Related Experiment Video
Updated: Oct 12, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.4K
Improving Uncertainty Estimation With Semi-Supervised Deep Learning for COVID-19 Detection Using Chest X-Ray Images.
Saul Calderon-Ramirez1,2, Shengxiang Yang1, Armaghan Moemeni3
1School of Computer Science and InformaticsDe Montfort University Leicester LE1 9BH U.K.
IEEE Access : Practical Innovations, Open Solutions
|November 23, 2021
Summary
This study enhances COVID-19 detection from X-rays using semi-supervised learning for better uncertainty estimation. Incorporating unlabeled data significantly improves the reliability of AI diagnostic tools, aiding radiologists.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Machine Learning for Diagnostics
Background:
- Computer-aided diagnosis (CAD) systems for COVID-19 detection using chest X-rays require reliable uncertainty estimation for safe clinical application.
- High uncertainty in model predictions necessitates careful review by radiologists, highlighting the need for improved uncertainty quantification.
Purpose of the Study:
- To enhance uncertainty estimation in COVID-19 detection models by leveraging unlabeled data via the MixMatch semi-supervised framework.
- To evaluate the effectiveness of different uncertainty estimation techniques, including Softmax scores, Monte-Carlo dropout, and deterministic uncertainty quantification.
- To introduce a statistically robust metric, Jensen-Shannon distance, for assessing the reliability of uncertainty estimates.
Main Methods:
- Implementation of a COVID-19 detection system using chest X-ray images.
- Application of the MixMatch semi-supervised learning framework to incorporate unlabeled data.
- Testing and comparison of Softmax scores, Monte-Carlo dropout, and deterministic uncertainty quantification for uncertainty estimation.
- Utilizing Jensen-Shannon distance to evaluate the reliability of uncertainty estimates by comparing distributions of correct and incorrect predictions.
Main Results:
- Significant improvements in uncertainty estimation were observed when utilizing unlabeled data.
- The Monte Carlo dropout method demonstrated the best performance in enhancing uncertainty estimates.
- The proposed Jensen-Shannon distance metric proved effective in reliably comparing uncertainty estimation methods.
Conclusions:
- Semi-supervised learning with unlabeled data substantially improves the reliability of uncertainty estimation in AI-driven medical diagnostic tools.
- Monte Carlo dropout, combined with semi-supervised learning, offers a promising approach for robust uncertainty quantification in COVID-19 detection.
- The Jensen-Shannon distance provides a statistically sound method for evaluating the quality of uncertainty estimates in medical AI.
