Self-Supervised Contrastive Learning for Covid-19 Classification from Computed Tomography Images

Abstract

Computer-aided diagnosis (CAD) emerges as an exhaustive diagnostic tool in the Covid-19 pandemic outbreak and is enormously investigated for automatic and more accurate detections. Artificial intelligence (AI) based radiographic images (Computed Tomography, X-Ray, Lung Ultrasound) interpretation improves the overall diagnosis efficiency of Covid-19 infections. In this paper, CAD based deep meta learning approach has been discussed for automatically quick analysis of chest computed tomography (CT) images regarding the early detection of corona virus (Covid-19) presence inside a subject. We incorporated a self-supervised contrastive-learning neural network for unbiased feature representation and classifications using fine-tuned pre-trained Inception module on 28203 chest CT images. This trainable multi-shot end-to-end deep learning architecture is validated on public dataset of normal and covid-19 CT images obtaining normalized accuracy of 0.9708. Results verify our model to be able enough to assist radiologists and specialists in screening and correct diagnosis of Covid-19 patients in less span of time.