Heartbeat audio signal analysis for cardiovascular abnormality diagnosis using neural-networks

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Cardiovascular diseases (CVDs) are the leading global cause of death, necessitating early detection and management.
• Accurate analysis of heartbeat signals is vital for identifying CVDs and improving patient outcomes.
• Prompt cardiovascular disease forecasting is critical for timely intervention and better patient results.

Purpose of the Study:

• To develop a resilient neural network architecture for automated heartbeat signal classification.
• To integrate Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) networks for enhanced diagnostic capabilities.
• To assess the model's performance on publicly available datasets for cardiovascular disease detection.

Main Methods:

• Implemented a hybrid CNN-LSTM neural network architecture.
• Applied comprehensive data preprocessing techniques including balancing, normalization, and augmentation.
• Utilized feature extraction to enhance heartbeat signal quality for analysis.

Main Results:

• Achieved up to 96.04% accuracy in binary classification of heartbeat signals.
• Reached 97.83% accuracy in three-class classification tasks.
• Demonstrated competitive performance compared to recent baseline models on similar datasets.

Conclusions:

• The proposed audio-based diagnostic framework shows significant potential for automated heart sound classification.
• Incorporating auscultation signals into decision-support systems is feasible and beneficial.
• Future work will focus on model enhancement, dataset augmentation, and exploring additional predictive variables for increased clinical relevance.