Explainable and Lightweight Model for COVID-19 Detection Using Chest Radiology Images

TL;DR

This paper presents an explainable, lightweight CNN model for COVID-19 detection from chest radiology images, emphasizing interpretability and robustness across unseen datasets, with performance comparable to state-of-the-art models.

Contribution

The study introduces a novel CNN model with visualization techniques for COVID-19 detection, addressing generalization issues of existing models.

Findings

Model achieves comparable performance to state-of-the-art CNNs.

Grad-CAM visualizations help understand model predictions.

Model generalizes well on unseen datasets.

Abstract

Deep learning (DL) analysis of Chest X-ray (CXR) and Computed tomography (CT) images has garnered a lot of attention in recent times due to the COVID-19 pandemic. Convolutional Neural Networks (CNNs) are well suited for the image analysis tasks when trained on humongous amounts of data. Applications developed for medical image analysis require high sensitivity and precision compared to any other fields. Most of the tools proposed for detection of COVID-19 claims to have high sensitivity and recalls but have failed to generalize and perform when tested on unseen datasets. This encouraged us to develop a CNN model, analyze and understand the performance of it by visualizing the predictions of the model using class activation maps generated using (Gradient-weighted Class Activation Mapping) Grad-CAM technique. This study provides a detailed discussion of the success and failure of the proposed model at an image level. Performance of the model is compared with state-of-the-art DL models and shown to be comparable. The data and code used are available at https://github.com/aleesuss/c19.