Overview
This master’s thesis addressed a core methodological gap in physiological signal generation: the lack of a standardized, reproducible way to evaluate synthetic ECG outputs from generative deep learning models.
The code repository for this project is available on GitHub.
Thesis
- Title: A Systematic Evaluation Framework of Generative Deep Learning for 10-second 12-lead Synthetic ECG Signals
- Institution: KU Leuven, Department of Electrical Engineering (ESAT), STADIUS
- Promotor: Prof. Maarten De Vos
- Supervisor Lore Van Santvliet
- Grade: 17/20
- Full text: KU Leuven repository
Contribution
I developed an end-to-end benchmarking pipeline that enables reproducible training and systematic evaluation of generative models for 12-lead ECG synthesis. The framework supports fair comparison across model families and provides a foundation for more reliable reporting in this area.
Focus Areas
- Generative deep learning for physiological time series
- Benchmark design and model comparison
- Reproducibility in biomedical AI