Related Experiment Video
Updated: May 10, 2025
06:57
Author Spotlight: Advancing Cardiovascular Imaging - Introducing the Spatially Weighted Calcium Score for Early Disease Detection
Published on: September 22, 2023
891
Automated, Standardized, Quantitative Analysis of Cardiovascular Borders on Chest X-Rays Using Deep Learning.
June-Goo Lee1, Tae Joon Jun2, Gyujun Jeong1
1Biomedical Engineering Research Center, Asan Institute for Life Sciences, Asan Medical Center, Seoul, South Korea.
JACC. Advances
|April 26, 2025
Summary
Deep learning quantifies cardiovascular borders on chest X-rays, establishing normal ranges and enabling risk stratification for heart conditions. This AI approach offers improved accuracy over subjective assessments.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Cardiovascular Disease Diagnostics
- Radiology and Imaging Analysis
Background:
- Traditional analysis of cardiovascular borders (CVBs) in chest X-rays (CXRs) lacks objective measures and established normal ranges.
- Subjective interpretation of CXRs can lead to variability in assessing cardiac size and shape.
Purpose of the Study:
- To develop and validate a deep learning (DL) model for quantitative CVB analysis on CXRs.
- To establish age- and sex-specific normal ranges for CVBs using a large dataset of normal CXRs.
- To explore the clinical utility of DL-derived CVB z-scores in classifying and stratifying cardiovascular diseases.
Main Methods:
- A prevalidated deep learning model was employed to analyze CVBs in a large cohort of 96,129 normal CXRs from four institutions.
- Normal ranges for CVBs were established and standardized into z-scores for newly inputted CXRs.
- The clinical utility of z-score analysis was assessed using 44,567 diseased CXRs across various cardiovascular conditions, including valve disease, coronary artery disease, and congenital heart disease.
Main Results:
- The DL model demonstrated strong performance in distinguishing valve disease from normal controls, with AUCs of 0.80 for cardiothoracic ratio and 0.83 for combined right atrium and left ventricle borders.
- Significant differences in CVB z-scores were observed between mitral and aortic stenosis, correlating with disease pathophysiology.
- Cardiothoracic ratio z-scores were independently associated with a 5-year risk of death or myocardial infarction in patients with coronary artery disease.
Conclusions:
- Deep learning-based z-score analysis of CXRs shows significant potential for objective quantification of CVBs.
- This AI-driven approach can aid in the classification and risk stratification of diverse cardiovascular abnormalities.
- The established normal ranges and z-scores offer a standardized method for interpreting CVBs in clinical practice.
