Abstract 13223: Assessing Cardiomediastinal Abnormalities Through Quantitative Evaluation of Chest X-rays: A Multicenter Validation in Various Clinical Cohorts

Abstract

Objective: The ADC study, entitled 'Quantitative A nalysis of Chest X-ray for D iagnosing C ardiomediastinal Abnormality,' is aiming to establish the normal reference values for cardiovascular borders (CB) using Chest X-ray (CXR) and to investigate its clinical application. Methods: The normal reference values of CBs, including right upper/lower, aortic knob, pulmonary conus, left atrial appendage [LAA], and, left lower were determined by pre-validated deep learning from 71,494 'normal' CXRs, as defined by echocardiography. All CBs were standardized into age-and gender specific z-score and underwent validation using three distinct normal external datasets (n=25,429). Subsequently, the clinical relevance was investigated across different disease cohorts - coronary artery disease (CAD, n=34,251), valvular heart diseases (VHD, n=9,893), and mediastinal mass (Mass, n=109). Results: External normal cohorts aligned with the normal CBs established by this study, with a low range of z-scores (-0.49 - 0.40). After adjusting for cardiovascular risks in CAD cohorts, an increased pulmonary conus (z-score > 2.5) on CXR was found to be independently linked to adverse outcomes (HR 1.56 - 3.76, p < 0.001). The pattern of cardiomegaly varies depending on the type of VHDs. In mitral stenosis (MS), the LAA are broader than in aortic stenosis (AS) (1.73 vs. 0.55), while the ascending aorta is more expansive in AS (1.20 vs. 0.47). The subgroup of MS with significant TR demonstrated worse CXR measurements as opposed to MS in isolation (LAA diameter 62.2mm vs. 57.4mm). Among 109 mediastinal masses, the borders of 66 tumors (60.6%) were merged with the CBs and consequently detected via CB analysis. Conclusion: This fully automated, deep learning-guided chest X-ray analysis may contribute to various clinical tasks such as detection, classification, and risk stratification of cardiomediastinal abnormalities.