VGG-AM: Towards a New Hybrid Medical Imaging Analysis Based on VGG Classification Model and Deep DATA Preparation
摘要
During the COVID-19 pandemic, the conventional RT-PCR method for diagnosing SARS-CoV-2 was limited by its processing time and accuracy rate. Consequently, thoracic CT images have been explored as a reliable and accurate alternative solution. However, analyzing a large volume of CT images by radiologists remains a major challenge. Deep learning solutions, specifically convolutional neural networks (CNN), have proven useful in radiology for image classification, object detection, semantic segmentation, and instance segmentation. In this study, DL approaches were evaluated for the automatic detection of COVID-19 on thoracic CT images using advanced deep networks such as XCEPTION, VGG, RESNET, and INCEPTION. A new advanced deep architecture called “VGG-advanced-Model” was also proposed, which improved upon the original VGG architecture that had achieved the best results. Various data preparation and preprocessing techniques were combined to enhance performance. The results demonstrate the effectiveness and efficiency of our method, achieving 100% precision, 99.52% accuracy, and 99% recall.