zaheerjk/AI-Powered-Heart-Murmur-Detection-System

🫀 AI-Powered Heart Murmur Detection System

An end-to-end AI-powered heart sound classification system that uses audio signal processing, MFCC feature extraction, and an LSTM deep learning model to analyze heart sound recordings and classify them into target categories.

⚠️ Medical Disclaimer: This project is for educational and research purposes only. It is not a medical device and should not be used for clinical diagnosis or treatment decisions.

🚀 Key Features

• 🎧 Heart sound audio classification
• 📊 MFCC feature extraction
• 🧠 LSTM deep learning model
• 🔊 Audio preprocessing and normalization
• ⏱️ Up to 10-second audio clips
• 🌐 Streamlit deployment
• 🤗 Hugging Face model hosting

🖼️ Project Preview

🏗️ System Architecture

🧠 Model Architecture

Heart Sound Audio
       ↓
Audio Preprocessing
       ↓
Resampling & Normalization
       ↓
MFCC Feature Extraction
       ↓
LSTM Deep Learning Model
       ↓
Classification
       ↓
Prediction + Confidence

📋 Model Details

Parameter	Details
Task	Audio Classification
Architecture	LSTM
Features	MFCC
Sample Rate	22,050 Hz
Maximum Duration	10 seconds
Framework	TensorFlow / Keras
Audio Processing	Librosa

🎧 Preprocessing

1. Load heart sound audio.
2. Resample to 22,050 Hz.
3. Normalize the audio.
4. Standardize the clip duration.
5. Extract MFCC features.
6. Pass the feature sequence to the LSTM model.

Important: Inference must use the same preprocessing and feature-extraction pipeline used during training.

📤 Output

The model produces a predicted class and confidence score:

Prediction: <CLASS_NAME>
Confidence: <VALUE>

Confidence should not be interpreted as medical certainty.

💻 Run Locally

git clone https://github.com/mdzaheerjk/AI-Powered-Heart-Murmur-Detection-System.git

cd AI-Powered-Heart-Murmur-Detection-System

pip install -r requirements.txt

streamlit run app.py

🌐 Live Demo

🚀 Streamlit App: https://ai-powered-heart-murmur-detection--system.streamlit.app/

🛠️ Tech Stack

Python • TensorFlow • Keras • Librosa • NumPy • Pandas • Matplotlib • Streamlit

⚠️ Limitations

Model performance can be affected by:

• Audio quality and background noise
• Recording devices
• Dataset size and diversity
• Class imbalance
• Differences between training and real-world data

This model has not been clinically validated and should not be used as a standalone diagnostic system.

🔮 Future Improvements

• CNN + LSTM hybrid architecture
• Transformer-based audio models
• Larger and more diverse datasets
• Audio augmentation
• Explainable AI
• Cross-dataset evaluation
• Model optimization
• Clinical validation

👨‍💻 Author

Md Zaheer JK

AI/ML • Deep Learning • Generative AI • NLP • Computer Vision • MLOps

🔗 GitHub: https://github.com/mdzaheerjk 🤗 Hugging Face: https://huggingface.co/zaheerjk

📜 License

MIT License.

---

🫀 Turning Heart Sounds into Machine Learning Insights