High-Accuracy ECG Arrhythmia Classification Using a Hybrid Deep Learning Model BiLSTM-GRU With Enhanced Features Extraction Algorithms

An electrocardiogram (ECG) analyzes the electrical activity of the heart, which diagnoses cardiac arrhythmias in the biomedical field. The ECG signals are highly susceptible to noise, and the signal waveform varies from person to another based on the age of the person and the heart disease, so accurate ways were needed for feature extraction that led to the best ECG arrhythmias identification. The global advantage is making heart disease diagnosis easier and more accurate. This study analyzes 7 ECG classes, which are normal class and arrhythmia classes. In the preprocessing step, a second-order IIR filter and a fourth-order high-pass Butterworth filter were used. Then, enhanced QRS detection algorithms, modified ST analyzer, and wave delineation analysis were applied to extract ECG morphologic features. And also, statistical and frequency domain features were calculated. The ECG features were fed into a novel model (BiLSTM-GRU) that combined and modified 2 deep learning techniques. Our methods showed an accurate 18 features were extracted correctly for 1800 ECG signals. The features were fed to the BiLSTM-GRU deep learning model, which classified the data into 7 classes using a softmax regressor at the end of the new model. Obtained an accuracy 99.9204, precision 99.3786, recall 100, specificity 99.909, F-Score 99.6859, and error 0.0796.