Automated, Standardized, Quantitative Analysis of Cardiovascular Borders on Chest X-Rays Using Deep Learning
Authors
June-Goo Lee, PHD, Tae Joon Jun, PHD, Gyujun Jeong, MS, Hongmin Oh, MS, Sijoon Kim, MS, Heejun Kang, MS, Jung Bok Lee, PHD, Hyun Jung Koo, MD, Jong Eun Lee, MD, Joon-Won Kang, MD, Yura Ahn, MD, Sang Min Lee, MD, Joon Beom Seo, MD, Seong Ho Park, MD, Min Soo Cho, MD, Jung-Min Ahn, MD, Duk-Woo Park, MD, Joon Bum Kim, MD, Cherry Kim, MD, Young Joo Suh, MD, Iksung Cho, MD, Marly van Assen, MD, Carlo N. De Cecco, MD, Eun Ju Chun, MD, Young-Hak Kim, MD
Dong Hyun Yang, MD,d the ADC Investigators
This study, named the ADC (Automated Diagnosis of Cardiovascular abnormalities using chest X-ray), developed a deep learning-based software for automated, quantitative analysis of cardiovascular borders (CVBs) in chest X-rays (CXRs) and evaluated its clinical utility. Using 96,129 normal CXRs from Korea and the United States, the authors established age- and sex-specific reference ranges for CVBs. Z-score mapping was then applied to 44,567 CXRs from patients with cardiovascular diseases. Z-scores of CVBs demonstrated strong diagnostic and prognostic capabilities, especially in valvular heart disease, coronary artery disease, and congenital heart disease. Notably, the cardiothoracic (CT) ratio was independently associated with 5-year mortality or myocardial infarction in coronary artery disease patients.
목록으로
