Interpretable COVID-19 chest X-ray detection based on handcrafted feature analysis and sequential neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Deep learning enhances medical image analysis for COVID-19 detection in chest X-rays.
• Existing methods face challenges including limited interpretability, high computational costs, and data requirements.

Purpose of the Study:

• Introduce novel algorithms, DC-GLM and CAMSGNeT, to address limitations in deep learning for COVID-19 detection.
• Improve diagnostic precision, interpretability, and resource efficiency in medical image analysis.

Main Methods:

• Developed Dynamic Co-Occurrence Grey Level Matrix (DC-GLM) to capture key COVID-19 indicators like ground-glass opacity, consolidation, and fibrosis.
• Introduced Contextual Adaptation Multiscale Gabor Network (CAMSGNeT) for enhanced fine feature extraction using Contextual Adaptive Diffusion.
• Utilized a simple optimized sequential neural network for classification and feature importance analysis for interpretability.

Main Results:

• DC-GLM effectively captures coarse texture patterns and spatial correlations in chest X-rays.
• CAMSGNeT preserves fine details and enhances edge/texture characteristics, maintaining patterns like air bronchograms.
• The combined approach achieved high classification accuracy, reaching 98.27% and 100% on two distinct datasets.

Conclusions:

• The proposed DC-GLM and CAMSGNeT algorithms offer a more interpretable, precise, and resource-efficient solution for COVID-19 detection.
• Feature importance analysis enhances model transparency, aiding clinical decision-making.
• This methodology represents a significant advancement in AI-driven medical image analysis for infectious disease detection.