Application and development direction of deep learning in COVID-19 identification based on Computed Tomography images

Abstract

Caused by the novel coronavirus SARS-CoV-2, COVID-19 is highly contagious via respiratory droplets from sneezing, coughing, or talking, and it can lead to severe respiratory issues, organ failure, and death. Early detection, treatment, and isolation of those at risk help slow its spread, it has challenged traditional diagnostic methods like RT-PCR due to limitations in sensitivity. CT imaging, aided by deep learning models, offers advantages in the early detection of lung abnormalities. This paper reviews the use of deep learning in analyzing CT images for COVID-19 diagnosis, highlighting advancements like image segmentation with U-Net and FPN, it also tracks the evolution of deep learning models in this domain, starting from initial applications focused on image classification and recognition to later advancements incorporating techniques like U-Net for image segmentation and feature pyramid networks. Novel techniques like multi-task learning and quantitative analysis show promise in improving accuracy. Future research focuses on enhancing training datasets, refining model architectures, and integrating methods to support clinical decision-making for COVID-19 management.