Deep Neural Network Trained on Surface ECG Improves Diagnostic Accuracy of Prior Myocardial Infarction Over Q Wave Analysis

Ozal Yildirim1, Ulas Baloglu2, Muhammed Talo3, Prasanth Ganesan4, Jagteshwar Tung4, Guson Kang4, Mahmood Alhusseini4, Tina Baykaner4, Paul Wang4, Marco Perez4, Larissa Tereshchencko5, Albert Rogers4
1Munzur University, 2University of Bristol, 3Firat University, 4Stanford University, 5OHSU


Background: Clinical screening of myocardial infarction (MI) is important for preventative treatment and risk stratification for cardiology practice. Many of these events may be silent (without clinical symptoms) but have the same impact on health. Current detection by ECG Q-wave analysis is quick and inexpensive but has poor accuracy for assessing prior MI.

Objective: To evaluate the ability of a deep neural network (DNN) trained on the surface ECG to identify patients with clinical history of myocardial infarction.

Methods: We assessed 608 well-characterized patients (61.4 ± 14.5 years, 31.2% female) at 2 academic centers. From one 12-lead ECG, median beats were calculated in 3 orthogonal planes (X, Y, Z; [A]) and used to train a DNN to identify a history of prior myocardial infarction. Accuracy was compared to manual assessment of pathologic Q waves, defined as a deflection > 25% of the subsequent R wave, >40ms in width, and > 0.2mV amplitude in 1 of 3 ECG planes.

Results: Of 608 patients, 175 had history of MI (28.7%). The DNN outperformed the accuracy of pathologic Q waves. In training, DNN converged to >98% accuracy and in testing, its accuracy was 71 ± 5% [B] (k=5-fold cross validation). This outperformed the 62% accuracy of pathologic Q waves in this study (red dotted line, Fig. B). In the validation cohort, DNN provided an area under the receiver operating characteristics curve of 0.730 [C].

Conclusion: Deep learning of a 12-lead ECG can identify features of prior myocardial injury more accurately than clinical Q-wave analysis. Such a platform could be used for frequent screening of interval MI events. In attempting to improve these results further, studies should explain what inputs weighted DNN decisions, and identify those that reflect abnormalities detectable clinically or on imaging.