3D CNN with Attention Modules for Enhanced COVID-19 and Pneumonia Screening from CT Scans

During the COVID-19 outbreak, deep-learning techniques have been extensively studied for computer-assisted diagnosis. However, distinguishing COVID-19 from other types of pneumonia remains challenging. This research describes an automated chest CT imaging technique for COVID-19 pneumonia screening. CT scans provide valuable three-dimensional information, aiding in detecting COVID-19 symptoms. The limited availability of comprehensive 3D datasets has slowed the use of attention-based models in this field. To address this, we propose a novel 3D Convolutional Neural Network (CNN) with a residual block incorporating self-attention and spatial attention modules. Our primary goal is accurately classifying COVID-19 and pneumonia cases based on CT scans. Integrating attention modules enhances spatial correlation analysis across CT slices, enabling dynamic re-calibration of the CNN output to identify complex patterns. Training our <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3D$</tex> CNN on CT scan images from the CC-CCII dataset achieved an impressive accuracy rate of 99.42%. Subsequently, we used the CNN's prediction probabilities to construct feature vectors for each sample. By training various machine learning classifiers independently and combining them through soft voting, we achieved a remarkable 98.66% accuracy. This study highlights how our approach can improve pulmonary patient care and diagnostic processes, leveraging state-of-the-art deep learning techniques for accurate and efficient COVID-19 pneumonia screening.