Denoising autoencoder for f-wave extraction – semester project
Atrial fibrillation (AF) is a major public health concern worldwide. Cardiologists studying AF observe the electrical activity of the heart using the surface electrocardiogram (ECG), which can be considered as a superposition of three components: ventricular activity (QRST complexes), atrial activity, and noise. It is often desirable to extract only the atrial activity, called fibrillatory waves (f-waves), which have irregular amplitude, shape, and frequency. Extracting f-waves is not trivial; they have small amplitude and overlap with ventricular activity in the frequency domain.
Many techniques have been developed for extracting the atrial signal, with varying levels of success. The goal of this project is to explore a self-supervised deep learning-based architecture for the task of atrial activity extraction. The architecture will be trained using simulated signals and performance will be evaluated on both simulated and real signals.
Requirements: The project will be implemented in Matlab/Python so good knowledge is required. Prior experience with Pytorch would be beneficial, as would previous experience with time series signal processing and machine learning. This project is ideal for a computer scientist or engineer interested in learning about the intersection between machine learning and signal processing with a biomedical application.