Enhanced Heart Sound Classification Using Mel Frequency Cepstral Coefficients
Author Information
Author(s): Hosseinzadeh Mehdi, Haider Amir, Malik Mazhar Hussain, Adeli Mohammad, Mzoughi Olfa, Gemeay Entesar, Mohammadi Mokhtar, Alinejad-Rokny Hamid, Khoshvaght Parisa, Porntaveetus Thantrira, Rahmani Amir Masoud
Primary Institution: School of Computer Science, Duy Tan University, Da Nang, Vietnam
Hypothesis
This study aims to enhance the performance of Mel Frequency Cepstral Coefficients (MFCCs) for detecting abnormal heart sounds.
Conclusion
The ensemble-classifier strategy improved the classification accuracy of heart sounds compared to the single-classifier strategy.
Supporting Evidence
- MFCCs were more effective than other features for heart sound classification.
- The ensemble classifier improved the accuracy of the SVM and DT by 1.64% and 4.89%, respectively.
- Classification accuracy was 93.59% for the SVM in the ensemble strategy.
- Both single and ensemble classifiers were tested on a large dataset of 2137 PCGs.
Takeaway
The study found a better way to tell if heart sounds are normal or not by using special sound features and comparing different methods.
Methodology
Heart sounds were pre-processed, segmented, and classified using single and ensemble classifier strategies based on MFCCs.
Limitations
The study did not evaluate the segmentation step and only focused on spectral features, neglecting temporal features.
Participant Demographics
The study used a publicly available database containing signals from healthy subjects and patients with heart diseases.
Statistical Information
Confidence Interval
[92.31, 93.25] for SVM sensitivity; [91.11, 91.71] for kNN sensitivity
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website