Detection of brief atrial fibrillation in continuous 12-lead ECG recordings using an algorithm for computerized resting ECG interpretation combined with convolutional neural network for reducing false detections

Abstract

Existing criteria recommended by ACC/ESC for identifying patients with ST elevation myocardial infarction (STEMI) from the 12-lead ECG perform with high specificity (SP), but low sensitivity (SE). In our previous studies, we found that the SE of ischemia detection can be markedly improved without any loss of SP by calculating, from the 12-lead ECG, ST deviation in 3 “optimal” vessel-specific leads (VSLs). Our original VSLs, based on ΔST body-surface potential maps (BSPMs), have been modified by using the more appropriate J-point BSPMs at peak ischemia (without subtraction of pre-occlusion distributions). The aim of the present study was to compare the performance of these new VSLs with that achieved by the STEMI criteria used in current practice.Two independent datasets of 12-lead ECGs were used: the STAFF III dataset acquired during ischemic episodes caused by balloon inflation in LAD (n = 35), RCA (n = 47), and LCx (n = 17) coronary arteries, and the Glasgow dataset comprising admission 12-lead ECGs of 116 patients who were hospitalized for chest pain and underwent contrast-enhanced cardiac MRI that confirmed AMI in 58 patients (50%).We found that, in the STAFF III dataset, the detection of ischemic state by the STEMI criteria attained SE/SP of 60/97%, whereas SE/SP values of VSLs were 72/98%. In the Glasgow dataset, STEMI criteria yielded SE/SP of 43/98%, whereas the VSLs improved SE/SP to 60/98%. The most significant increase in diagnostic performance appeared in patients with LCx coronary artery occlusion: in STAFF III data (n = 17) SE achieved by STEMI criteria was improved by the VSLs from 35% to 71%; in Glasgow data (n = 12) SE of 31% achieved by STEMI criteria was improved by the VSLs to 69%.In our study population, existing ACC/ESC STEMI criteria complemented by the new VSLs yielded much improved sensitivity of ischemia detection without any detrimental effect on specificity. This finding needs to be corroborated on a larger chest-pain patient population with typical prevalence of acute ischemia presented to the emergency rooms.