Development of Machine Learning Algorithm for Diagnosis of Arrhythmia from ECG Signals

Abstract

ARRHYTHMIA is a kind of disorder which occurs due to newlinevariations in the heart beat rate than the normal. There are several kinds newlinein this heart beat rate. Some beats are too fast which is called as newlinetachycardia and some beats are too slow and those are named as newlinebradycardia. There are many reasons which can be listed out like heart newlinedisease, abnormalities in the structure of the heart, life style, imbalances newlinein electrolytes of the body medications etc. Some of the kinds of newlineARRHYTHMIA are harmless and some were threatened to life. So, in newlinesuch scenario a diagnostic tool is necessary to identify the early symptoms newlineof ARRHYTHMIA. The conventional method followed for the diagnosis newlineof ARRHYTHMIA is by analysing the Electrocardiograms (ECG) newlinesignals. These analyses are done manually. So, the objective of this newlineresearch is to develop some computed aided diagnostic tool that aids the newlineclinicians in diagnosis of ARRHYTHMIA. newlineIn the first phase of the study, feed forward networks were used newlineto analyse the performance of classification algorithms between the newlinenormal and abnormal features of ARRHYTHMIA. Feedforward neural newlinenetworks are the one in which the connection between the nodes do not newlineform cycles. So, the information from input node to output node flows newlineonly in one direction. This architecture is well suited for concluding the newlineefficiency of conventional classifiers between the normal and abnormal newlinefeatures of ARRHYTHMIA. In this phase of works, the wavelet features newlinevi newlineextracted from the ECG signals were fed to feed forward neural networks. newlineThe classification accuracy achieved in this phase of study is 88 percent. newlineIn the next phase of the study, the work is narrowed to diagnose newlinea specific kind of ARRHYTHMIA. So the features of ECG signals which newlinebelongs to the category of Atrial Fibrillation type ARRHYTHMIA was newlinechosen for the study. The features belong to the normal class and the class newlineof Atrial Fibrillation were fed to Long Short-Term Memory (LSTM) newlinenetwork. The classification accuracy achieved with this method is 68 newlinepercent.