This repository explores multiple physics-based preprocessing techniques—Euler Differentiation, Hilbert Transform, and Gauss-Legendre Integration—to enhance the segmentation of ECG waveforms (P, QRS, and T) using an LSTM model.
Electrocardiogram (ECG) signals contain three primary waveforms:
- P-wave: low-frequency, low-amplitude
- QRS-complex: sharp, high-frequency
- T-wave: smooth, moderate amplitude
Each wave is preprocessed using a tailored mathematical method to match its physical characteristics before being segmented using an LSTM model.
| Method | Description | Best for |
|---|---|---|
| Euler Differentiation | Highlights sharp slope changes | QRS-complex |
| Hilbert Transform | Enhances amplitude & phase info (envelope) | P & T waves |
| Gauss-Legendre | Smooth integration, retains morphology | All segments |
| High-pass Filtering | Removes baseline drift | All |
- Architecture: 2-layer LSTM
- Input: Preprocessed 1D ECG signal
- Output: Predicted wave labels (P, QRS, T)
- Evaluation Metrics: Accuracy, Test Loss, Inference Time
Figure: Loss convergence across preprocessing techniques
| Method | CPU Time (ms) | GPU Memory (MB) |
|---|---|---|
| No Preprocessing | 40.2 ms | 110 MB |
| Euler | 41.5 ms | 112 MB |
| Hilbert | 43.0 ms | 115 MB |
| Gauss-Legendre | 45.0 ms | 120 MB |
| File | Description |
|---|---|
src/preprocessing.py |
All signal transformation methods |
src/model_lstm.py |
LSTM-based segmentation model |
main.py |
Run complete pipeline (preprocess → segment → plot) |
git clone https://github.com/username/ecg-segmentation-preprocessing.git
cd ecg-segmentation-preprocessing
pip install -r requirements.txt