AtrialFibrillation-Classification-SMoLK

Overview

Atrial Fibrillation Classification SMoLK is a deep learning project designed to classify arrhythmias, specifically focusing on atrial fibrillation (Afib), using the MIT-BIH Arrhythmia Database and SMoLK (Sparse Mixture of Learned Kernels). The project leverages a neural network model with learned filters to achieve high classification accuracy across multiple classes.

Features

• Data Loading & Preprocessing: Efficient handling of the MIT-BIH dataset with resampling and window extraction.
• Neural Network Model: Implementation of the SMoLK (Sparse Mixture of Learned Kernels) model with multiple convolutional layers.
• Training & Evaluation: Comprehensive training scripts with cross-validation and holdout set evaluation.
• Metrics Calculation: Detailed metrics including Sensitivity, Specificity, AUC, and F1 Score.
• Reproducibility: Seed setting for consistent results across runs.

Installation

Prerequisites

• Python 3.8 or higher
• Conda (optional but recommended)

Clone the Repository

git clone https://github.com/abdulvahapmutlu/atrial-fibrillation-smolk.git
cd atrial-fibrillation-smolk

Using Conda

conda create -n afib-classification python=3.8
conda activate afib-classification
pip install -r requirements.txt

Using pip

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

Update Configuration:

Open src/config.py and set the DATASET_PATH to point to the extracted dataset directory.

DATASET_PATH = "/path/to/mit-bih-database"

Configuration

All configurable parameters such as hyperparameters, file paths, and other settings are defined in src/config.py. Adjust these settings as needed for your experiments.

Usage

Training the Model

To train the model, execute the training script using the provided shell script:

bash scripts/run_training.sh

Alternatively, you can run the training script directly:

python src/train.py

Evaluating the Model

After training, evaluate the model using the evaluation script:

bash scripts/run_evaluation.sh

Alternatively, run the evaluation script directly:

python src/evaluate.py

Results

Cross-Validation Results

Class	Sensitivity	Specificity	AUC
Normal	0.939	0.957	0.988
Afib	0.869	0.965	0.972
Other	0.947	0.977	0.993

F1 Score: 0.825 ± 0.165

Holdout Set Results

Class	Sensitivity	Specificity	AUC
Normal	0.939	0.967	0.991
Afib	0.917	0.965	0.984
Other	0.955	0.978	0.995

F1 Score: 0.832 ± 0.160

License

This project is licensed under the MIT License.