ASSOCIATION OF CHEST RADIOGRAPHIC GEOMETRIC CHANGES IN MEDIASTINUM WITH COVID-19 CONDITIONS

Abstract

In this work, an attempt is made to investigate the association of geometric changes in mediastinum and lungs with Coronavirus Disease-2019 (COVID-19) using chest radiographic images. For this, the normal and COVID-19 images are considered from a public database. Reaction-diffusion level set is employed to segment the lung fields. Further, Chan Vese level set mechanism is used to delineate the mediastinum. Features, such as area, convex area, and bounding box area, are extracted from the mediastinum and lung masks. Then, mediastinum to lungs ratiometric features are derived, and statistical analysis is performed. The results demonstrate that the proposed methods are able to segment both regions by capturing significant anatomical landmarks. The ratiometric indices, along with mediastinum measures, are observed to be statistically significant for normal and COVID-19 conditions. Mediastinum convex area for COVID-19 conditions is found to be two times greater than normal subjects indicating the maximum difference in values between the classes. An AUC of 94% is obtained using SVM classifier for differentiating normal and COVID-19 conditions. Thus, the investigation of the mechanics of structural alterations of lungs and mediastinum is significant in COVID-19 diagnosis. As the proposed approach is able to detect COVID-19 conditions, it could act as a decision support system to assist clinicians in early detection.