1D neural network design to detect cardiac arrhythmias

Abstract

This article shows a neuronal network for deep learning focused on recognizing and classification five types of cardiac signals (Sinus, Ventricular Tachycardia, Ventricular Fibrillation, Atrial Flutter, and Atrial Fibrillation). The final objective is to obtain an architecture that can be implemented in an embedded system as a pre-diagnostic device linked to a Holter monitoring system. The network was designed using the Keras API programmed in Python, where it is possible to obtain a comparison of different types of networks that vary the presence of a residual block, with the result that the network with said block obtains the best response (100% success rate) and a model loss of approximately 0.15%. On the other hand, a validation by means of confusion matrices was carried out to verify the existence of false positives in the network results and evidence what type of arrhythmia can be presented according to the network output against an input signal through the console.