Detection of bradycardia from electrocardiogram signals using feature extraction and snapshot ensembling

One of the most common diagnostic techniques for detecting certain cardiovascular diseases is using electrocardiogram (ECG) readings. Doctors around the world mostly rely on human insight and processing to determine and interpret these ECG graphs. This process is thus often prone to human error introduced to the increasing cognitive burden of doctors and might introduce delays in diagnosis, which could be fatal. Ongoing research has focused on the design of automated algorithms to accurately diagnose and speed up the process of analyzing and interpreting an ECG signal. In this paper, we present a novel approach that utilizes a neural network pipeline with Snapshot ensembling to enable automated Bradycardia detection from ECG signals. Before the modeling phase, a cross-correlation and segmentation method is used for detecting relevant features in the ECG signals, using which the detection performance is improved. The proposed approach gave good results, with around 95% accuracy and an AUC score of about 0.96, implying an efficient and accurate classification.