An Approach to Identify Regions of Interest in Chest X-Ray Images of COVID-19 Patients and Its Clinical Validation: An Indian Study

Abstract

Background & Objective: This study aims to develop a simple and low-cost approach in identifying the Regions of Interest (ROI) inside the lung fields and accessory structures in the respiratory system. Methods: To achieve this goal, Marker Based Watershedding (MBW) segmentation operation has been applied on ten Chest X-Ray (CXR) images of COVID-19 patients. CXR is economic compared to CT images and MBW is computationally simple compared to the Deep Learning (DL) techniques. Raw images have been tested for inherent noise by computing Noise Variance (NV) and Signal-to-Noise ratio (SNR). Then, 'Simple Median Filter (SMF)' was used to denoise raw images before MBW operation and the denoising performance was checked by estimating the Mean Squared Errors (MSE), and Peak Signal-to-Noise Ratio (PSNR). Regions of Interest (ROI), thus obtained, are then validated by radiologists and then clinically correlated with the symptoms. Results: The study shows that the SMF is an efficient filter for CXR images and MBW can identify the ROIs, which are supported by the symptoms. Conclusion: The paper presents a simple, low-cost (both financially and computationally), and reliable method to get a complete clinical picture of COVID-19 cases by applying the MBW segmentation technique on CXRs that is further validated with radiologists. The social implication of this work is that it can be used by general physicians and nurses in remote areas as a ready reference to important regions of the lungs, where, radiologists are unavailable. The approach can be incorporated in telemedicine during a pandemic period.