harryjdavies/HeartGPT
📦 Open Source Projectharryjdavies
An interpretable, pre-trained transformer model designed for analyzing heart-related time-series data like ECG and PPG signals.
HeartGPT represents a significant step forward in applying transformer-based architectures to the domain of physiological time-series analysis. Unlike generic time-series models, HeartGPT is specifically architected to handle the unique morphology and temporal dependencies found in cardiac biosignals like Electrocardiograms (ECG) and Photoplethysmograms (PPG). The core innovation lies in its focus on interpretability, addressing the 'black box' nature of deep learning in medical settings. By utilizing pre-trained transformer blocks, the model can effectively capture long-range dependencies in heart rhythm data, which is crucial for detecting subtle anomalies that might indicate cardiovascular issues. The repository provides a robust PyTorch-based implementation, enabling developers to fine-tune the model on specific heart datasets. Its design facilitates the extraction of attention-based features, which can be mapped back to specific segments of the heart signal, providing clinicians with a visual and quantitative rationale for the model's predictions. This makes it a powerful tool for researchers building next-generation eHealth monitoring systems, wearable health devices, and automated diagnostic pipelines.
💡Highlights
- ├─Interpretable transformer architecture
- ├─Optimized for ECG and PPG signals
- └─PyTorch-based deep learning framework
🎯For
- ├─AI Researchers
- ├─Biomedical Engineers
- └─Healthcare Data Scientists