GM-CBAM-ResNet: A Lightweight Deep Learning Network for Diagnosis of COVID-19

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• COVID-19 poses significant health risks, including potential cardiac damage.
• Electrocardiograms (ECGs) are a cost-effective, non-invasive tool for assessing heart health.

Purpose of the Study:

• To develop a lightweight deep learning model for diagnosing COVID-19 using ECG images.
• To evaluate the performance, complexity, and interpretability of the proposed model against existing methods.

Main Methods:

• Proposed a novel lightweight deep learning network, GM-CBAM-ResNet, integrating Ghost Modules (GM) and Convolutional Block Attention Modules (CBAM) into ResNet.
• Compared GM-CBAM-ResNet with ResNet, GM-ResNet, and CBAM-ResNet using the 'ECG Images dataset of Cardiac and COVID-19 Patients'.
• Assessed model complexity via parameter count and interpretability using Gradient-weighted Class Activation Mapping (Grad-CAM).

Main Results:

• GM-CBAM-ResNet19 reduced model parameters by 45.4% compared to ResNet19.
• Achieved approximately 5% higher diagnostic accuracy than ResNet19 with reduced model complexity.
• Interpretability analysis confirmed CBAM's ability to mitigate background interference, enhancing diagnostic accuracy.

Conclusions:

• GM-CBAM-ResNet provides a lightweight and accurate solution for COVID-19 diagnosis from ECG images.
• The model demonstrates practical value for rapid and efficient deployment in clinical settings.
• Integration of GM and CBAM modules enhances deep learning model performance for medical image analysis.