Temporal HRV Feature Extraction Using QRS Peaks For Arrhythmia ECG Heart Disease Patterns

Abstract

Electrocardiogram (ECG) signals are commonly used to diagnose heart electrical activity and disorders. Many different heart ECG patterns are detected and used to make diagnoses. This makes it critical to classify the appropriate ECG pattern prior to diagnosis and treatment. This research sought to create an effective feature extraction strategy for ECG classification problem. Paper initially processes the raw ECG data employing the modified Pan Tompkins basic filtering technique to remove noise and artifacts. Then average filter is used to smooth the difference of notch filtered data and its 10^th level wavelet approximation. The filtered data is used to detect the QRS wave peaks to be utilised as a criterion for further classification. Paper proposed an adaptive threshold selection based ECG peak detection method. The temporal ECG features are extracted based on the detected peaks. Rich set of 8 time domain heart rate variability (HRV) features are considered and extracted for each ECG pattern. The algorithm is tested over the MIT-BIH ECG database signal with 10 ECG data for each of four patterns. Machine learning (ML) classification algorithms can be employed to recognize distinct ECG disease patterns based on the QRS set of features detection. The basic proposed method correctly recognizes the ECG signal peaks including the Q, R, and S peaks. The visual results indicate the ECG's QRS detection efficiency in the case of artifacts.