Standard Training Dataset vs. Different Testing Dataset to Compare Deep Learning Architectures Models in Diagnosing COVID-19

The Coronavirus-2 disease epidemic has spurred global interest in adopting architectures for deep learning, especially the deep convolutional neural network -DCNN, often developing high-performance deep CNN models that require access to additional data for training. Unfortunately, such data is frequently unavailable. In this paper, we present a novel standardized dataset designed for training deep CNN models. This dataset encompasses chest CT scan images, with both COVID-19 and normal cases, by two distinct input sample sizes. Additionally, we employ this dataset to evaluate the effectiveness of recent DCNN techniques in diagnosing new Coronavirus disease, comparing it with the SARS-CoV-2 dataset. Notably, we observe variations in the behavior of DCNN models when using different datasets for training and testing, emphasizing the model’s dependency on dataset characteristics. Furthermore, our findings indicate that the proposed approach exhibits exceptional performance in detecting Coronavirus disease. Consequently, this approach holds promise for integration into medical diagnosis systems reliant on image analysis, offering potential advantages for resource-constrained computing devices, thus mitigating the need for expensive GPUs. In the forthcoming study, we propose applying this approach to evaluate the effectiveness of various classification techniques suggested by other studies to generate substantial CT imaging datasets for COVID-19 research.