CardioGraphFormer based automated cardiomegaly detection from chest radiographs using graph-enhanced deep learning

Abstract

Cardiomegaly, or enlargement of the heart, is a serious condition that often indicates underlying cardiovascular disease. Early detection is important because delayed diagnosis can lead to severe complications, including heart failure and increased mortality. Chest X-ray imaging is widely used as an initial screening tool; however, manual interpretation is time-consuming and depends heavily on clinical expertise. To address this challenge, this study presents an automated deep learning–based framework for detecting cardiomegaly from chest X-ray images. The dataset used in this research was collected from a publicly available Kaggle repository and includes both cardiomegaly and normal cases. All images were preprocessed using contrast enhancement and resizing to improve visual quality and model performance. The proposed model integrates convolutional neural networks with transformer and graph-based reasoning to capture both local anatomical features and global structural relationships of the cardiac region. Experimental results show that the proposed approach achieves superior performance compared with several baseline models, reaching an accuracy of 95.28% along with strong precision, recall, and AUC-ROC values. Additionally eXplainable AI analyses, including attention maps, Grad-CAM heatmaps, morphological assessment, and feature embedding, demonstrate that the model provides reliable and predictions.