Master's thesis opportunity about "Conditional Diffusion Generative Models for Cardiac MRI"

31.10.2025, Abschlussarbeiten, Bachelor- und Masterarbeiten

This thesis explores conditional diffusion models for cardiac MRI super-resolution by leveraging auxiliary information such as routine localizer scans. Despite their low quality, localizers provide comprehensive 3D coverage. The goal is to reconstruct high-quality images from these scans, narrowing the gap between non-diagnostic and diagnostic MRI and enabling faster, potentially localizer-only cardiac imaging workflows.

Super-Resolution Cardiac MRI using Conditional Diffusion Models

The aim of this MS thesis is to exploit readily available side information (e.g., auxiliary images or vectors) to generate missing slices for MR super-resolution using conditional diffusion models. Specifically, in cardiac MRI protocols, non-diagnostic localizer scans are routinely acquired to visualize the heart before detailed, time-resolved diagnostic sequences. Although localizers are static and of lower image quality, they offer a more comprehensive 3D overview of the heart. This work explores conditional diffusion-based super-resolution techniques to reconstruct high-quality 2D and potentially 3D cardiac images from these localizers, bridging the quality gap between non-diagnostic and diagnostic scans and paving the way for faster, potentially localizer-only cardiac imaging pipelines.

The student will first explore the landscape of super resolution and generative AI models. Evaluation will be done using ground truth diagnostic cardiac MRI data at hand. You'll have a chance to: -Investigate and compare state-of-the-art super-resolution techniques -Work with real-world cardiac MRI data -Explore clinical relevance and reconstruction fidelity metrics

References:

1- Wang, Shuo, et al. "Joint motion correction and super resolution for cardiac segmentation via latent optimisation." International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer International Publishing, 2021. 2- He, Xiaoxiao, et al. "Dmcvr: Morphology-guided diffusion model for 3d cardiac volume reconstruction." International conference on medical image computing and computer-assisted intervention. Cham: Springer Nature Switzerland, 2023. 3- Z. Zhan et al., "Conditional Image Synthesis with Diffusion Models: A Survey," arXiv preprint arXiv:2409.19365, 2024. 4- R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, "High-resolution image synthesis with latent diffusion models," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10684-10695.

Kontakt: s.kafali@tum.de, niklas.bubeck@tum.de

Mehr Information

https://thesis.aet.cit.tum.de/topics/8e863ca2-9741-4a0c-887e-19303c1caee4