Heartbeat Sound Signal Classification Using Deep Learning.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Heart disease remains a leading cause of mortality worldwide.
• Accurate diagnosis of heart conditions through heartbeat sound analysis is crucial.
• Heart sound classification presents challenges in segmentation and feature extraction.

Purpose of the Study:

• To develop a robust framework for heartbeat sound classification.
• To improve the accuracy and efficiency of diagnosing heart conditions using audio signals.
• To address the limitations of existing methods in heart sound analysis.

Main Methods:

• Applied band-pass filtering to remove noise from heartbeat sound signals.
• Standardized sampling rates and employed down-sampling for feature discrimination and dimensionality reduction.
• Utilized a Recurrent Neural Network (RNN) model incorporating Long Short-Term Memory (LSTM), Dropout, Dense, and Softmax layers.

Main Results:

• The proposed framework achieved competitive performance in classifying heartbeat sounds.
• Down-sampling techniques reduced computational power and time without compromising results.
• The RNN-based model demonstrated effectiveness in distinguishing between Normal, Murmur, and Extrasystole heartbeats.

Conclusions:

• The developed RNN-based model offers a promising approach for automated heart sound classification.
• The methodology enhances diagnostic capabilities for various heart conditions.
• This framework provides a more efficient and accurate tool for cardiovascular health assessment.