Artificial intelligence-enabled electrocardiographic screening for left ventricular systolic dysfunction and mortality risk prediction

• DOI: 10.3389/fcvm.2023.1070641
• Publication Date: 2023-03-03T06:09:21Z
• Authors: Yu-Chang Huang; Yu-Chun Hsu; Zhi-Yong Liu; Ching-Heng Lin; Richard Tsai; Jung-Sheng Chen; Po-Cheng Chang; Hao-Tien Liu; Wen-Chen Lee; Hung-Ta Wo; Chung-Chuan Chou; Chun-Chieh Wang; Ming-Shien Wen; Chang-Fu Kuo
• Keywords: RC666-701; deep neural network; all-cause mortality; Diseases of the circulatory (Cardiovascular) system; left ventricular ejection fraction; electrocardiogram; Cardiovascular Medicine; left ventricular systolic dysfunction; 3. Good health

Left ventricular systolic dysfunction (LVSD) characterized by a reduced left ejection fraction (LVEF) is associated with adverse patient outcomes. We aimed to build deep neural network (DNN)-based model using standard 12-lead electrocardiogram (ECG) screen for LVSD and stratify prognosis.This retrospective chart review study was conducted data from consecutive adults who underwent ECG examinations at Chang Gung Memorial Hospital in Taiwan between October 2007 December 2019. DNN models were developed recognize LVSD, defined as LVEF <40%, original signals or transformed images 190,359 patients paired echocardiogram within 14 days. The divided into training set of 133,225 validation 57,134. accuracy recognizing subsequent mortality predictions tested ECGs 190,316 data. Of these patients, we further selected 49,564 multiple echocardiographic predict incidence. additionally used 1,194,982 only assess prognostication. External performed 91,425 Tri-Service General Hospital, Taiwan.The mean age the testing dataset 63.7 ± 16.3 years (46.3% women), 8,216 (4.3%) had LVSD. median follow-up period 3.9 (interquartile range 1.5-7.9 years). area under receiver-operating characteristic curve (AUROC), sensitivity, specificity signal-based (DNN-signal) identify 0.95, 0.91, 0.86, respectively. signal-predicted age- sex-adjusted hazard ratios (HRs) 2.57 (95% confidence interval [CI], 2.53-2.62) all-cause 6.09 (5.83-6.37) cardiovascular mortality. In echocardiograms, positive prediction preserved an adjusted HR CI) 8.33 (7.71 9.00) incident Signal- image-based DNNs equally well primary additional datasets.Using DNNs, becomes low-cost, clinically feasible tool facilitate accurate