Multi-task driven explainable diagnosis of COVID-19 using chest X-ray images.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology and Diagnostic Imaging
• Infectious Disease Diagnostics

Background:

• Global surge in COVID-19 cases necessitates increased testing capacity.
• Limited availability of RT-PCR kits and processing centers in remote areas poses a challenge.
• Chest X-ray (CXR) emerges as a viable, accessible screening modality.

Purpose of the Study:

• To develop an automated, end-to-end deep learning network for COVID-19 screening using CXR.
• To enhance model explainability through semantic segmentation of relevant lung regions.
• To create and release annotated datasets for COVID-19 symptom identification in CXRs.

Main Methods:

• Development of the COVID-19 Multi-Task Network (COMiT-Net) for automated screening.
• Implementation of semantic segmentation for identifying and highlighting COVID-19 features in CXRs.
• Manual annotation of lung regions and COVID-19 symptoms in CXRs from multiple datasets (ChestXray-14, CheXpert, COVID-19 dataset).

Main Results:

• COMiT-Net demonstrates high performance in COVID-19 screening from CXR images.
• The network achieves 96.80% sensitivity at 90% specificity.
• Semantic segmentation provides explainable insights into the model's predictions.

Conclusions:

• COMiT-Net offers a reliable, automated solution for COVID-19 screening using chest X-rays.
• The model's explainability through segmentation aids clinical interpretation.
• The released annotated datasets will support further research in AI-driven medical image analysis for infectious diseases.