Transfer learning with fine-tuned deep CNN model for COVID-19 diagnosis from chest X-ray images

The COVID-19 pandemic has had a significant impact on people's lives, necessitating accurate detection and early diagnosis to control the dissemination of virus. Reverse transcription-polymerase chain reaction (RT-PCR) is the most prevalent diagnostic strategy, but its accuracy is influenced by various factors such as sample collection, timing, and processing. Deep Convolutional Neural Networks (DCNNs) have shown great promise in medical image analysis and are consequently being utilized for the diagnosis of COVID-19 from radiographic images. This study evaluates the effectiveness of different convolutional neural network (CNN) architectures with optimum hyperparameters for COVID-19 diagnosis using publicly available chest radiography datasets. The evaluated models included both CNN architectures built from scratch and pre-trained CNN architectures, such as residual network (ResNet-50), visual geometry group (VGG-16), VGG-19, Inception-V3, and MobileNet-V2. The experimental results demonstrate that MobileNet-V2 achieved 96% accuracy, precision, recall, F1 score, and area under the curve (AUC), making it a prospective and acceptable model for COVID-19 diagnosis. In contrast to existing models, the proposed model's evaluation also includes an assessment of network training time and memory consumption. The study also describes the web deployment of a deep CNN-based computer-aided diagnosis (CAD) system that can assist doctors in diagnosing COVID-19 faster, more accurately, and more consistently. This advancement leads to better patient outcomes and improved efficiency within the healthcare system.