Abstract 311: Machine Learning of the Electrocardiogram to Detect Regional Structural Abnormalities of the Heart

Abstract

Introduction: Regional dysfunction of cardiac myocardium can result from diverse underlying conditions and cause adverse events even in patients without clinically apparent cardiovascular disease over ejection fraction. Assessment of regional wall motion abnormalities (WMA) currently requires sophisticated imaging that is not feasible for population screening or frequent disease monitoring. Hypothesis: We hypothesized that an AI-enabled ECG model trained on a dataset of qualitative labels could identify and localize wall motion abnormalities with higher accuracy than traditional ECG analysis Methods: In a large academic center, a deep convolutional network was developed with raw 12-lead ECGs and associated reports from 82,424 transthoracic echocardiograms (N = 50,960 patients, age 64 ± 17 years, 54.5% male). 80% of the data was used in model development and the remaining 20% was used for testing. ECGs with ventricular pacing and non-finalized echocardiography reports were excluded. Pre-processing included Kors transformation and beat extraction for input size of 3x300 samples (Figure 1A) and output labels were established by NLP of semi-structured echocardiography reports. Results: ML ECG based models predicted inferior/posterior regional wall motion abnormality with C-statistic of 0.772 (Figure 1B). Physician interpretation of Q-wave ECG pattern was inferior (C-statistic of 0.505). Other segments had similar AUC (apex 0.81, anterior 0.80, lateral 0.75, septal 0.75). Patients with positive ECG screen had +LR 2.70 of having a wall motion abnormality on echocardiography and patients with WMA by echocardiography had decreased survival (LR 0.41, CI: 0.34-0.47, p<0.005) mortality over follow-up of >1500 days (Figure 1C). Conclusion: This study is the first to evaluate the presence and location of regional wall motion abnormalities from inexpensive and noninvasive 12-lead ECG. This approach could be applied for widespread ambulatory screening.