Heart Sound Classification for Murmur Detection
A deep learning system for detecting heart murmurs from phonocardiogram (PCG) recordings, aiding in early cardiovascular disease diagnosis.
This project focuses on automatic murmur detection from heart sound recordings using deep learning.
By analyzing phonocardiogram (PCG) signals, the system classifies heart sounds as normal or abnormal (murmur) to assist in early cardiovascular disease (CVD) detection.
Features
- Signal Preprocessing & Feature Extraction: Utilizes MFCCs, Wavelet Transforms, and Spectrograms to extract meaningful audio features.
- Deep Learning-Based Classification: Employs CNNs and RNNs to classify heart sounds into normal and murmur categories.
- Real-Time Inference: Designed for integration into medical diagnostic systems for real-time analysis.
- Clinical Relevance: Provides an automated, non-invasive solution for early screening of heart conditions.
Model architecture.
Challenges & Solutions
- Noise in PCG Recordings: Applied denoising filters and spectral analysis to improve signal clarity.
- Imbalanced Data: Used data augmentation techniques to balance normal and murmur samples.
- Real-Time Performance: Optimized model inference to enable real-time classification for clinical applications.
Results
The system achieves high classification accuracy, offering a cost-effective and scalable solution for early heart disease screening.