Evaluating Pretrained Deep Learning Models for COVID-19 Detection in Medical Imaging

COVID-19 is a respiratory disease for which reverse transcription-polymerase chain reaction (RT-PCR) is the standard detection method. This study introduces a hybrid deep learning approach to support the diagnosis of respiratory abnormalities, using COVID-19 as a case study. Multiple datasets were utilized, including the COVID-19 Radiography Dataset (Kaggle), the SARS-CoV-2 CTScan Dataset, and proprietary data from the King Faisal Medical Complex. Seven pre-trained deep learning models—VGG16, VGG19, MobileNet, DenseNet201, ResNet50, InceptionResNetV2, and InceptionV3—were evaluated for their performance. InceptionResNetV2 and InceptionV3 achieved perfect accuracy (100%) on the COVID-19 Radiography Dataset and 99% accuracy on the SARS-CoV-2 CT-Scan Dataset. Additionally, augmentation techniques were explored to improve model robustness. This research contributes significantly to medical imaging, offering practical insights for advancing diagnostic and monitoring practices for COVID-19 and other respiratory conditions.