ECG Signal Analysis and Classification

This project provides a complete pipeline for ECG signal classification using deep learning. It includes raw and processed data, preprocessing scripts, model implementations, Jupyter notebooks, and evaluation results.

---

Project Structure

1. Data

• Folder: data/
• Description:
  • Contains raw and processed ECG data.
  • Raw data includes unprocessed ECG signals.
  • Processed data contains features extracted from ECG signals, such as wavelet coefficients, R-R intervals, and PQRS complexes.
• Files:
  • physionet2017.csv: Raw ECG signals with labels.
  • dwt_features_ecg.csv: Processed ECG features for machine learning.
• Details:
  • Raw Data:
    • Columns: Time-series signal values with a label column.
    • Example:

      Signal1	Signal2	...	Label
      0.1	0.2	...	0

  • Processed Data:
    • Contains extracted features like PQRS complexes, R-R intervals, wavelet variance, and entropy.

---

2. Models

• Folder: models/
• Description:
  • Contains Python scripts implementing various deep learning architectures for ECG classification.
• Files:
  • resnet.py: Implements the ResNet model.
  • resnet_A.py: ResNet with attention.
  • cnn_bilstm.py: Basic CNN-BiLSTM model.
  • cnn_bilstm_a.py: CNN-BiLSTM with attention.
  • transformer.py: Transformer-based model.
• Model Highlights:
  • ResNet:
    • Deep residual learning for feature extraction.
    • Includes attention-based variants.
  • CNN-BiLSTM:
    • Combines convolutional feature extraction with BiLSTM for sequence modeling.
    • Includes attention-enhanced variants.
  • Transformer:
    • Uses multi-head self-attention for long-range dependency modeling.

---

3. Notebooks

• Folder: notebooks/
• Description:
  • Jupyter notebooks demonstrating model training, evaluation, and experimentation.
• Files:
  • resnet.ipynb: ResNet training and evaluation.
  • resnet_a.ipynb: ResNet with attention.
  • resnet-encoder.ipynb: ResNet for feature extraction tasks.
  • cnn_bilstm.ipynb: CNN-BiLSTM training and evaluation.
  • cnn_bilstm_a.ipynb: CNN-BiLSTM with attention.
  • bilstm_a.ipynb: BiLSTM with attention for temporal analysis.
  • transformer_resnet16.ipynb: Hybrid Transformer + ResNet model.
  • evaluation.ipynb: Unified model evaluation.

---

4. Preprocessing

• Folder: preprocessing/
• Description:
  • Scripts for preprocessing ECG data, tailored to different models.
• Files:
  • cnn_bi_lstm_preprocessing.py: Prepares data for CNN-BiLSTM.
  • resnet_preprocessing.py: Prepares data for ResNet.
• Functionality:
  • Load and normalize ECG signals.
  • Extract features like PQRS complexes and R-R intervals.
  • Split data into training and testing sets.

---

5. Results

• Folder: results/
• Description:
  • Contains visualizations and performance metrics for all models.
• Files:
  • Classification Reports:
    • cnnbilstm_classification_report.png: CNN-BiLSTM classification report.
    • resnet_classification_report.png: ResNet classification report.
  • Confusion Matrices:
    • cnnbilstm_best_conf_matrix.png: Best CNN-BiLSTM confusion matrix.
    • resnet_best_conf_matrix.png: Best ResNet confusion matrix.
    • resnet+encoder_conf_matrix.png: Confusion matrix for ResNet + Encoder.
  • Training Metrics:
    • resnet_epoch_loss.png: Training loss for ResNet.
    • resnet_epoch_accuracy.png: Training accuracy for ResNet.
  • ROC Curve:
    • AUC-ROC_resnet+encoder.png: ROC curve for ResNet + Encoder.