Bayesian Modelling of the ECG

This project focuses on developing a Bayesian Neural Network model to analyze physiological signals, specifically Electrocardiogram (ECG) data. The primary goal is to accurately model ECG signals while quantifying the model's uncertainty. The implementation consists of a Convolutional Autoencoder architecture that processes ECG signals and applies approximate Bayesian inference using Monte Carlo dropout. This approach provides a robust way to model ECG signals, enabling better understanding and interpretation of the data while accounting for uncertainties. The project utilizes ECG data from the EPHNOGRAM database, and the implemented model has undergone several iterations to optimize its performance.

Prerequisites

• pandas
• numpy
• matplotlib
• keras
• tensorflow
• sklearn
• scipy

Running the project

1. Clone the project by using git clone https://github.com/hamza-mughees/ECG-Modelling.git
2. Insall the data from the EPHNOGRAM database and place it into the root directory.
3. Create a res directory in root.
4. Navigate into the src directory.
5. Create the data:
   1. For a single patient, run python create_data_singlePatient.py
   2. For all patients, run python create_data_allPatients.py
6. Train the model: run python autoencoder.py.
   1. For bayesian inference, make sure to assign True to the bayes variable in globals.py.
7. To analyze a trained model, run performance.py. Update the ID to that of the model in the out directory. The latest model would always be the one at the very bottom.