Lightweight Deep Learning Model for Automated COVID-19 Diagnosis from CXR Images

The novel coronavirus has caused a huge number of infections and continues to be a concern. Rapid detection of infection is paramount in order to (i) ensure timely quarantine and segregation, and (ii) to plan suitable treatment trajectories. The conventional RT-PCR test for COVID-19 is both expensive and slow. Therefore, researchers looked towards leveraging deep learning (DL) for diagnosis of the disease from a variety of images such as CT Scans, Chest X-Ray images, etc. Such deep models have shown promise, and are increasingly been improved upon. This paper presents one such bespoke deep learning model for COVID-19 detection from CXR scans. It is shown that the proposed model matches the performance of most other available DL models while requiring only a fraction of the model size and number of parameters, thereby making it the most lightweight high-performance model available for automated COVID-19 detection from CXR images.