Classification of atrial ectopic origins into spatial segments based on the 12-lead ECG

Abstract

The aim of this thesis is to study, whether the standard 12-lead ECG can be used to localize the origin of atrial ectopic beats. A neural network shall be used to classify ECG signals into predefined spatial segments of the atria. This work will involve the following steps: *Literature research about: -Atrial ectopic beats and their relation to changes in the ECG features -Methods to split the atria into spatial segments -Similar ML approaches to solve the inverse problem -Design, training and validation of neural networks *Splitting the atria into spatial segments *Excitation simulations using Fast Marching *Forward calculations *Analysis of ECG variations and feature extraction *Neural network design and training *Application of to simulated data / validation *(Application to measured data)