Marvin Seyfarth
PhD Student
Institute for Artificial Intelligence in Cardiovascular Medicine
University Hospital Heidelberg
About me
I develop and evaluate generative models for medical imaging, with a focus on efficient 3D and 4D synthesis.
My research connects high-resolution image synthesis with computational efficiency, anatomical and temporal structure, diversity, and memorization.
Publications
Generative medical imaging, from models to evaluation.
01 / 02Publications
Generative medical imaging, from models to evaluation.
02 / 02Rethinking Diversity Metrics in Medical Imaging with Wasserstein Distance
Latent Pollution Model: The Hidden Carbon Footprint in 3D Image Synthesis
Latent diffusion for single-pass full 4D cine cardiac magnetic resonance imaging synthesis with spatiotemporal autoencoding
Beyond visual realism: assessing coronary artery dominance diversity in AI-generated synthetic coronary computed tomography angiography images
News
Papers, milestones, and research updates.
Research milestone
CardioDiT and VolDiT accepted at MICCAI 2026
Our work on 4D cardiac MRI and controllable 3D synthesis will appear at MICCAI.
-
CardioDiT preprint released
Joint spatial and temporal modeling for 4D cardiac MRI synthesis.
-
VolDiT preprint released
Controllable volumetric medical image synthesis with diffusion transformers.
-
Rethinking diversity evaluation
WAD-Div measures diversity beyond visual similarity.
-
MedLoRD published online
Our SASHIMI 2025 paper examines high-resolution 3D CT synthesis under limited computational resources.
-
Patient-data memorization study published
Our Nature Biomedical Engineering study investigates patient-image memorization in unconditional latent diffusion models.
Talks & research notes
Presentations and explanations of my research.
Talks & outreach
Upcoming ·
From papers to models and back
How to become an AI-empowered scientist using AI agents
ESC Digital & AI Summit 2026
Basel, Switzerland
Research notes
Foundations ·
From noise to anatomy
Diffusion models for medical image synthesis
An introduction to denoising, conditioning, latent spaces, and evaluation.
- From noise to images
- Structure and conditioning
- Evaluating synthetic data