Figure 1.

(a) The multi-stage DL-based model for ECG abnormality classification consists of a feature extraction module, an optimal ECG-lead subset selection module and a decision-making module. The blocks with max pooling shortcuts are the residual blocks. BN, batch normalization layer; Conv, convolutional layer; ReLU, rectified linear unit activation; LSTM, long short-term memory layer; AF, atrial fibrillation; I-AVB, first-degree atrioventricular block; LBBB, left bundle branch block; RBBB, right bundle branch block; PAC, premature atrial contraction; PVC, premature ventricular contraction; STD, ST-segment depression; STE, ST-segment elevated. (b) An illustration of the forward stepwise subset selection process. In each step, the candidate single-lead ECG that introduces the greatest performance improvement, i.e. the one with the smallest p-value < 0.05, is selected to the target set. (Online version in colour.)