Resumen

AJITKUMAR SINGH, Sinam; ASHINIKUMAR SINGH, Sinam; DINITA DEVI, Ningthoujam  y  MAJUMDER, Swanirbhar. Heart Abnormality Classification Using PCG and ECG Recordings. Comp. y Sist. [online]. 2021, vol.25, n.2, pp.381-391.  Epub 11-Oct-2021. ISSN 2007-9737.  https://doi.org/10.13053/cys-25-2-3447.

Both PCG (Phonocardiogram) and Electrocardiogram (ECG) carry helpful features that aid in the fundamental analysis of heart-related disorders. Although they contain varying physical characteristics, some characteristics may predict some parameters better than the other. The research of such electrical and mechanical signals reveals a beneficial topic for any researcher. Hence, the study for automated detection and prediction of an anomaly of the heart using PCG and ECG signal is essential. The proposed method introduced modified preprocessing techniques along with features extraction techniques using both ECG and PCG datasets in tandem based on a different classification approach. The preprocessing of ECG signals comprises of the delineation and elimination of noise and artifacts whereas, the preprocessing of PCG signals includes the removal of unwanted noise and murmurs by applying a band-pass filter. The time-frequency features using PCG signals were extracted based on wavelet decomposition, Homomorphic filtering, Hilbert transforms, and Power spectral density. Using the ECG signals, the QRS based feature extraction method based on the Pan-Tompkins algorithm was performed. The extracted features from PCG and ECG signals were independently trained and tested using different classifiers (SVM, KNN, and Ensemble). Finally, the merged features of both the PCG and ECG signals were again trained and tested. The proposed model was validated using publicly available data-sets 'A' of PhysioNet 2016/ CinC challenges that comprise of both ECG and PCG data-sets. The results show that ECG and PCG signals can efficiently be employed for predicting cardiovascular disorders.

Palabras llave : Phonocardiogram; electrocardiogram; QRS complex; wavelet decomposition; Hilbert transform; homomorphic envelope; K nearest neighbors; power spectral density; support vector machine; ensemble of classifiers.