Assessment of Cardiomegaly from Chest X-Ray Using Deep Learning Artificial Intelligence

Abstract

Chest radiography is essential in assessing heart size and shape, contributing to the screening of lung and heart diseases. The cardiothoracic ratio (CTR), defined as the ratio of the heart's largest transverse dimension to the chest's largest transverse dimension, serves as a crucial indicator of cardiac abnormalities. Traditional CTR measurement relies on segmentation-based approaches, which are prone to inaccuracies due to anatomical variations and inconsistencies in image quality. This study introduces a novel regression-based deep learning approach for automated CTR prediction, leveraging ProGAN-generated synthetic chest X-ray data. The proposed method eliminates the dependency on segmentation while enhancing accuracy and efficiency. A CNN model is developed and trained on ProGAN-generated images. The effectiveness of the regression approach is compared with conventional segmentationbased techniques, with accuracy assessments and Grad-CAM visualizations. The findings demonstrate improved performance and robustness in CTR prediction, offering a potential screening tool to assist medical professionals in assessing cardiomegaly.