A New Ensemble Approach for Congestive Heart Failure and Arrhythmia Classification Using Shifted One-Dimensional Local Binary Patterns with Long Short-Term Memory

Abstract The electrocardiogram (ECG) is a vital diagnostic tool for identifying a variety of cardiac disorders, including cardiac arrhythmia (ARR), sinus rhythms and heart failure. However, rapid interpretation of ECG recordings is quite important in the diagnosis of heart-related diseases. Many patients can be saved using the systems developed for the rapid and accurate analysis of ECG signals. A novel ensemble method based on shifted one-dimensional local binary patterns (S-1D-LBP) and long short-term memory (LSTM) is presented for the prognosis of ARR, normal sinus rhythm (NSR) and congestive heart failure (CHF) in this study. The ECG signals were first subjected to the S-1D-LBP method. Depending on the R and L parameters of this method, nine different signals are generated. Each of the histograms of these signals is given to LSTM models with the same hyperparameters. ECG signals are classified according to the common decisions of LSTM models with nine different input signals. The suggested method was tested using ECG signals (ARR, NSR and CHF) from the MIT-BIH and BIDMC datasets. Considering the results obtained in the applications carried out with various scenarios, it was observed that a high (99.6%) success rate was attained by the proposed approach. The suggested S-1D-LBP + ELSTM (Ensemble LSTM) model is expected to be safe to employ in the classification of various signals.