A robust dataset-agnostic heart disease classifier from Phonocardiogram.

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Automatic classification of heart sounds is crucial but challenging due to signal quality and demographic variations.
• Developing robust algorithms for Phonocardiogram (PCG) analysis remains an active research area.

Purpose of the Study:

• To create a robust and discriminative feature set for dataset-agnostic classification of normal and abnormal heart sounds.
• To develop a classification methodology for cardiovascular diseases using diverse PCG features.

Main Methods:

• Analysis of a wide range of PCG features in time, frequency, morphological, and statistical domains.
• Utilized the Physionet 2016 challenge dataset for feature selection, validation, and model training.
• Evaluated the methodology on an in-house dataset collected via a smartphone-based digital stethoscope.

Main Results:

• Achieved sensitivity of 0.76 and specificity of 0.75 on the independent test dataset for cardiovascular disease classification.
• The proposed feature set and classification methodology demonstrated robustness and discriminative power.
• Outperformed three popular prior art approaches on the same evaluation dataset.

Conclusions:

• The developed methodology offers a robust approach for classifying cardiovascular diseases from heart sounds.
• The proposed feature set enhances the generalizability of classification models across different datasets.
• This work contributes to advancing automated cardiac diagnostics using accessible technology.