Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Deep learning model calibration for improving performance in class-imbalanced medical image classification tasks.

Sivaramakrishnan Rajaraman1, Prasanth Ganesan2, Sameer Antani1

  • 1National Library of Medicine, National Institutes of Health, Bethesda, MD, United States of America.

Plos One
|January 27, 2022
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

Survival Tree01:19

Survival Tree

178
Survival trees are a non-parametric method used in survival analysis to model the relationship between a set of covariates and the time until an event of interest occurs, often referred to as the "time-to-event" or "survival time." This method is particularly useful when dealing with censored data, where the event has not occurred for some individuals by the end of the study period, or when the exact time of the event is unknown.
 Building a Survival Tree
Constructing a...
178

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Identifying the Presence and Characteristics of Mid-Myocardial and Epicardial Fibrosis From Intracardiac Electrograms in Patients Undergoing Ventricular Arrhythmia Ablation Using a Transformer-Based Self-Supervised Classifier.

Circulation. Arrhythmia and electrophysiology·2026
Same author

Antithetic Sampling Enhanced Probabilistic Diffusion for Denoising Cardiac Time Series.

IEEE journal of biomedical and health informatics·2026
Same author

Primary Platinum-Resistant Ovarian Cancer: Clinicopathologic Correlates and Outcomes From a Multicenter Prospective Indian Registry.

JCO global oncology·2026
Same author

Efficacy of Metronomic Oral Capecitabine, Methotrexate and Cyclophosphamide in Locally Advanced Operable Oral Cavity Squamous Cell Carcinoma - A Phase II Study.

Indian journal of surgical oncology·2026
Same author

Multimodal Learning with Privileged Report Supervision for Generalizable Tuberculosis Detection on Chest Radiographs.

Journal of medical systems·2026
Same author

Oral Cancer Detection By Using Tabular Data Synthesis and Classification.

Proceedings ... ICDM workshops. IEEE International Conference on Data Mining·2026

Model calibration significantly improves deep learning performance in medical image classification at a default threshold of 0.5. However, calibration benefits diminish when using optimal thresholds derived from precision-recall curves.

Area of Science:

  • Medical Imaging
  • Artificial Intelligence
  • Machine Learning

Background:

  • Class imbalance in medical datasets poses challenges for deep learning models, biasing predictions towards majority classes.
  • Model calibration is a technique to address these biases, but its effectiveness requires further systematic analysis.

Purpose of the Study:

  • To systematically analyze the impact of model calibration on deep learning classifier performance for medical image classification.
  • To investigate how calibration effects vary with dataset imbalance, calibration methods, and classification thresholds.

Main Methods:

  • Evaluated model calibration on chest X-ray and fundus image datasets with varying degrees of class imbalance.
  • Compared performance using different calibration techniques and two classification thresholds: a default 0.5 and an optimal precision-recall (PR) curve-guided threshold.

Related Experiment Videos

  • Utilized various deep learning classifier backbones for the analysis.

    • Main Results:

    • Calibration significantly improved performance (p < 0.05) over uncalibrated probabilities at the default 0.5 threshold.
    • Performance gains from calibration were not statistically significant (p > 0.05) when using the PR-guided threshold.
    • These findings were consistent across both medical image modalities and varying imbalance levels.

    Conclusions:

    • Model calibration is beneficial for medical image classification tasks using the default 0.5 threshold, especially in imbalanced datasets.
    • The utility of calibration is threshold-dependent, offering limited advantages when optimal thresholds are employed.
    • Further research should consider threshold selection alongside calibration strategies for optimal model performance.