Advanced Deep Learning Models for COVID-19 Prediction: A Multi-convolutional Neural Network Perspective

The novel coronavirus (COVID-19), is a very consequential illness that has a significant influence on the respiratory system and several physiological systems inside the human body. The virus originated in Wuhan, China, in December 2019 and then escalated into a worldwide pandemic, swiftly disseminating around the globe. In order to mitigate the transmission of viral infections, it is imperative to promptly detect those who have tested positive and promptly administer appropriate medical interventions to those who have been infected. The demand for COVID-19 testing kits has experienced a significant increase, resulting in a depletion of supplies in several developing nations due to the continuous emergence of new cases. In this particular instance, recent studies have demonstrated the efficacy of radiology imaging techniques, including X-ray and CT scan, in the detection of COVID-19. CT scan images provide crucial insights into the pathology associated with the COVID-19 virus. To address these challenges, many convolutional neural network (CNN) models have been suggested, including traditional convolutional, and Separable Convolution. These models aim to provide precise and efficient illness prediction, hence aiding in the alleviation of the scarcity of testing kits. The design has three convolutional layers, each consisting of 32 filters with a kernel size of 3 × 3. Additionally, the pooling size is set at 2 × 2. The architecture also includes a fully connected layer with 1024 units. In order to conduct a performance evaluation, a dataset consisting of 11,000 CT Scan images from the COVID-19 CT-Scan dataset was utilized. The assessment involved implementing a three-fold validation procedure, specifically employing three different k-fold validation techniques: threefold, fivefold, and tenfold. The experimental findings indicate that the tenfold cross validation model had superior performance in classifying the COVID-19 illness compared to the threefold and fivefold cross validation models. The accuracy rates achieved by the tenfold cross validation model were 94.85, 96.85, and 97.18% correspondingly for the two aforementioned models.