VAE-ECG

A VAE-based framework for reconstructing ECG signals and conducting statistical hypothesis testing. Detailed methodology and results can be found in the report.pdf.

To download the dataset used for training, run the following command in the project directory:

wget -r -N -c -np https://physionet.org/files/challenge-2021/1.0.3/training/chapman_shaoxing/

File Overview (src directory)

.py

• helper_code.py: Utility functions to support various tasks throughout the project.
• preprocessing_indexing.py: Preprocesses and indexes the ECG data before training.
• eda.py: Contains exploratory data analysis scripts for visualizing and understanding the dataset.
• datasets.py: Defines custom datasets and prepares the data structure for input to the models.
• dataloaders.py: Handles the loading and batching of ECG data for model training.
• models.py: Defines the VAE based neural network architectures used for ECG reconstruction.
• trainer.py: Manages the training loop, validation, and testing phases for the models.
• hypoth_testing.py: Performs hypothesis testing, including statistical tests like Mann-Whitney U.

constants

• label_index.json: Maps labels to their corresponding index in the dataset.
• labels.csv: Contains the only diagnosis labels.
• multi_hot_encodings.h5: Stores the multi-hot encoded diagnostics for the full dataset used for training.
• scaling_params.json: Stores global min-max values for scaling the ECG data and age.