Deep Learning-Based COVID-19 Detection in CT Images.

COVID-19, a highly contagious viral disease, has had a significant global impact, affecting a substantial number of individuals in Algeria. This study develops and evaluates a deep learning model for COVID-19 detection in chest CT images. The dataset utilized in this study consists of 4986 images validated by RT-PCR tests from Tlemcen Hospital, Algeria, as well as an additional dataset focused on SARS-CoV-2 infection. The latter includes 1252 CT scans positive for COVID-19 and 1230 CT scans from non-infected patients, totaling 2482 scans. These data were obtained from medical institutions in São Paulo, Brazil. This initiative aims to facilitate AI research for identifying SARS-CoV-2 infection through CT scan analysis and encourages collaboration within the global scientific community. Leveraging DenseNet-201, known for its robust performance on ImageNet, transfer learning techniques were applied. An interpretable model DenseNet-201 algorithm was proposed to enhance diagnostic transparency. Experimental results on the test set demonstrated strong performance metrics: Accuracy of 98.12%, Precision of 97.24%, Specificity of 97.88%, Recall of 98.44% and F1-score of 98.35%. During training, the model achieved high accuracy (99.48% on training, 98.12% on validation) with minimal loss (0.0181). These findings underscore the clinical potential of the model in accurately diagnosing and monitoring COVID-19 patients.