Feifei Yang1, Xiaotian Chen2, Xixiang Lin3, Xu Chen3, Wenjun Wang1, Bohan Liu1, Yao Li3, Haitao Pu2, Liwei Zhang4, Dangsheng Huang4, Meiqing Zhang4, Xin Li5, Hui Wang6, Yueheng Wang7, Huayuan Guo1, Yujiao Deng8, Lu Zhang9, Qin Zhong3, Zongren Li1, Liheng Yu3, Yongjie Duan3, Peifang Zhang2, Zhenzhou Wu2, Daniel Burkhoff10, Qiushuang Wang4, Kunlun He11. 1. Medical Big Data Research Center, Beijing Key Laboratory for Precision Medicine of Chronic Heart Failure, Key Laboratory of Ministry of Industry and Information Technology of Biomedical Engineering and Translational Medicine, Chinese PLA General Hospital, Beijing, China. 2. BioMind Technology, Zhongguancun Medical Engineering Center, Beijing, China. 3. Medical School of Chinese PLA, Beijing, China; and Medical Big Data Research Center, Chinese PLA General Hospital, Beijing, China. 4. Department of Cardiology, The Fourth Medical Center of Chinese PLA General Hospital, Beijing, China. 5. Department of Ultrasound Diagnosis, The Sixth Medical Center of Chinese PLA General Hospital, Beijing, China. 6. Department of Special Examination, The Second Affiliated Hospital of Shandong University of Traditional Chinese Medicine, Jinan, China. 7. Department of Ultrasound Diagnosis, The Second Hospital of Hebei Medical University, Shijiazhuang, China. 8. Department of Ultrasound Diagnosis, The First Medical Center of Chinese PLA General Hospital, Beijing, China. 9. Department of Cardiology, The Second Medical Center of Chinese PLA General Hospital, Beijing, China. 10. Cardiovascular Research Foundation, New York, New York, USA. 11. Medical Big Data Research Center, Beijing Key Laboratory for Precision Medicine of Chronic Heart Failure, Key Laboratory of Ministry of Industry and Information Technology of Biomedical Engineering and Translational Medicine, Chinese PLA General Hospital, Beijing, China. Electronic address: kunlunhe@plagh.org.
Abstract
OBJECTIVES: This study sought to develop a deep learning (DL) framework to automatically analyze echocardiographic videos for the presence of valvular heart diseases (VHDs). BACKGROUND: Although advances in DL have been applied to the interpretation of echocardiograms, such techniques have not been reported for interpretation of color Doppler videos for diagnosing VHDs. METHODS: The authors developed a 3-stage DL framework for automatic screening of echocardiographic videos for mitral stenosis (MS), mitral regurgitation (MR), aortic stenosis (AS), and aortic regurgitation (AR) that classifies echocardiographic views, detects the presence of VHDs, and, when present, quantifies key metrics related to VHD severities. The algorithm was trained (n = 1,335), validated (n = 311), and tested (n = 434) using retrospectively selected studies from 5 hospitals. A prospectively collected set of 1,374 consecutive echocardiograms served as a real-world test data set. RESULTS: Disease classification accuracy was high, with areas under the curve of 0.99 (95% CI: 0.97-0.99) for MS; 0.88 (95% CI: 0.86-0.90) for MR; 0.97 (95% CI: 0.95-0.99) for AS; and 0.90 (95% CI: 0.88-0.92) for AR in the prospective test data set. The limits of agreement (LOA) between the DL algorithm and physician estimates of metrics of valve lesion severities compared to the LOAs between 2 experienced physicians spanned from -0.60 to 0.77 cm2 vs -0.48 to 0.44 cm2 for MV area; from -0.27 to 0.25 vs -0.23 to 0.08 for MR jet area/left atrial area; from -0.86 to 0.52 m/s vs -0.48 to 0.54 m/s for peak aortic valve blood flow velocity (Vmax); from -10.6 to 9.5 mm Hg vs -10.2 to 4.9 mm Hg for average peak aortic valve gradient; and from -0.39 to 0.32 vs -0.31 to 0.32 for AR jet width/left ventricular outflow tract diameter. CONCLUSIONS: The proposed deep learning algorithm has the potential to automate and increase efficiency of the clinical workflow for screening echocardiographic images for the presence of VHDs and for quantifying metrics of disease severity.
OBJECTIVES: This study sought to develop a deep learning (DL) framework to automatically analyze echocardiographic videos for the presence of valvular heart diseases (VHDs). BACKGROUND: Although advances in DL have been applied to the interpretation of echocardiograms, such techniques have not been reported for interpretation of color Doppler videos for diagnosing VHDs. METHODS: The authors developed a 3-stage DL framework for automatic screening of echocardiographic videos for mitral stenosis (MS), mitral regurgitation (MR), aortic stenosis (AS), and aortic regurgitation (AR) that classifies echocardiographic views, detects the presence of VHDs, and, when present, quantifies key metrics related to VHD severities. The algorithm was trained (n = 1,335), validated (n = 311), and tested (n = 434) using retrospectively selected studies from 5 hospitals. A prospectively collected set of 1,374 consecutive echocardiograms served as a real-world test data set. RESULTS: Disease classification accuracy was high, with areas under the curve of 0.99 (95% CI: 0.97-0.99) for MS; 0.88 (95% CI: 0.86-0.90) for MR; 0.97 (95% CI: 0.95-0.99) for AS; and 0.90 (95% CI: 0.88-0.92) for AR in the prospective test data set. The limits of agreement (LOA) between the DL algorithm and physician estimates of metrics of valve lesion severities compared to the LOAs between 2 experienced physicians spanned from -0.60 to 0.77 cm2 vs -0.48 to 0.44 cm2 for MV area; from -0.27 to 0.25 vs -0.23 to 0.08 for MR jet area/left atrial area; from -0.86 to 0.52 m/s vs -0.48 to 0.54 m/s for peak aortic valve blood flow velocity (Vmax); from -10.6 to 9.5 mm Hg vs -10.2 to 4.9 mm Hg for average peak aortic valve gradient; and from -0.39 to 0.32 vs -0.31 to 0.32 for AR jet width/left ventricular outflow tract diameter. CONCLUSIONS: The proposed deep learning algorithm has the potential to automate and increase efficiency of the clinical workflow for screening echocardiographic images for the presence of VHDs and for quantifying metrics of disease severity.