Artificial intelligence framework for heart disease classification from audio signals.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Cardiovascular disorders are a leading cause of mortality globally.
• Accurate and early diagnosis of heart disease is crucial for effective treatment and patient outcomes.
• Current diagnostic methods can be invasive or require specialized equipment, creating a need for accessible alternatives.

Purpose of the Study:

• To investigate the efficacy of machine learning (ML) and deep learning (DL) techniques for detecting heart disease using audio sound signals.
• To analyze the performance of various ML and DL models in classifying cardiovascular disorders from noisy heart sound recordings.
• To explore data augmentation and feature ensembling strategies to enhance diagnostic accuracy.

Main Methods:

• Utilized two subsets of real heart audio datasets from the PASCAL CHALLENGE.
• Employed signal visualization techniques including spectrograms and Mel-Frequency Cepstral Coefficients (MFCCs).
• Applied data augmentation to introduce synthetic noise and developed a feature ensembler for integrated audio feature extraction. Several ML and DL classifiers were evaluated.

Main Results:

• The multilayer perceptron model demonstrated superior performance among the evaluated classifiers.
• The best-performing model achieved an accuracy rate of 95.65% in detecting heart disease from audio signals.
• Signal processing techniques and data augmentation significantly contributed to model robustness and accuracy.

Conclusions:

• Audio signal-based analysis using ML and DL presents a promising, non-invasive approach for heart disease detection.
• The high accuracy achieved highlights the potential for integrating this methodology into clinical practice.
• This research offers opportunities for improved medical diagnosis, enhanced patient care, and more accessible cardiovascular screening.