of seven model on the same image. All seven models were trained on the dataset contained noisy labels with a noisy ratio of 0.8
To substantiate the efficacy of our proposed meth, we have contrasted the performance of our model with the aforementioned model by training them on three ACDC datasets with different noisy ratios. Figure 7 illustrates the results inferred from models trained on the dataset with a noisy ratio of 0.8. The arrangement of Fig. 7 is as follows. The first column presents the images from the test dataset; the second column exhibits the true labels. Subsequently, columns 3 to 7 confer the prediction results from seven different models. Similarly to Fig. 4 , within the mask, the background is denoted by gray, and the LV, RV, and Myo components are represented by yellow, green, and red, respectively. Observationally, our proposed method’s prediction mask demonstrates a higher degree of resemblance to the true label. Evidently, it exhibits an enhanced capability to more accurately segment each section of the heart depicted in the figure.
A comprehensive comparison of evaluation metrics across our proposed approach and other methods are shown in Table 1 . We present results of four different noisy ratios (r = 0.2, 0.4, 0.6, 0.8). To enhance the clarity of the results, we have highlighted the most superior values in each set in bold and underlined those that are second best. It is discernible that the performance distinction among the various methods is relatively minute when applied to the group with a lower noisy ratio of 0.4. However, the disparity in effectiveness becomes more evident in the group with a significantly higher noisy ratio of 0.8. In this case, our proposed method demonstrates a sensational improvement, enhancing both the Dice and JAC metrics by an average of approximately 2.5%. Consequently, this implies that our methodology exhibits exceptional performance, even the training labels with high noisy ratio.
Comparison of segmentation results of seven model on the same image. All seven models were trained on the LA dataset contained noisy labels with a noisy ratio of 0.9
Analogously, we train our model along with other existing models on three distinct LA datasets infused with different noisy ratios. The results yielded from the training on the dataset characterized by a considerably high noisy ratio of 0.9 are shown in Fig. 8 . To improve the comprehensibility of the results, we overlay the masks with the images.
Echoing the arrangement in Fig. 7 , Fig. 8 presents, from left to right, the images, the ground truth, and the inference results corresponding to seven models. We can see that when the segmentation of the target region is small, some methods have poor performance. Indeed, in severe instances, some methods fail to recognize the target region at all. However, our proposed method continues to yield commendable results, demonstrating a formidable ability to accurately segment even in circumstances where the target area is small or the boundary is intricate.
Furthermore, when our proposed method compares with other methods on the LA dataset, the evaluation metrics, as shown in Table 2 . The proposed method notably outperforms the others, especially in cases of high noisy ratios. For the performance indicator 95HD, with a noisy ratio as high as 0.9, our method successfully reduces the metric to 13.15 mm.
In affirmation of the efficacy of our proposed method’s components, we have performed a series of ablation experiments. These trials were conducted on two distinct datasets—the ACDC dataset with a noisy ratio of 0.6 and the LA dataset with a noisy ratio of 0.9. The results of these experiments are consolidated in Table 3 .
In the first row, we observe performance of the baseline model. Following which, the second row of results elucidates the effects of incorporating Label Correction (LC) during the training phase for one network. LC is implemented by a single network, corresponding to the third column, here the label correction is guided by the student model (SG). In the third line, the model also uses Teacher model Guided (TG) in addition to label correction during training. Finally, the fourth row represents the introduction of Feature Perturbation (FP) to the student model. This set of experiments serves as a robust check to validate the effectiveness of individual components contributing to our proposed method.
It is worth noting that the second and third rows in Table 3 represent the results of experiments using a single model and a teacher model to guide the correction of noisy labels, respectively. In our proposed method, the teacher model solely focuses on correcting noisy labels and does not partake in model updates. Therefore, noisy label correction guided by a single model is equivalent to correction guided by the student model.
The results of using student model and teacher model for label correction of ACDC dataset with a noisy ratio of 0.6. The “SG” represents the student model guided, and “TG” represents the teacher model guided
The results of using student model and teacher model for label correction of LA dataset with a noisy ratio of 0.9. The “SG” represents the student model guided, and “TG” represents the teacher model guided
Since the ACDC dataset contains four target organs requiring segmentation, mere label correction proves challenging with low accuracy, yielding minimal enhancement in model performance. With the integration of the TG component, substantial improvements are observed across all metrics, validating the efficacy of TG. The Fig. 9 shows that, when the image contours are clear, the results are close to the ground truth regardless of whether the noisy label correction is guided by the student model or the teacher model. However, when the image is not clear or interference occurs around the target (the second and third rows), the results obtained using the student model for label correction have larger errors. And the LA dataset comprises one class, thus, correcting noisy labels during training markedly enhances model performance. It can be seen in Fig. 10 that the results of SG and TG are similar. Our analysis reveals a significant hike in the model’s performance on datasets containing multiple class labels, when TG is introduced.
Experiments were carried out to determine the influence of the hyperparameter g , which governs the commencement timing of correcting the noisy labels. The outcomes of these tests are presented in Table 4 . A discernible degradation in performance was observed when g was set at 0.6 and 0.7. In contrast, the performance variation was relatively insignificant when g was adjusted to 0.8, 0.9, and 0.99.
Moreover, we conducted experiments on the hyperparameter \(\phi _{max}\) within \(\gamma (t)\) , which controls the proportion of the loss function. The corresponding results are shown in Table 5 . According to the tabulated data, the model exhibits optimal performance when \(\phi _{max}\) is set to 0.8.
In this paper, we propose a novel method aimed at addressing the issue of noisy label learning within the field of medical image segmentation. Our method incorporates both a student model and a teacher model, enabling learning from high-quality and low-quality datasets and yielding exceptional outcomes. The teacher model adaptively rectifies the noisy labels within the dataset during the training phase. Meanwhile, the student model learns from the labels of the current dataset and the denoised labels. Alongside, feature perturbations are incorporated into the student model to bolster robustness. The results from our experiments indicate that our method is highly effective. When applied to the ACDC dataset with a noisy ratio of 0.8, the mean Dice is enhanced by 2.58%.
The data used to support the findings of this study are included within the paper.
Dou H, Liu Y, Chen S, Zhao H, Bilal H (2023) A hybrid ceemd-gmm scheme for enhancing the detection of traffic flow on highways. Soft Comput 27(21):16373–16388
Article Google Scholar
Wu Q, Li X, Wang K, Bilal H (2023) Regional feature fusion for on-road detection of objects using camera and 3d-lidar in high-speed autonomous vehicles. Soft Comput 27(23):18195–18213
Kora P, Ooi CP, Faust O, Raghavendra U, Gudigar A, Chan WY, Meenakshi K, Swaraja K, Plawiak P, Acharya UR (2022) Transfer learning techniques for medical image analysis: a review. Biocybernet Biomed Eng 42(1):79–107
Qureshi I, Yan J, Abbas Q, Shaheed K, Riaz AB, Wahid A, Khan MWJ, Szczuko P (2023) Medical image segmentation using deep semantic-based methods: a review of techniques, applications and emerging trends. Inform Fus 90:316–352
Vec T, Rupnik Vec T, Žorga S (2014) Understanding how supervision works and what it can achieve. The Wiley international handbook of clinical supervision, 103–127
Song H, Kim M, Park D, Shin Y, Lee J-G (2022) Learning from noisy labels with deep neural networks: a survey. IEEE Transactions on Neural Networks and Learning Systems
Chen Y, Liu F, Wang H, Wang C, Liu Y, Tian Y, Carneiro G (2023) Bomd: bag of multi-label descriptors for noisy chest x-ray classification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 21284–21295
Oh Y, Kim B, Ham B (2021) Background-aware pooling and noise-aware loss for weakly-supervised semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6913–6922
Wang J, Zhou S, Fang C, Wang L, Wang J (2020) Meta corrupted pixels mining for medical image segmentation. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part I 23, pp. 335–345. Springer
Han B, Yao Q, Yu X, Niu G, Xu M, Hu W, Tsang I, Sugiyama M (2018) Co-teaching: Robust training of deep neural networks with extremely noisy labels. Advances in neural information processing systems 31
Zhang T, Yu L, Hu N, Lv S, Gu S (2020) Robust medical image segmentation from non-expert annotations with tri-network. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part IV 23, pp. 249–258. Springer
Fang C, Wang Q, Cheng L, Gao Z, Pan C, Cao Z, Zheng Z, Zhang D (2023) Reliable mutual distillation for medical image segmentation under imperfect annotations. IEEE Transactions on Medical Imaging
Liu S, Liu K, Zhu W, Shen Y, Fernandez-Granda C (2022) Adaptive early-learning correction for segmentation from noisy annotations. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 2606–2616
Luo W, Yang M (2020) Semi-supervised semantic segmentation via strong-weak dual-branch network. In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16, pp. 784–800. Springer
Xu Z, Lu D, Luo J, Wang Y, Yan J, Ma K, Zheng Y, Tong RK-Y (2022) Anti-interference from noisy labels: Mean-teacher-assisted confident learning for medical image segmentation. IEEE Trans Med Imag 41(11):3062–3073
Tarvainen A, Valpola H (2017) Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems 30
Tarvainen A, Valpola H (2017) Mean teachers are better role models: weight-averaged consistency targets improve semi-supervised deep learning results. International Conference on Learning Representations,International Conference on Learning Representations
Liu Y, Tian Y, Chen Y, Liu F, Belagiannis V, Carneiro G (2022) Perturbed and strict mean teachers for semi-supervised semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4258–4267
Yang L, Qi L, Feng L, Zhang W, Shi Y (2023) Revisiting weak-to-strong consistency in semi-supervised semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7236–7246
Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18, pp. 234–241. Springer
Yin X-X, Sun L, Fu Y, Lu R, Zhang Y, et al (2022) U-net-based medical image segmentation. J Healthcare Eng 2022
Cao H, Wang Y, Chen J, Jiang D, Zhang X, Tian Q, Wang M (2022) Swin-unet: Unet-like pure transformer for medical image segmentation. In: European Conference on Computer Vision, pp. 205–218. Springer
Xu Q, Ma Z, Na H, Duan W (2023) Dcsau-net: A deeper and more compact split-attention u-net for medical image segmentation. Comput Biol Med 154:106626
Yan X, Tang H, Sun S, Ma H, Kong D, Xie X (2022) After-unet: Axial fusion transformer unet for medical image segmentation. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 3971–3981
Wang H, Xie S, Lin L, Iwamoto Y, Han X-H, Chen Y-W, Tong R (2022) Mixed transformer u-net for medical image segmentation. In: ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2390–2394. IEEE
Chen B, Liu Y, Zhang Z, Lu G, Kong AWK (2023) Transattunet: Multi-level attention-guided u-net with transformer for medical image segmentation. IEEE Transactions on Emerging Topics in Computational Intelligence
Yuan F, Zhang Z, Fang Z (2023) An effective cnn and transformer complementary network for medical image segmentation. Pattern Recognit 136:109228
Jiao R, Zhang Y, Ding L, Xue B, Zhang J, Cai R, Jin C (2023) Learning with limited annotations: a survey on deep semi-supervised learning for medical image segmentation. Comput Biol Med 107840
Luo X, Hu M, Song T, Wang G, Zhang S (2022) Semi-supervised medical image segmentation via cross teaching between cnn and transformer. In: International Conference on Medical Imaging with Deep Learning, pp. 820–833. PMLR
Wang S, Li C, Wang R, Liu Z, Wang M, Tan H, Wu Y, Liu X, Sun H, Yang R (2021) Annotation-efficient deep learning for automatic medical image segmentation. Nat Commun 12(1):5915
Zhang L, Tanno R, Xu M-C, Jin C, Jacob J, Cicarrelli O, Barkhof F, Alexander D (2020) Disentangling human error from ground truth in segmentation of medical images. Adv Neural Inform Process Syst 33:15750–15762
Google Scholar
Li M, Soltanolkotabi M, Oymak S (2020) Gradient descent with early stopping is provably robust to label noise for overparameterized neural networks. In: International Conference on Artificial Intelligence and Statistics, pp. 4313–4324. PMLR
Cheng J, Liu T, Ramamohanarao K, Tao D (2020) Learning with bounded instance and label-dependent label noise. In: International conference on machine learning, pp. 1789–1799. PMLR
Goldberger J, Ben-Reuven E (2016) Training deep neural-networks using a noise adaptation layer. In: International conference on learning representations
Bekker AJ, Goldberger J (2016) Training deep neural-networks based on unreliable labels. In: 2016 IEEE International conference on acoustics, speech and signal processing (ICASSP), pp. 2682–2686. IEEE
Zhang Z, Sabuncu M (2018) Generalized cross entropy loss for training deep neural networks with noisy labels. Adv Neural Inform Process Syst 31
Ma X, Huang H, Wang Y, Romano S, Erfani S, Bailey J (2020) Normalized loss functions for deep learning with noisy labels. In: International Conference on Machine Learning, pp. 6543–6553. PMLR
Song H, Kim M, Park D, Shin Y, Lee J-G (2021) Robust learning by self-transition for handling noisy labels. In: Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, pp. 1490–1500
Cordeiro FR, Sachdeva R, Belagiannis V, Reid I, Carneiro G (2023) Longremix: Robust learning with high confidence samples in a noisy label environment. Pattern Recognit 133:109013
Zheng G, Awadallah AH, Dumais S (2021) Meta label correction for noisy label learning. In: Proceedings of the AAAI conference on artificial intelligence, vol. 35, pp. 11053–11061
Jiang H, Gao M, Hu Y, Ren Q, Xie Z, Liu J (2023) Label-noise-tolerant medical image classification via self-attention and self-supervised learning. arXiv preprint arXiv:2306.09718
Tan C, Xia J, Wu L, Li SZ (2021) Co-learning: Learning from noisy labels with self-supervision. In: Proceedings of the 29th ACM international conference on multimedia, pp. 1405–1413
Liu C, Albrecht C, Wang Y, Zhu XX (2024) Task specific pretraining with noisy labels for remote sensing image segmentation. arXiv preprint arXiv:2402.16164
Guo X, Yuan Y (2022) Joint class-affinity loss correction for robust medical image segmentation with noisy labels. In: International conference on medical image computing and computer-assisted Intervention, pp. 588–598. Springer
Li S, Gao Z, He X (2021) Superpixel-guided iterative learning from noisy labels for medical image segmentation. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part I 24, pp. 525–535. Springer
Guo R, Xie K, Pagnucco M, Song Y (2023) Sac-net: Learning with weak and noisy labels in histopathology image segmentation. Med Image Anal 86:102790
Karimi D, Rollins CK, Velasco-Annis C, Ouaalam A, Gholipour A (2023) Learning to segment fetal brain tissue from noisy annotations. Med Image Anal 85:102731
Zhou Y, Yu H, Shi H (2021) Study group learning: Improving retinal vessel segmentation trained with noisy labels. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part I 24, pp. 57–67. Springer
Liao Z, Hu S, Xie Y, Xia Y (2024) Modeling annotator preference and stochastic annotation error for medical image segmentation. Med Image Anal 92:103028
Liu S, Li Y, Chai Q-W, Zheng W (2024) Region-scalable fitting-assisted medical image segmentation with noisy labels. Expert Syst Appl 238:121926
Zhang M, Gao J, Lyu Z, Zhao W, Wang Q, Ding W, Wang S, Li Z, Cui S (2020) Characterizing label errors: Confident learning for noisy-labeled image segmentation. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part I 23, pp. 721–730. Springer
Dolz J, Desrosiers C, Ayed IB (2021) Teach me to segment with mixed supervision: Confident students become masters. In: Information processing in medical imaging: 27th international conference, IPMI 2021, virtual event, June 28–June 30, 2021, Proceedings 27, pp. 517–529. Springer
Liu S, Niles-Weed J, Razavian N, Fernandez-Granda C (2020) Early-learning regularization prevents memorization of noisy labels. Adv Neural Inform Process Syst 33:20331–20342
Northcutt C, Jiang L, Chuang I (2021) Confident learning: estimating uncertainty in dataset labels. J Artificial Intell Res 70:1373–1411
Article MathSciNet Google Scholar
Cui W, Liu Y, Li Y, Guo M, Li Y, Li X, Wang T, Zeng X, Ye C (2019) Semi-supervised brain lesion segmentation with an adapted mean teacher model. In: Information Processing in Medical Imaging: 26th International Conference, IPMI 2019, Hong Kong, China, June 2–7, 2019, Proceedings 26, pp. 554–565. Springer
Yu L, Wang S, Li X, Fu C-W, Heng P-A (2019) Uncertainty-aware self-ensembling model for semi-supervised 3d left atrium segmentation. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part II 22, pp. 605–613. Springer
Download references
The authors acknowledge the funding of National Natural Science Foundation of China, Grant No. 11974373.
Authors and affiliations.
Shandong University of Science and Technology, Qingdao, 266590, Shandong, China
Shangkun Liu, Minghao Zou, Ning Liu, Yanxin Li & Weimin Zheng
Nanjing University of Information Science and Technology, Nanjing, 210044, JiangSu, China
Weimin Zheng
You can also search for this author in PubMed Google Scholar
Correspondence to Weimin Zheng .
Conflict of interest.
The authors declare there are no Conflict of interest regarding the publication of this paper.
Publisher's note.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .
Reprints and permissions
Liu, S., Zou, M., Liu, N. et al. A teacher-guided early-learning method for medical image segmentation from noisy labels. Complex Intell. Syst. (2024). https://doi.org/10.1007/s40747-024-01574-1
Download citation
Received : 08 February 2024
Accepted : 21 July 2024
Published : 13 August 2024
DOI : https://doi.org/10.1007/s40747-024-01574-1
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Published on 13.8.2024 in Vol 26 (2024)
Authors of this article:
1 Department of Nursing, the First Affiliated Hospital of Anhui Medical University, Hefei, China
2 School of Nursing, Anhui Medical University, Hefei, China
3 College of Traditional Chinese Medicine, Bozhou University, Bozhou, China
4 Department of Gastrointestinal Surgery, the First Affiliated Hospital of Anhui Medical University, Hefei, China
*these authors contributed equally
Shaohua Hu, PhD
Department of Nursing
the First Affiliated Hospital of Anhui Medical University
218 Jixi Road
Hefei, 230009
Phone: 86 62922005
Email: [email protected]
Background: People who undergo sphincter-preserving surgery have high rates of anorectal functional disturbances, known as low anterior resection syndrome (LARS). LARS negatively affects patients’ quality of life (QoL) and increases their need for self-management behaviors. Therefore, approaches to enhance self-management behavior and QoL are vital.
Objective: This study aims to assess the effectiveness of a remote digital management intervention designed to enhance the QoL and self-management behavior of patients with LARS.
Methods: From July 2022 to May 2023, we conducted a single-blinded randomized controlled trial and recruited 120 patients with LARS in a tertiary hospital in Hefei, China. All patients were randomly assigned to the intervention group (using the “e-bowel safety” applet and monthly motivational interviewing) or the control group (usual care and an information booklet). Our team provided a 3-month intervention and followed up with all patients for an additional 3 months. The primary outcome was patient QoL measured using the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30. The secondary outcomes were evaluated using the Bowel Symptoms Self-Management Behaviors Questionnaire, LARS score, and Perceived Social Support Scale. Data collection occurred at study enrollment, the end of the 3-month intervention, and the 3-month follow-up. Generalized estimating equations were used to analyze changes in all outcome variables.
Results: In the end, 111 patients completed the study. In the intervention group, 5 patients withdrew; 4 patients withdrew in the control group. Patients in the intervention group had significantly larger improvements in the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 total score (mean difference 11.51; 95% CI 10.68-12.35; Cohen d =1.73) and Bowel Symptoms Self-Management Behaviors Questionnaire total score (mean difference 8.80; 95% CI 8.28-9.32; Cohen d =1.94) than those in the control group. This improvement effect remained stable at 3-month follow-up (mean difference 14.47; 95% CI 13.65-15.30; Cohen d =1.58 and mean difference 8.85; 95% CI 8.25-9.42; Cohen d =2.23). The LARS score total score had significantly larger decreases after intervention (mean difference –3.28; 95% CI –4.03 to –2.54; Cohen d =–0.39) and at 3-month follow-up (mean difference –6.69; 95% CI –7.45 to –5.93; Cohen d =–0.69). The Perceived Social Support Scale total score had significantly larger improvements after intervention (mean difference 0.47; 95% CI 0.22-0.71; Cohen d =1.81).
Conclusions: Our preliminary findings suggest that the mobile health–based remote interaction management intervention significantly enhanced the self-management behaviors and QoL of patients with LARS, and the effect was sustained. Mobile health–based remote interventions become an effective method to improve health outcomes for many patients with LARS.
Trial Registration: Chinese Clinical Trial Registry ChiCTR2200061317; https://tinyurl.com/tmmvpq3
The Global Cancer Statistics 2020 showed that colorectal cancer ranks third in incidence of malignant tumors and second in cause of death worldwide [ 1 ]. Colorectal cancer incidence is also on the rise in China, with rectal cancer accounting for 60% of cases and middle and lower rectal cancers being the most common [ 2 ]. With the advancement of medical technology, optimal management of middle and lower rectal cancers increasingly favors sphincter-preserving surgery (SPS) [ 3 ]. This operation preserves anal function and avoids the inconvenience and pressure caused by permanent colostomy [ 4 ]. However, 70%-90% of patients after SPS struggle with long-term anorectal functional disturbances called low anterior resection syndrome (LARS) [ 5 , 6 ].
The presence of LARS has a severe adverse effect on the quality of life (QoL) of patients [ 7 ]. Postoperative LARS induces a spectrum of adverse physical and psychological effects in patients; for example, up to 50% of patients with LARS report toilet dependence during rehabilitation [ 8 , 9 ], 36% of patients experience pain, and approximately 13% of patients report high psychological distress [ 10 , 11 ]. Furthermore, LARS can restrict a patient’s social life, leading to further impact on their QoL [ 12 ]. Recently, longitudinal studies have found that patients’ QoL is still affected by LARS even 15 years after surgery [ 13 ]. Research has shown that patients can improve their QoL through methods, such as pelvic floor muscle exercises and dietary adjustments during home care; however, the effectiveness of these methods is limited by patients’ lack of knowledge of LARS and rehabilitation guidance [ 14 , 15 ].
Owing to the frequent occurrence of LARS in patients post discharge, patients must have a high level of self-management behavior [ 16 ]. However, in China, the majority of patients have a passive response to LARS, and their self-management behavior is at a low level [ 17 ]. Enhancing self-management awareness and providing information on supportive care can improve the self-management behavior of patients with LARS [ 18 ]. Research has demonstrated that motivational interviewing (MI) enhances self-management awareness and supports behavioral change [ 19 ].
Therefore, to improve patients’ QoL and self-management behaviors, providing supportive care information to patients is crucial. A qualitative exploration of patients with LARS’s perspectives on information needs revealed that timely symptom management measures are critical during home-based rehabilitation [ 20 ]. However, it is difficult to maintain continuity and instantaneity with existing management measures [ 21 , 22 ]. Owing to current advances in mobile technology, mobile health (mHealth) has been widely considered a means of patient health management, which can improve the effects of symptoms and assist patients in timely access to the required information [ 23 , 24 ].
To date, remote follow-up tools for patients with LARS have yielded promising results [ 25 ]. For patients with LARS, mHealth-based remote interventions may become an effective method to assist them in improving symptoms. However, mHealth intervention measures constructed for patients with LARS are rare. Most studies have only completed the development and pilot research of remote intervention programs, leading to insufficient data on the effectiveness of remote interventions in improving patient health outcomes [ 26 , 27 ]. WeChat (Tencent Corp) is China’s most frequently used instant messaging and social media application [ 28 ]. Evidence suggests that WeChat-based mHealth interventions effectively improve health outcomes in various health conditions [ 29 , 30 ].
This study aimed to assess the effectiveness of a remote digital management intervention designed for patients with LARS. The effectiveness of the intervention measure is determined by improvement in QoL, self-management behaviors, gastrointestinal symptoms, and social support. We hypothesized that the remote digital management intervention can effectively improve the health outcomes of patients with LARS.
This study was conducted from July 15, 2022, to March 15, 2023, in Hefei, China. Our team provided a 3-month intervention and followed up with all patients for an additional 3 months. The intervention group used the “e-bowel safety” applet and received monthly MI. The control group received the usual care and was provided with a handbook containing information related to LARS. The CONSORT (Consolidated Standards of Reporting Trials) checklist is in Multimedia Appendix 1 .
This randomized controlled trial (RCT) was approved by the ethics committee of the First Affiliated Hospital of Anhui Medical University (PJ2022-07-53) and registered on the Chinese Clinical Trial Registry (ChiCTR2200061317). All data were identified with a code number to ensure the confidentiality of the subjects’ data. No compensation was provided to participants.
The patients were recruited from a tertiary hospital in Hefei, Anhui Province, China. Patients were eligible to participate in our study if they met the following criteria: age older than 18 years, a diagnosis of rectal cancer, underwent SPS, LARS scores ≥21, ostomy closure surgery performed at least 3 months prior, the ability to read and write text, and proficiency in using WeChat. Patients with chronic gastrointestinal conditions, prior or current mental health disorders, cognitive impairments, communication disorders, or those who have participated in other clinical studies are ineligible for participation in this research. When patients meeting the recruitment criteria appeared in the hospital database, the system sent recruitment information to these patients with the approval of doctors not directly involved in the research design.
In this study, the sample size was determined based on the QoL. Previous research has shown that the QoL for patients with rectal cancer is 77±19 [ 31 ]. In an RCT using the EORTC QLQ-C30, a difference of 10 points is considered clinically significant [ 32 ]. With a two-sided test level of 0.05 and 80% test efficacy, each group requires a sample size of 45. Accounting for a 20% dropout rate, 112 patients are needed.
Our previous study provided a comprehensive description of the intervention protocol [ 33 ]. The patients in the intervention group used the “e-bowel safety” applet for 3 months. They were required to check in on the applet daily and record their daily gastrointestinal symptoms. Our “e-bowel safety” applet comprises 4 main sections: a rehabilitation plan, LARS knowledge, web-based consultation, and patient stories. The rehabilitation plan module involves the collaborative development of home dietary and exercise plans by patients and researchers. The applet features intelligent reminders to monitor daily plan completion and provide prompts. After completing the rehabilitation plan, patients must fill out a daily health diary, and researchers dynamically adjust the rehabilitation plan based on patients’ feedback and physical condition. The LARS knowledge module offers evidence-based information on LARS and symptom management strategies. The web-based consultation module provides patients with an opportunity to interact with health care professionals, offering personalized guidance and feedback. The patient stories module allows patients to share symptom management experiences or engage with other patients, with all published content subject to researcher approval. Additionally, an incentive system has been designed to encourage participation. For instance, patients earn points by sharing personal stories or comments, which can later be exchanged for rewards after accumulating a certain number of points.
Moreover, our team members conducted monthly MIs with patients. MIs were led by 4 researchers with expertise in health coaching and disease management, including 1 clinical psychologist (Shangxin Zhang) and 3 registered nurses (TW, HH, and Ling Fang). The researchers engage with patients via WeChat for 30-60 minutes per call. The aim of MIs is to assist patients in setting rehabilitation goals, reinforcing self-management awareness, and promoting health behavior changes. The content of MIs is based on the interview guide determined by the research team, which guides the conversation from the initial session to explore the participant’s motivation to identify the facilitating factors and barriers to achieving their health goals. The interview guide is outlined in Multimedia Appendix 2 .
Patients in the control group received the usual care and were provided with a handbook containing information related to LARS. At the same time, our team members followed up with patients, using the same timing and frequency as the MI intervention group.
This study was a single-blind, two-arm RCT. After obtaining consent from eligible patients, assistants who were not involved in the study randomly assigned them to the intervention and control groups at a 1:1 ratio. The randomization process was performed by the assistants and anonymized envelopes were used with block randomization, including block sizes randomly varying between 4 (2:2) and 6 (3:3). The research assistants (Ping Ni and Ai Wang) who collected the data were unaware of the patient assignments throughout the study. Patients used the QR codes provided by the research team to access the “e-bowel safety” applet, effectively reducing contamination between the 2 groups. Patients were blinded to their group assignments throughout the entire research process.
Several strategies were used to ensure quality control and participant retention. Our “e-bowel safety” applet can monitor patients’ plan execution and provide reminders, which ensures the daily plans are followed strictly by patients. Before the formal intervention, we conducted a pilot experiment and gathered participant feedback to enhance our plan. The specific results are included in Multimedia Appendix 3 . Furthermore, patients received consistent guidance from our research assistants (Ping Ni and Ai Wang) when they had questions about the questionnaire content. Before the start of the study, all research assistants must undergo training and assessment on the use of all questionnaires by research team members. Only research assistants who pass the assessment can participate in data collection. Additionally, team members regularly check the progress of research assistants’ work to ensure that they are following the questionnaire collection process, identifying issues promptly, and making corrections.
The patients’ demographic and clinical information were obtained from the hospital database. Data were collected from patients using scales for their QoL, social support, self-management behaviors, and LARS scores at different time periods (0, 3, and 6 months). The research assistants (Ping Ni and Ai Wang) who collected the data assisted patients in completing questionnaires over the phone or through direct personal interaction.
The EORTC-QLQ-C30 (European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30) was used to measure QoL. This questionnaire comprises 30 items divided into 15 dimensions, including 1 dimension for QoL, 5 dimensions for functionality, 3 dimensions for symptoms, and 6 dimensions for additional symptoms. All dimension scores were linearly transformed to a scale of 0-100 points. Elevated scores on the 5 functionality dimensions and the QoL dimension were linked to improved functional status, whereas the reverse pattern was observed for the symptom dimensions and additional symptom dimensions. The Cronbach α coefficient ranged from 0.764 to 0.809 [ 34 ].
Self-management.
The self-management behavior of patients was assessed by the Bowel Symptoms Self-Management Behaviors Questionnaire (BSSBQ). This questionnaire comprises 24 items divided into 5 functional scales, with each item scored on a scale of 0 (never) to 7 (always). Higher scores indicate better bowel symptom self-management behavior. The Cronbach α coefficient was 0.81 [ 17 ].
The LARS score consists of 5 items, with a total score ranging from 0 to 42. Patients’ gastrointestinal symptoms are classified into no LARS, minor LARS, and major LARS based on the total score. The LARS score is a validated instrument for assessing bowel symptoms. The Cronbach α coefficient was 0.767 [ 35 ].
The Perceived Social Support Scale (PSSS) consists of 12 items, with each item scored on a scale of 1 (extreme disagreement) to 7 (strong consent). The total scores ranged from 12 to 84. The higher the score, the stronger the perceived social support by the patient. This scale is widely used to assess the level of social support among patients in China. The Cronbach α coefficient of this scale was 0.899 [ 36 ].
The feasibility of intervention was assessed through the completion status of MI sessions and the adherence to health diary entries. The 3-month intervention corresponds to 3 MI sessions and 84 days of health diary entries.
All data were analyzed using SPSS Statistics (version 23.0; IBM Corp). An intention-to-treat analysis was performed in this study. We used the last observed values of the patients to replace missing data. Chi-square analysis was used to analyze the remaining demographic characteristics, and a 2-tailed independent sample t test was used to analyze the age and tumor height. Descriptive data were computed, including means with SD, medians with ranges, and frequencies with proportions where appropriate. The statistical significance was established at P <.05 (2-tailed test). Generalized estimating equations were used to analyze changes in QoL, self-management behaviors, LARS, and social support scores at different time points. The calculation of effect sizes was performed using Cohen d for the mean differences at various time periods.
Initially, 60 patients were recruited in the control and intervention groups. During the study, 9 patients dropped out (dropout rate 7.5%). In the intervention group, 5 patients withdrew from the study, including 2 patients who received a reostomy because of an anastomotic fistula and 3 patients whose condition worsened. In the control group, 4 patients dropped out, including 2 patients whose condition worsened and 2 patients who refused to continue the intervention because of the side effects of chemotherapy. No statistically significant differences were observed between the patients who dropped out and those who completed all evaluations ( P =.17). Figure 1 shows the CONSORT flowchart of this study. Table 1 demonstrates no statistically significant differences in the demographics and clinical information between the control and intervention groups at baseline.
Characteristics | Intervention group (n=60) | Control group (n=60) | test ( ) or chi-square value ( ) | value | ||||||
0.93 (1) | .34 | |||||||||
Male | 42 (70) | 37 (62) | ||||||||
Female | 18 (30) | 23 (38) | ||||||||
Age (years), mean (SD) | 62.72 (7.91) | 61.78 (11.80) | 0.51 (118) | .61 | ||||||
0.07 (2) | .96 | |||||||||
Junior high school or lower | 33 (55) | 32 (53) | ||||||||
High school | 19 (32) | 19 (32) | ||||||||
College or higher | 8 (13) | 9 (15) | ||||||||
0.21 (1) | .65 | |||||||||
Married | 58 (97) | 57 (95) | ||||||||
Single | 2 (3) | 3 (5) | ||||||||
1.42 (3) | .70 | |||||||||
I | 14 (23) | 13 (22) | ||||||||
II | 24 (40) | 30 (50) | ||||||||
III | 20 (33) | 15 (25) | ||||||||
IV | 2 (4) | 2 (3) | ||||||||
Tumor height, mean (SD) | 7.62 (1.708) | 7.80 (1.811) | –0.57 (118) | .57 | ||||||
0.378 (2) | .83 | |||||||||
<6 | 18 (30) | 17 (28) | ||||||||
6-12 | 27 (45) | 25 (42) | ||||||||
>12 | 15 (25) | 18 (30) | ||||||||
0.24 (1) | .62 | |||||||||
Laparoscopy | 51 (85) | 49 (82) | ||||||||
Laparotomy | 9 (15) | 11 (18) | ||||||||
0.34 (1) | .56 | |||||||||
LAR | 58 (9) | 59 (98) | ||||||||
TaTME | 2 (3) | 1 (2) | ||||||||
0.53 (1) | .47 | |||||||||
Yes | 29 (48) | 33 (55) | ||||||||
No | 31 (52) | 27 (45) | ||||||||
0.88 (2) | .65 | |||||||||
Preoperative | 8 (13) | 5 (8) | ||||||||
Postoperative | 49 (82) | 51 (85) | ||||||||
No | 3 (5) | 4 (7) | ||||||||
1.20 (1) | .27 | |||||||||
Countryside | 28 (47) | 34 (57) | ||||||||
City | 32 (53) | 26 (43) | ||||||||
EORTC-QLQ-C30 | 69.67 (4.26) | 69.42 (3.66) | 0.35 (118) | .72 | ||||||
BSSBQ | 30.33 (1.90) | 30.58 (2.01) | –0.70 (118) | .49 | ||||||
LARS score | 31.07 (3.88) | 31.32 (4.73) | –0.32 (118) | .75 | ||||||
PSSS | 34.42 (1.62) | 34.3 (1.48) | 0.29 (118) | .77 |
a LAR: low anterior resection.
b TaTME: transanal total mesorectal excision.
c EORTC-QLQ-C30: European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30.
d BSSBQ: Bowel Symptoms Self-Management Behaviors Questionnaire.
e LARS: Low anterior resection syndrome.
f PSSS: Perceived Social Support Scale.
Table 2 shows that the patients’ QoL improved for both groups. Patients in the intervention group demonstrated greater improvements in the EORTC-QLQ-C30 total score than those in the control group after intervention (mean difference 11.51; 95% CI 10.68-12.35; Cohen d =1.73). Furthermore, this improvement effect remained stable at 3-month follow-up (mean difference 14.47; 95% CI 13.65-15.30; Cohen d =1.58). Table 3 shows that the EORTC-QLQ-C30 total score in both groups exhibited a trend of change over the 6-month period ( P <.001). Differences were observed between the 2 groups and the interaction between group and time. A subgroup analysis was conducted on patients receiving preoperative chemotherapy versus postoperative chemotherapy. Among the 49 patients in the intervention group and 51 in the control group undergoing postoperative chemotherapy, a nominally significant improvement in the change from baseline in the EORTC-QLQ-C30 total score at 3 months was observed compared to the control group (difference of 4.42; P <.001). However, this effect was not seen in patients receiving preoperative chemotherapy. The specific results are included in Multimedia Appendix 4 .
Outcomes | Intervention group, mean (SD) | Control group, mean (SD) | Cohen | GEE statistical tests | ||
Score, (95% CI) | value | |||||
T0 | 69.67 (4.26) | 69.42 (3.66) | N/A | N/A | N/A | |
TI | 83.41 (2.46) | 78.71 (2.72) | 1.73 | 11.51 (10.68 to 12.35) | <.001 | |
T2 | 86.22 (2.49) | 81.82 (2.79) | 1.58 | 14.47 (13.65 to 15.30) | <.001 | |
T0 | 30.33 (1.90) | 30.58 (2.01) | N/A | N/A | N/A | |
TI | 41.23 (2.26) | 37.28 (2.04) | 1.94 | 8.80 (8.28 to 9.32) | <.001 | |
T2 | 42.25 (2.58) | 36.37 (2.63) | 2.23 | 8.85 (8.25 to 9.42) | <.001 | |
score | ||||||
T0 | 31.07 (3.88) | 31.32 (4.73) | N/A | N/A | N/A | |
TI | 26.95 (3.51) | 28.87 (4.83) | –0.39 | –3.28 (–4.03 to 2.54) | <.001 | |
T2 | 22.87 (3.09) | 26.13 (4.67) | –0.69 | –6.69 (–7.45 to 5.93) | <.001 | |
T0 | 34.42 (1.62) | 34.3 (1.48) | N/A | N/A | N/A | |
TI | 36.63 (1.44) | 33.05 (1.98) | 1.81 | 0.47 (0.22 to 0.71) | <.001 | |
T2 | 34.80 (1.19) | 34.40 (1.55) | 0.25 | 0.23 (–0.20 to 0.45) | .07 |
a GEE: Generalized estimating equations.
b Difference in mean change from baseline to endpoint between the groups.
d Baseline.
e N/A: Not applicable.
f After the intervention.
g 3-month follow-up.
h BSSBQ: Bowel Symptoms Self-Management Behavior Questionnaire.
i LARS: Low anterior resection syndrome score.
j PSSS: Perceived Social Support Scale.
Outcomes | Group effect | Time effect | Group×time | |||
test ( ) | value | test ( ) | value | test ( ) | value | |
EORTC-QLQ-C30 | 68.50 (1) | <.001 | 53.81 (2) | <.001 | 27.79 (2) | <.001 |
BSSBQ | 48.15 (1) | <.001 | 74.31 (2) | <.001 | 3.24 (2) | .03 |
LARS Score | 7.78 (1) | .05 | 74.94 (2) | <.001 | 21.34 (2) | <.001 |
PSSS | 29.97 (1) | <.001 | 14.47 (2) | .001 | 71.71 (2) | <.001 |
a EORTC-QLQ-C30: European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30.
b BSSBQ: Bowel Symptoms Self-Management Behaviors Questionnaire.
c LARS: Low anterior resection syndrome.
d PSSS: Perceived Social Support Scale.
Table 2 shows that the patients’ self-management behavior was enhanced for both groups. The BSSBQ total score had significantly larger improvements after intervention (mean difference 8.80; 95% CI 8.28-9.32; Cohen d =1.94) and at 3-month follow-up (mean difference 8.85; 95% CI 8.25-9.42; Cohen d =2.23) between groups. The BSSBQ total score showed statistically significant time effects ( P <.001; Table 3 ).
The LARS score total score had significantly larger decreases after intervention (mean difference –3.28; 95% CI –4.03 to –2.54; Cohen d =–0.39) and at 3-month follow-up (mean difference –6.69; 95% CI –7.45 to –5.93; Cohen d =–0.69). Table 3 shows that the LARS score total score in both groups exhibited a trend of change over the 6-month period. The intergroup effect exhibits homogeneity ( P =.05).
The PSSS total score had significantly larger improvements after intervention (mean difference 0.47; 95% CI 0.22-0.71; Cohen d =1.81); however, the improvement in this effect did not persist at 3-month follow-up (mean difference 0.23; 95% CI –0.20 to 0.45; P =.07; Table 2 ). Table 3 shows that the PSSS total score in both groups exhibited a trend of change over the 6-month period.
Among the 55 patients who completed the intervention, 45 patients completed 3 MI sessions on time, 7 patients postponed 1 MI session because of scheduling conflicts, and 3 patients only completed 2 MI sessions. The mean number of attended MI sessions was 2.95 (SD 0.23). Additionally, 40 patients completed 84 health diary entries, while the remaining 11 patients did not submit completed entries or fulfill the required entries. The mean number of days of health diary entries was 82.87 (SD 3.15). We invited patients from the intervention group to complete a survey to evaluate their perceptions of the intervention's usability. In the end, 49 people completed the survey. The specific results are included in Appendix 5.
To the best of our knowledge, the “e-bowel safety” applet is the first mobile app developed for patients with LARS in China. This study offers a valuable reference point for future initiatives in mHealth interventions for patients with LARS. A mHealth-based intervention was found to be feasible and effective in helping patients with LARS relieve bowel dysfunction, improve their self-management behavior, and improve their QoL compared to usual care.
This study found that the EORTC-QLQ-C30 total score of the intervention group increased significantly more than that of the control group after the intervention, indicating that the mHealth-based remote interaction could improve the QoL of patients with LARS. These results can be attributed to multiple factors. First, uncontrollable changes in intestinal function, concerns about prognosis, and fear of the future make patients with LARS feel uncertain [ 37 ]. A sense of uncertainty influences a patient’s QoL [ 38 ]. Patients using the “e-bowel safety” applet can provide timely feedback on their problems to the medical staff and obtain solutions, which can effectively reduce the uncertainty of patients during home rehabilitation. Second, decreased bowel dysfunction severity positively affected the QoL [ 39 ]. Third, peer support reportedly enhances cancer adaptation and QoL [ 40 ]. The patients’ stories module offers a channel for communication and emotional support among patients with LARS. In this section, patients can share their experiences related to disease management or self-management and receive responses from their peers through comments.
As expected, the BSSBQ total score in the intervention group after the intervention was significantly higher than that in the control group. The findings supported our hypothesis that health-based remote interaction can enhance the self-management behavior of patients with LARS. After the intervention, the results of enhanced self-management behavior were consistent with a previous face-to-face 6-month self-management program study for LARS, which may indicate that mHealth-based remote interaction may yield intervention effects on self-management behavior similar to those observed in face-to-face interventions [ 41 ]. However, a more significant effect was observed at 3-month follow-up. This may be because monthly motivational interviews help patients adopt positive health behaviors and improve their self-management awareness [ 42 ]. Moreover, current web-based self-management information on LARS is overly intricate for patients, and the information fails to meet the patient’s needs [ 43 ]. The strength of our “e-bowel safety” applet is the credibility of the information provided and medical consultation from experts, which can meet the information needs of patients. Finally, our team members created an individualized self-management plan for each participant in the intervention group and reminded them to follow the plans on the applet, which ensured that the patients developed good habits.
Consistent with previous studies [ 41 ], this study found that the intervention group demonstrated a more significant decline in the LARS score than the control group. The LARS score also showed significant time effects, indicating that the patient’s bowel dysfunction changed significantly during the 6-month period. This may be because our team members guided patients in rehabilitation exercises and diet adjustments, which have been proven effective in improving bowel dysfunction [ 44 - 46 ]. Meanwhile, the severity of bowel dysfunction decreased over time [ 13 ].
Unlike those of previous studies, our findings indicated that mHealth-based remote interaction management intervention could improve the social support levels in the short term; however, sustaining a stable long-term effect on social support was not realized [ 47 ]. The patients in the study might have used the “e-bowel safety” applet only for 3 months, and the impact of the intervention on social support may not yield a residual advantage at 3-month follow-up. Furthermore, most patients’ physical and social functions gradually stabilized at 6 months. Our “e-bowel safety” applet focuses on intensive support for symptom management and lacks support knowledge for patients when symptoms plateau, which should be refined in future studies to achieve long-term effects.
In this study, MI was used to stimulate behavioral change and maintenance. The dual intervention of mHealth and MI promotes effective engagement and motivation for health behavior changes. Nearly all the patients (55/60) successfully completed the 3-month intervention and the follow-up during the intervention process, signifying that the mHealth-based remote interaction management intervention is feasible and acceptable. In addition, none of the patients in the intervention group experienced adverse consequences caused by the intervention, indicating that the intervention was safe.
This study has some limitations. First, this study enrolled patients from a tertiary hospital in China, which restricts the generalizability of our results. In the future, we will recruit patients from more hospitals to confirm our research findings. Second, patients were subjected to a limited 3-month follow-up period, thereby restricting our assessment of the enduring effects of the mHealth-based remote interaction management intervention on self-management behavior and QoL. Finally, patients were required to use WeChat and smartphones, which presents the potential for selection bias.
The mHealth-based remote interaction management intervention effectively enhanced the self-management behavior and QoL of patients with LARS, and the impact remained consistent during the 3-month follow-up. Bowel dysfunction also significantly improved throughout the entire research process. This study suggests that mHealth intervention could provide an effective and new option for many patients with LARS. Multicenter studies are necessary to establish the generalizability and effectiveness of these interventions.
This work was supported by the 2021 Anhui Higher Education Institutions Provincial Quality Engineering Project (grant 2021jyxm0718) and the Scientific Research and Cultivation project of the School of Nursing, Anhui Medical University (grant hlqm12023055).
None declared.
CONSORT-EHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile HEalth Applications and onLine TeleHealth) checklist (version 1.6.1).
The interview guide of motivational interviewing.
Results of pilot experiment.
The results of subgroup analysis.
Comments and attitudes towards intervention of intervention group.
Bowel Symptoms Self-Management Behaviors Questionnaire |
Consolidated Standards of Reporting Trials |
European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 |
low anterior resection syndrome |
mobile health |
motivational interviewing |
Perceived Social Support Scale |
quality of life |
randomized controlled trial |
sphincter-preserving surgery |
Edited by A Mavragani; submitted 24.10.23; peer-reviewed by V Sun, C Thomson; comments to author 13.03.24; revised version received 07.05.24; accepted 03.06.24; published 13.08.24.
©Peng Zhou, Hui Li, Xueying Pang, Ting Wang, Yan Wang, Hongye He, Dongmei Zhuang, Furong Zhu, Rui Zhu, Shaohua Hu. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 13.08.2024.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research (ISSN 1438-8871), is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.
George Lundberg, MD
Disclosure: George D. Lundberg, MD, has disclosed no relevant financial relationships.
The Heritage Foundation sponsored and developed Project 2025 for the explicit, stated purpose of building a conservative victory through policy, personnel, and training with a 180-day game plan after a sympathetic new President of the United States takes office. To date, Project 2025 has not been formally endorsed by any presidential campaign.
More than 100 conservative organizations are said to be participating. More than 400 conservative scholars and experts have collaborated in authorship of the mandate's 40 chapters. Chapter 14 of the "Mandate for Leadership" is an exhaustive proposed overhaul of the Department of Health and Human Services (HHS), one of the major existing arms of the executive branch of the US government.
The mandate's sweeping recommendations, if implemented, would impact the lives of all Americans and all healthcare workers, as outlined in the following excerpts.
MORE RESULTS
Home / Research & Medical / Metabolic / Diagnosing and Managing Hyperinsulinemia in Horses
Hyperinsulinemia is an important risk factor for laminitis in horses. Identifying insulin-resistant horses is key to preventing and managing laminitis. At the 2024 Tex Cauthen Farrier/Veterinarian/Researcher Seminar in Kentucky, Andrew van Eps, BVSc, PhD, MACVSc, Dipl. ACVIM, professor at University of Pennsylvania’s New Bolton Center, proposed multiple ways to evaluate a horse’s insulin levels.
Adiponectin (high molecular weight) is a protective adipokine from fat cells that improves insulin sensitivity and serves as a metabolic health marker. It is insulin-sensitizing and acts as an AMP-activated protein kinase (AMPK) agonist to inhibit mTor (which promotes the activation of insulin receptors and insulin-like growth factor receptors). Adiponectin might also protect tissues against cell stress—both endoplasmic reticulum and oxidative—and ischemia. Low serum adiponectin is associated with insulin dysregulation. It is also an independent risk factor for laminitis when its levels are low.
Adiponectin doesn’t fluctuate with stress or feeding. Researchers have compared adiponectin levels in horses fed carbohydrate or fat-based diets for more than 20 weeks. Horses in both categories developed high leptin levels, but only the carbohydrate-fed horses developed insulin dysfunction (ID) along with low serum adiponectin.
Veterinarians can evaluate baseline fed insulin as a random test in nonfasted horses. This test involves taking blood samples two hours after feeding or pasture access. It is a reasonable predictor of insulin dysregulation. Insulin levels of 25 mIU/ml are 80% sensitive and 85% specific for ID.
Van Eps reported that a combination of high insulin and low adiponectin is a red flag. Normal insulin and low adiponectin is a warning sign, but it is possible that a single sample doesn’t account for insulin fluctuations throughout the day. Doing a sugar or feed challenge might identify hyperinsulinemia with a potentially higher risk of laminitis. It is important to monitor and screen these horses.
Attempts to increase adiponectin through diet and exercise are not particularly helpful, especially because certain breeds are inherently at risk for low adiponectin, including Welsh ponies and donkeys. Thoroughbreds have a lower risk. Sick and hospitalized horses with laminitis reportedly also have low adiponectin levels.
Dietary changes, along with exercise and pasture management, are critical to controlling hyperinsulinemia in horses. Besides feeding forage with < 10% nonstructural carbohydrates (NSC), van Eps emphasized the benefits of soaking hay to reduce sugars and starches in the feed. Study results indicate soaking grass hay in five gallons of room-temperature water for up to two hours provides the maximum benefit for controlling sugar and starch levels. Alfalfa is lower in NSCs than grass hay, but soaking alfalfa is also helpful.
The European College of Equine Internal Medicine offers guidelines for dietary control of obesity in horses , which is a risk factor for insulin dysregulation. Rather than feeding horses 2% of their bodyweight in forage, owners can feed 1.4-1.7% of their bodyweight to achieve weight loss. These horses should not eat supplementary feeds. Exercise can also help horses lose weight.
Van Eps described available medications for controlling insulin levels in horses:
Researchers evaluated pioglitazone’s effects on adiponectin concentrations and insulin responses after the oral sugar test. Fifteen horses and ponies received 2 mg/kg orally of HMW pioglitazone once daily for 28 days. The subjects were grouped into horses, ponies, and insulin dysregulated (ID) animals. Oral sugar tests were performed before and after treatment to measure adipokines on Days 0, 14, and 28. The researchers measured insulin and glucose levels on Days 14 and 28. The results indicated significant decreases in insulin responses to the oral sugar test at 90 and 120 minutes, especially after pioglitazone treatment in the ponies and insulin-dysregulated individuals. “Adiponectin concentrations were significantly increased in all groups after pioglitazone treatment,” the authors stated. They concluded that this medication provides positive effects for treating metabolic derangements in horses with ID and equine metabolic syndrome.
Legere RM, Taylor DR, Davis JL, et al. Pharmacodynamic Effects of Pioglitazone on High Molecular Weight (HMW) Adiponectin Concentrations and Insulin Response After Oral Sugar in Equids. J Equine Vet Sci. Nov 2019; DOI: 10.1015/j.jevs.2019.102797
Stay in the know! Sign up for EquiManagement’s FREE weekly newsletters to get the latest equine research, disease alerts, and vet practice updates delivered straight to your inbox.
" * " indicates required fields
If you're on a Galaxy Fold, consider unfolding your phone or viewing it in full screen to best optimize your experience.
Published on Aug. 13, 2024
By: Dana George
Medical debt is no joke. Not only does the U.S. have the most expensive healthcare of any country, but medical debt is the No. 1 reason Americans file for bankruptcy, according to the American Medical Association (AMA).
Medical debt can also play havoc on a person's budget and personal finances, causing their credit score to plummet and making it harder for them to get credit when it's needed.
Two years ago, Experian, Equifax, and TransUnion -- the big three national credit reporting agencies -- announced that they were removing medical debts under $500 from credit reports. Removing relatively small medical debts certainly represented a step in the right direction for many U.S. consumers. Still, 15 million Americans were left with $49 billion worth of medical bills on their credit reports.
The Biden administration plan would force credit reporting agencies to go much further.
What matters is how erasing medical debt from credit reports will impact the everyday American. If your credit report currently lists medical expenses among your unpaid debts, here's how this change may benefit you.
The Consumer Financial Protection Bureau (CFPB) estimates that Americans with medical debt on their credit reports will see their credit scores rise by an average of 20 points once medical debt is removed. A sudden 20-point boost may be just what some households need to qualify for a loan or open a credit card.
While credit reports only show that you have outstanding medical debt and don't spell out the specifics, anyone who sees a copy of your report can tell that you (or someone in your household) have been ill.
That may not matter if you're applying for a car loan, but it could impact whether a potential employer checking your credit report decides to take a chance on you or a landlord believes you can pay your rent every month. In other words, it raises questions that should never be raised.
The proposed rule would prevent lenders from repossessing medical equipment like wheelchairs if you can't repay a loan.
According to the CFPB, under the current system, medical debt collectors use the credit reporting system to force people to pay debts -- some of which they may not owe. CFPB reports that many debt collectors use a practice known as "debt parking."
Here's how it works: Debt collectors purchase medical debt at a discount. They then place the debt on your credit report, typically without your knowledge. It's only when you apply for credit that you discover that medical debt is holding you back from loan approval.
If you really need that loan, you may feel forced to pay the medical bill just to get it off your credit report and improve your credit score. Once medical debt no longer shows up on credit reports, bill collectors can no longer use this manipulative tactic.
There are undoubtedly benefits associated with having medical debts removed from your credit report, but you will still be responsible for repaying the debt . The point of removing the debt from your report is to make it easier for you to carry on with your financial responsibilities and get back on your financial feet.
The CFPB will continue to accept comments and feedback on the Biden administration proposal through Aug. 12, 2024. If all goes well, the rule will be finalized early next year.
This credit card is not just good – it’s so exceptional that our experts use it personally. It features a lengthy 0% intro APR period, a cash back rate of up to 5%, and all somehow for no annual fee! Click here to read our full review for free and apply in just 2 minutes.
Dana is a full-time personal finance writer, with more than two decades of experience. Her focus is on helping readers feel less alone as they navigate their personal finances and offering actionable insights.
Share this page
We're firm believers in the Golden Rule, which is why editorial opinions are ours alone and have not been previously reviewed, approved, or endorsed by included advertisers. The Ascent, a Motley Fool service, does not cover all offers on the market. The Ascent has a dedicated team of editors and analysts focused on personal finance, and they follow the same set of publishing standards and editorial integrity while maintaining professional separation from the analysts and editors on other Motley Fool brands.
Related Articles
By: Cole Tretheway | Published on June 7, 2024
By: Lyle Daly | Published on June 5, 2024
By: Christy Bieber | Published on June 5, 2024
By: Lyle Daly | Published on June 4, 2024
The Ascent is a Motley Fool service that rates and reviews essential products for your everyday money matters.
Copyright © 2018 - 2024 The Ascent. All rights reserved.
E-learning pills on immunotherapy in urothelial carcinoma: the e-pimuc program for continuing medical education.
The final, formatted version of the article will be published soon.
You have multiple emails registered with Frontiers:
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
The high incidence and mortality rates of urothelial carcinoma mean it remains a significant global health concern. Its prevalence is notably pronounced in industrialized countries, with Spain registering one of the highest incidences in Europe. Treatment options are available for various stages of bladder cancer. Moreover, the management landscape for this disease has been significantly transformed by the rapid advances in immunotherapy. Healthcare professionals who diagnose, treat, and follow up with bladder cancer patients need comprehensive training to incorporate these advances into their clinical practice. To bridge these knowledge gaps, we set up the E-PIMUC program to educate healthcare professionals on bladder cancer management and specifically immunotherapy.Methods: E-PIMUC used an innovative microlearning methodology comprising bitesize learning pills that support efficient acquisition of specialized expertise. We used a mixed methods, quantitative and qualitative approach to assess the success of the E-PIMUC program. Data collection encompassed prepost testing, participation metrics, satisfaction surveys, and self-perceived performance assessments.Results: A total of 751 healthcare professionals enrolled in the program. Of these, 81.0% actively engaged with the content and 33.2% passed all tests and were awarded the course certificate and professional credits. The course received satisfaction ratings of 94.3% to 95.1% and significantly improved the declarative knowledge of participants who had a range of professional profiles (p<0.001). Participants reported increased confidence in applying immunotherapy principles in their practice (average improvement of 1.4 points). Open-ended responses also underscored participants' perceived benefits, including expanded knowledge and enhanced patient interaction skills.The E-PIMUC program provided effective, comprehensive, cutting-edge training on bladder cancer management, particularly on the use of immunotherapy in this area of oncology. The high participation rates, positive satisfaction scores, substantial knowledge enhancement, and improved self-perceived performance, are all testament to the program's success. E-PIMUC was endorsed by regulatory bodies as a trusted educational resource in urothelial carcinoma management. What is more, complementary initiatives brought together patients and medical experts to foster a holistic, patient-centered approach to the complexities of bladder cancer care.
Keywords: urothelial carcinoma, Bladder cancer, Immunotherapy, healthcare professionals, Continuing medical education, Microlearning, e-learning pills
Received: 02 Feb 2024; Accepted: 12 Aug 2024.
Copyright: © 2024 Romero-Clarà, Madrid, Pardo, Ruiz De Porras, Etxaniz, Moreno-Alonso and Font. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Olga Romero-Clarà, Catalan Institute of Oncology, Barcelona, 08908, Catalonia, Spain Juan Carlos Pardo, Catalan Institute of Oncology, Barcelona, 08908, Catalonia, Spain Olatz Etxaniz, Catalan Institute of Oncology, Barcelona, 08908, Catalonia, Spain Deborah Moreno-Alonso, Catalan Institute of Oncology, Barcelona, 08908, Catalonia, Spain Albert Font, Catalan Institute of Oncology, Barcelona, 08908, Catalonia, Spain
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.
IMAGES
COMMENTS
Medical research (or biomedical research ), also known as health research, refers to the process of using scientific methods with the aim to produce knowledge about human diseases, the prevention and treatment of illness, and the promotion of health.
Clinical research is the comprehensive study of the safety and effectiveness of the most promising advances in patient care. Clinical research is different than laboratory research. It involves people who volunteer to help us better understand medicine and health. Lab research generally does not involve people — although it helps us learn which new ideas may help people.
This is done on the basis of a selective literature search concerning study types in medical research, in addition to the authors' own experience. Results Three main areas of medical research can be distinguished by study type: basic (experimental), clinical, and epidemiological research.
Whereas basic research is looking at questions related to how nature works, translational research aims to take what's learned in basic research and apply that in the development of solutions to medical problems. Clinical research, then, takes those solutions and studies them in clinical trials. Together, they form a continuous research loop that transforms ideas into action in the form of ...
Health research entails systematic collection or analysis of data with the intent to develop generalizable knowledge to understand health challenges and mount an improved response to them. The full spectrum of health research spans five generic areas of activity: measuring the health problem; understanding its cause(s); elaborating solutions; translating the solutions or evidence into policy ...
Research conducted for the purpose of contributing towards science by the systematic collection, interpretation and evaluation of data and that, too, in a planned manner is called scientific research: a researcher is the one who conducts this research. The results obtained from a small group through scientific studies are socialised, and new ...
Medical research involves research in a wide range of fields, such as biology, chemistry, pharmacology and toxicology with the goal of developing new medicines or medical procedures or improving ...
Clinical research occurs in many formats and can involve anyone. Learn how you can participate and contribute to medical advances.
To gain more knowledge about illness and how the human body and mind work, volunteers can help researchers answer questions about health in studies of an illness. Studies might involve testing new drugs, vaccines, surgical procedures, or medical devices in clinical trials. For this reason, health research can involve known and unknown risks. To answer questions correctly, safely, and according ...
Clinical research is a branch of medical research that involves people and aims to determine the effectiveness ( efficacy) and safety of medications, devices, diagnostic products, and treatment regimens intended for improving human health.
It seems to happen almost every day - you hear about the results of a new medical research study. Learn how to evaluate medical findings.
Biomedical research is the broad area of science that looks for ways to prevent and treat diseases that cause illness and death in people and in animals. This general field of research includes many areas of both the life and physical sciences. Utilizing biotechnology techniques, biomedical researchers study biological processes and diseases ...
Clinical research describes many different elements of scientific investigation. Simply put, it involves human participants and helps translate basic research (done in labs) into new treatments and information to benefit patients. Clinical trials as well as research in epidemiology, physiology and pathophysiology, health services, education ...
1. The organized quest for new knowledge and better understanding (e.g., of the natural world or determinants of health and disease). Five types of research are recognized: observational (empiric), analytic, experimental, theoretic, applied. 2. To conduct such scientific inquiry.
Medical research has become an important part of the health care industry, and advances in technology have made it possible for much of it to be done on an outpatient basis, meaning that investigators sometimes don't need to do extensive studies in a research facility. The field of medical research is one of the most […]
Clinical research includes all research that involves people. Types of clinical research include: Epidemiology, which improves the understanding of a disease by studying patterns, causes, and effects of health and disease in specific groups. Behavioral, which improves the understanding of human behavior and how it relates to health and disease.
This chapter gives an overview of medical research endeavors, including the pre-eminent role of empiricism, the dominance of uncertainties, broad steps, and the essential ingredients of good research.
research the systematic, rigorous investigation of a situation or problem in order to generate new knowledge or validate existing knowledge. Research in health care takes place in a variety of areas and has many potential benefits; the areas include professional practice, environmental issues affecting health, vitality, treatments, theory development, health care economics, and many others ...
The general definition of research is, 'an investigation that is intentionally designed to help develop or contribute to knowledge'. When you add a medical purpose to 'research', the general definition stays the same, but the goal becomes more specific. Ultimately, the goal shifts to a focus on increasing medical knowledge, improving patient care, developing new medicines or procedures, and ...
(Clinical Research: A National Call to Action, November 1999) Clinical research is a component of medical and health research intended to produce knowledge valuable for understanding human disease, preventing and treating illness, and promoting health. Clinical Research embraces a continuum of studies involving interactions with patients, diagnostic clinical materials or data, or populations ...
Also called a clinical trial. (NCI) A clinical trial is a research study to answer specific questions about vaccines or new therapies or new ways of using known treatments. Clinical trials (also called medical research and research studies) are used to determine whether new drugs or treatments are both safe and effective.
New to clinical research? Learn the meaning of common industry acronyms and abbreviations including eCRF, IIT, PI, TMF, and more.
Definition of Terms The Mayo Clinic Institutional Review Board's definition of terms explains legal definitions related to research guidelines and the protection of human research subjects, including advocate, conflict of interest, emergency treatment, informed consent and more.
The success of current deep learning models depends on a large number of precise labels. However, in the field of medical image segmentation, acquiring precise labels is labor-intensive and time-consuming. Hence, the challenge of achieving a high-performance model via datasets containing noisy labels has attracted significant research interest. Some existing methods are unable to exclude ...
Background: People who undergo sphincter-preserving surgery have high rates of anorectal functional disturbances, known as low anterior resection syndrome (LARS). LARS negatively affects patients' quality of life (QoL) and increases their need for self-management behaviors. Therefore, approaches to enhance self-management behavior and QoL are vital.
Research is shedding light on different chromosomal make-ups and what advantages they may bring to sport.
Excerpts from Project 2025 demonstrate the likely impact of sweeping changes to US healthcare policy.
Hyperinsulinemia is an important risk factor for laminitis in horses. Identifying insulin-resistant horses is key to preventing and managing laminitis.
A new Biden administration proposal is aimed at medical bills that lower your credit score. Check out how the plan may benefit you.
The high incidence and mortality rates of urothelial carcinoma mean it remains a significant global health concern. ... This article is part of the Research Topic Education and training in ... What is more, complementary initiatives brought together patients and medical experts to foster a holistic, patient-centered approach to the complexities ...