Evaluating the risks of arrhythmia through big data: Automatic processing and neural networks to classify epicardial electrograms

Arrhythmic behaviors are a major risk to the population. These are diverse and can have their origin in cellular dynamics that affect the functioning of the heart. When trying to understand the mechanisms behind arrhythmogenesis the epicardial electrograms present themselves as a useful measurement because they reflect the electrical behavior of the cells surrounding the electrodes. Nevertheless, there is a lack of methods in the literature to automatically process and analyze these signals. In this paper, an algorithm to automatically detect the R, S and T wave peaks in epicardial electrogram signals is presented. This algorithm uses the derivative of the signal to find the activation and recovery times, and uses these as fiducial points to find the desired features. These features are then used as inputs to an artificial neural network, trained to classify individual beats into 'healthy' and 'pathological'. After optimization, both the detector and the neural network showed good performance in their tasks; furthermore, the robustness and amenability to real-time implementation of the methods here presented make them ideal for monitoring patients or experimental platforms when epicardial electrograms can be measured.