Deep Learning Based Early Diagnosis for SARS-CoV-2 Using Chest X-Ray Images

Although Vaccines seem like the first ray of hope in a long time, it is still far too early to assume that this pandemic calamity is over. The corona virus is a persistent and highly contagious disease which can evolve faster than most self-acclaimed SoundCloud Rappers’ dying careers. Fortunately, computed tomography (CT) and X-Ray chest images have been proven to be very effective in diagnosing pneumonia. Although, CT scans are more accurate, they are slower, less available and more expensive than X-Ray. Chest X-Ray diagnosis requires a highly experienced medical expert, though. The popularity of Deep Learning has been sky rocketing ever since 2012’s ImageNet Large Scale Visual Recognition Challenge (ILSVRC). Accordingly, deep learning techniques can provide an alternate computer aided diagnosis given that enough data is available for the machine learning process. This paper proposes 2 early COVID19 diagnosis deep learning models through transfer learning using frontal X-ray Chest Images, one for binary (as COVID19 and Normal) and another for multi-class (as COVID19, Normal and Viral Pneumonia) classification. The models are proposed after a comparative study on the performance of several state of the art Convolutional Neural Networks is made on both classification types. The images’ quality is first enhanced with Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm and augmented. After having its Hyper Parameter optimized, each model is fine tuned to fit the data. The recall, precision, specificity, f1 Score, and accuracy are used to evaluated the performance of the models. The results indicate that the fined tuned VGG16 performed the best in multi-class classification with 96.7% and 97.8% COVID19 f1 score and testing accuracy respectively. In binary classification, ResNet50 displayed superior results with COVID19 f1 score and testing accuracy of 96.4% and 98.3% respectively.