Two-Stage Deep Learning Architecture for Chest X-Ray-Based COVID-19 Prediction

Abstract

The coronavirus pandemic has led to an immense loss of human life worldwide. It poses an unprecedented challenge to the education system, economic system, food systems, public health, and the job market. Approximately 220 nations, regions, or territories are now affected by coronavirus disease, and the World Health Organization has recorded about 1,535,982 deaths. The reverse transcription polymerase chain reaction (RT-PCR) or rapid antigen test or computed tomography (CT)-based detection techniques lack rapid testing property due to their low sensitivity and availability issues. Therefore, the use of deep convolutional neural networks (DCNN) to learn and analyze the chest X-ray signatures of COVID-19 patients is becoming crucial. This chapter gives a brief introduction two various deep learning architectures used for chest X-ray analysis for COVID-19 detection, along with their pros and cons. Further in this chapter, it is proposed that there should be a two-stage DCNN architecture to learn the imaging hallmarks of normal, pneumonia, and COVID-19 diseases from chest X-ray images and predict those lung diseases without human intervention. The main objective of this study is to precisely extract the region of interest, i.e., the lung contours, from the chest X-ray images and feed them to the classification network. It directly helps to increase the diagnosis sensitivity, reduce diagnosis time, and enable easier analysis and classification of chest-related diseases. It also helps to efficiently train the deep learning network, even with a minimal number of training samples. The experimental results demonstrate that the proposed model’s misclassification rate is low. The proposed model outperforms other state-of-the-art studies and achieves a COVID-19 diagnosis sensitivity of 99% and an overall accuracy of 98%. The proposed model can be used as a reliable rapid testing tool for COVID-19 disease but not as an alternative to RT-PCR. The performance of the model was also tested in Indian COVID-19 patients. Ultimately, this low-cost, intelligent two-stage DCNN provides a detailed understanding that can be useful to improve patient diagnostic decisions made by medical professionals.