🫀 Arrhythmia Classification on CPSC 2018 Dataset

This project focuses on multi-class arrhythmia classification using the CPSC 2018 12-lead ECG dataset. It supports training and evaluation on both 12-lead signals and single-lead signals, with configurable model architectures and MLflow-based experiment tracking.

📌 Table of Contents

1. Project Structure
2. CPSC Dataset
3. Installation
4. Modules Breakdown
5. Installation
6. How to use

🧠 Project Structure

├── config.py                                   # Configuration file for paths and settings
├── cpsc_data_prep_csv.py                       # Script to prepare CSVs from raw data
├── data_processing
│   ├── cpsc_signal_loader.py                   # Signal loading and preprocessing
│   ├── data_generator.py                       # Data generator for model input
│   ├── data_loading.py                         # Load and preprocess dataset
│   ├── data_reduction.py                       # Dataset size reduction scripts
│   ├── data_splitter.py                        # Train/Val/test splitting utilities
├── models
│   ├── checkpoints                             # Model checkpoints (saved weights)        
│   ├── final                                   # Final trained models
│   ├── misclassified_samples                   # Incorrect predictions for analysis
│   ├── model
│   │   ├── cnn1d_resent_18.py                  # resnet model  
│   │   └── cnn_attentaion_bilstm_improve.py    # cnn_attentaion_bilstm_model
│   ├── model_paths                             # Model path utilities
│   ├── model_plots                             # Training and evaluation plots
│   ├── model_results                           # Evaluation results (metrics)
│   └── Spectrograms                            # Spectrogram representations of signals
├── README.md                                   # Project documentation
├── requirements.txt                            # Python package dependencies
├── test_cpsc.py                                # Testing script
├── train_custom_log_cpsc.py                    # Training script with MLflow logging
├── utils
│   ├── model_selector.py                       # Model selection utilities
│   ├── utils.py                                # Utility functions and global configs and input sizes paths
│   └── visualization.py                        # Visulization utility
└── visualize_signal.py                         # ECG signal visulization script

🛢️ CPSC Dataset

CPSC dataset is opensource dataset avaliable here

⚙️ Installation

1. Colne the repository:

   clone repo

2. Install dependencies:

pip install -r requirements.txt

3.Download the datasets:

• place the datasets as per your convinent give the correspoding path in the config file and utils file

🚀 How to use

1. To train a model

python3 train_custom_log_cpsc.py

• config.py: Contains model selection, MLflow enable/disable settings, number of epochs, and all other required configuration parameters.

• utils.py: Contains file paths, input sizes, and other utility constants needed across the project.

  • Models will be saved in models/checkpoints/.
  • Training history will be saved in mlflow server

2. To test a model

python3 test_cpsc.py

• training logs ,tesing logs, model performace metrics and model are saving in mflow server

3. To Visualization

python visualize_signal.py --file_path data/test_data.pkl --plot_signal --sampling_rate 250

• provide reqiured args for visulization of the signal