Project information
- Title: Heart Disease Recognition from Heart Beat Audio Signals
- Category: Advanced Biomedical Machine Learning
- Author: Andrea Alberti - Davide Ligari
- Project date: 3 July, 2024
- Project URL: Github-Heart Disease Recognition
Heart Disease Recognition
Advanced Machine Learning for Early Detection of Heart Diseases through Audio Analysis
Record your heartbeat, detect your health: Unlock early heart disease insights with just your phone
About
Heart diseases are the leading cause of death worldwide, often due to late diagnosis when symptoms become severe. Traditional diagnostic methods frequently miss early signs, delaying treatment. This project aims to develop an automated system to detect heart diseases early using machine learning techniques. The primary goals include creating accurate predictive models for classifying heart sound recordings, balancing computational efficiency with diagnostic precision, and incorporating explainability features to help medical professionals understand the model's predictions.
Results
The project report includes key findings from the Prevention and Support models:
-
Prevention Model:
Aimed at early heart disease diagnosis with minimal false normal predictions.
- Best AUC: 0.96 (MLP_Ensemble5)
- TPR at 1% FPR: 43.4%
- TPR at 5% FPR: 74.3%
- TPR at 10% FPR: 86.6%
- TPR at 20% FPR: 95.8%
-
Support Model:
Designed to assist clinicians with accurate classification across all heart disease classes.
- Highest Macro F1 Score: 81.58 (MLP_Ensemble2)
- Highest MCC: 81.53 (MLP_Ensemble2)
- Best Overall Risk Score: MLP_Rollercoaster
- Best for Extra Systoles: MLP_Ensemble3
For a comprehensive overview of the results, please refer to the full project report.
License
This project is licensed under the MIT License. Feel free to use, modify, and distribute the code as per the terms of the license.