M.Sc. Jule Bender
M.Sc. Julian Mierisch

Download als PDF

Motivation
Synthetic patient cohorts are a valuable alternative to in vivo studies. Generating synthetic ECGs requires models of both the heart and the torso; however, realistic torso models for synthetic patients are not publicly available. While some patient-specific and statistical shape models exist, including 25 derived models available at IBT, they lack sex-specificity. Recently, conditional variational autoencoders (cVAES) have been successfully used to model anatomical structures but have not been applied for torso models yet and could serve as a valuable tool to generate torso models. In general, a small number of representative models is preferred to minimize computational costs. Furthermore, previous studies have shown that torso characteristics primarily influence P-wave amplitude. However, it remains unclear how many torso models are required for in silico studies to draw relevant conclusions for the general population.

Project Description

Your task:
- Development of a (conditional) generative torso model (sex & BMI): Data preprocessing, model training, and evaluation
- Electrocardiogram (ECG) simulations
- Data Analysis: Determine how many different torso models and which kind of torso models are needed to capture relevant ECG variability

Notes
- Programming skills in Python are beneficial
- Basic knowledge about cardiac electrophysiology and ECGs is beneficial
- The thesis can be written in German or English

The specific focus of the project can be adjusted according to your interests.