CardioXNet: Representation Learning for Accurate Cardiovascular Disease Diagnosis from X-ray Images

Cardiovascular Disease (CVD) is a prevalent, incurable condition affecting the heart and blood vessels. Due to its significant impact on mortality in the United States, there is a pressing need for enhanced risk stratification methods. Coronary Artery Calcium Score (CACS), reliant on CT scans, is most commonly employed as a risk stratification method but suffers from limitations, including accessibility and early detection challenges. To address the problem, this research study proposes a representation learning-based CVD diagnosis framework utilizing X-ray images, offering expedited and earlier detection, as well as improved accessibility. This system comprises two distinct stages: representation learning to extract CVD-related features and transfer learning to train a CVD classifier. Representation learning enhances the quality of extracted features, thereby leading to more precise results in the subsequent CVD diagnosis network. Comprehensive experiments and training validate the efficacy of the proposed method, demonstrating its superiority over the existing methods. These promising results suggest the potential utility of X-rays as a valuable biomarker for diagnosing CVD disease.