<b>Automating COVID-19 Classification in Chest CT Scans Using Advanced CNN Architectures</b>

Abstract

This paper proposes an efficient and fully automated method for classifying COVID-19 using CT scan images of a patient's chest. The study utilizes the publicly available SARS-CoV-2 CT scan dataset, which contains 1252 CT scans positive for SARS-CoV-2 (COVID-19) infection, 1230 CT scans from SARS-CoV-2-negative patients, and a total of 2482 cross-sectional scans. This research explores various topologies designed to enhance the classification accuracy of convolutional neural networks, particularly when dealing with images containing small objects of interest. DenseNet169 is particularly effective in handling small-sized infection patterns commonly observed in COVID-19 cases, as it allows the model to analyze images at varying resolutions effectively without compromising small-object data integrity. Several approaches were evaluated, including VGG19, Xception, ResNet101, DenseNet169, and two custom models: a "custom vanilla" model based on the vanilla architecture and a "custom inception" model based on the inception architecture. Among these, DenseNet169 achieved the highest performance, attaining an impressive accuracy rate of 99.731%.