Involution Receptive Field Network for COVID-19 Diagnosis

Abstract

COVID-19 is a new infectious illness that has been sweeping the globe since its emergence, causing severe pneumonia-related respiratory failure. From the Large COVID-19 CT scan slice dataset, the Community-Acquired Pneumonia (CAP), Normal, and COVID-19 Computed Tomography (CT) scan images are identified using the Involution Receptive Field Network. For better embedding representation in latent dimension for CT scan slices, N-pair contrastive loss is introduced during the training of the network. The proposed lightweight Involution Receptive Field Network-Medium (InRFNet-M) uses a Receptive Field structure to improve feature map extraction. It is spatially specific and channel-agnostic. The InRFNet-M model evaluation results reveal a high level of validation accuracy (99 percent).  With high accuracy and recall scores, the proposed InRFNet-M: Involution Receptive Field Network-Medium has demonstrated efficient classification.