An Explainable AI driven Decision Support System for COVID-19 Diagnosis using Fused Classification and Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• The Reverse Transcription Polymerase Chain Reaction (RT-PCR) test is standard for COVID-19 detection but doesn't assess infection severity.
• Chest CT scans reveal COVID-19 lesions with high sensitivity, yet expert analysis is resource-intensive.
• A shortage of skilled radiologists during the pandemic highlighted the need for automated diagnostic support.

Purpose of the Study:

• To develop a real-time, explainable AI system (GGECS) for classifying COVID-19 severity and identifying lesions on chest CT scans.
• To enhance diagnostic capabilities by providing decision support for medical professionals.
• To leverage explainable AI techniques for transparent and interpretable model predictions.

Main Methods:

• Proposed a Guided Gradcam based Explainable Classification and Segmentation system (GGECS).
• Employed a Res2Net-inspired classification model and integrated explainable AI (GradCam, Guided GradCam) for CNN interpretability.
• Developed a segmentation model fusing VGG-16 and the classification network for precise localization of infected regions.

Main Results:

• The GGECS classification model achieved an overall accuracy of 98.51%.
• The segmentation model demonstrated a significant performance with an IoU score of 0.595.
• Explainable AI techniques successfully highlighted regions in CT scans crucial for the model's predictions.

Conclusions:

• GGECS provides an effective, explainable AI-driven decision support system for COVID-19 diagnosis using chest CT scans.
• The system demonstrates high accuracy in classification and reliable lesion localization, addressing the need for expert radiological interpretation.
• Explainable AI integration enhances trust and utility in clinical decision-making for infectious disease imaging.