Related Experiment Video
Updated: Oct 24, 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
Multi-label segmentation and detection of COVID-19 abnormalities from chest radiographs using deep learning
Ruchika Arora1, Indu Saini1, Neetu Sood1
1Department of Electronics and Communication Engineering, Dr. B. R. Ambedkar National Institute of Technology Jalandhar, Jalandhar 144011, India.
Optik
|August 16, 2021
Summary
A new deep learning model, CXAU-Net, accurately detects COVID-19 findings in Chest Radiographs (CXRs). This automated system enhances diagnostic capabilities for chest X-ray analysis.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Deep Learning for Diagnostic Radiology
- Computer-Aided Diagnosis (CAD) Systems
Background:
- The COVID-19 pandemic significantly increased the demand for Chest Radiographs (CXRs).
- Accurate and efficient detection of COVID-19 related pathologies in CXRs is crucial for timely diagnosis and patient management.
- Existing automated segmentation methods may require further enhancement for precise localization of multiple pathologies.
Purpose of the Study:
- To introduce a novel, fully automatic modified Attention U-Net (CXAU-Net) deep learning model for multi-class segmentation of COVID-19 findings in CXR images.
- To improve the precision and sensitivity of automated detection of various pathologies within CXRs.
- To evaluate the performance of the proposed model against state-of-the-art segmentation techniques.
Main Methods:
- Development of CXAU-Net, a modified Attention U-Net incorporating channel and spatial attention blocks for precise pathology localization.
- Integration of dilated convolution to enhance model sensitivity to foreground pixels and expand receptive fields.
- Implementation of a novel hybrid loss function combining area and size information for optimized model training.
Main Results:
- The model achieved high performance on the Chest X-ray 14 dataset, with average accuracy, DSC, and Jaccard index scores of 0.951, 0.993, 0.984 (image-based) and 0.921, 0.985, 0.973 (patch-based) for multi-class segmentation.
- Excellent results were obtained for binary-class segmentation on the JSRT CXR dataset, with average DSC and Jaccard index scores of 0.998 and 0.989, respectively.
- The proposed CXAU-Net demonstrated superior performance compared to existing state-of-the-art segmentation methods.
Conclusions:
- The CXAU-Net model offers a robust and accurate solution for automated multi-class segmentation of COVID-19 related findings in CXRs.
- The novel architectural components and hybrid loss function contribute to the model's enhanced diagnostic capabilities.
- This deep learning approach shows significant potential for improving the efficiency and accuracy of radiological assessments in clinical practice.
