Related Experiment Video
Updated: Aug 7, 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
14.3K
Deep Learning Algorithms with LIME and Similarity Distance Analysis on COVID-19 Chest X-ray Dataset
Kuan-Yung Chen1, Hsi-Chieh Lee2, Tsung-Chieh Lin2
1Department of Radiology, Chang Bing Show Chwan Memorial Hospital, Changhua 505, Taiwan.
Summary
This study enhances COVID-19 detection using deep learning on chest X-rays. By focusing on relevant lung regions and analyzing feature similarity, the model achieved high accuracy, improving diagnostic confidence.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Science
Background:
- COVID-19 remains a significant global health threat.
- Machine learning, particularly deep learning, shows promise in analyzing medical images for disease detection.
- Chest X-rays are a common tool for diagnosing respiratory illnesses, including COVID-19.
Purpose of the Study:
- To investigate the effectiveness of deep learning algorithms for COVID-19 detection using chest X-rays.
- To improve the interpretability and performance of deep learning models by focusing on relevant image regions and employing similarity analysis.
- To develop a more flexible and accurate diagnostic approach for COVID-19.
Main Methods:
- Utilized Local Interpretable Model-agnostic Explanations (LIME) to identify important image regions.
- Employed U-Net segmentation to create regions of interest (ROIs), masking non-lung areas.
- Applied similarity analysis to feature spaces for outlier detection and confidence assessment.
- Evaluated model performance using metrics such as accuracy, sensitivity, precision, and F1 score.
Main Results:
- Achieved high detection performance for COVID-19 on chest X-rays: 95.5% overall accuracy, 98.4% sensitivity, 94.7% precision, and 96.5% F1 score.
- Demonstrated the utility of ROIs in preventing model distraction by irrelevant features.
- Similarity analysis provided objective confidence references for diagnostic inferences.
- Identified low-accuracy subspaces requiring targeted improvements.
Conclusions:
- The proposed deep learning approach, incorporating ROI segmentation and similarity analysis, significantly enhances COVID-19 detection accuracy from chest X-rays.
- Focusing on specific feature subspaces and potentially deploying dedicated classifiers can lead to more robust and flexible diagnostic tools.
- This method offers a promising alternative to rigid end-to-end black-box models for medical image analysis.
