Master Thesis: Generating 3D Brain Tumor Images from Tumor Concentrations using Diffusion Models

10.03.2025, Abschlussarbeiten, Bachelor- und Masterarbeiten

Master Thesis: Generating 3D Brain Tumor Images from Tumor Concentrations using Diffusion Models

Accurate and realistic medical imaging is fundamental for effective diagnosis, treatment planning, and outcome prediction in clinical practice. However, generating detailed 3D medical images, such as MRI scans, from limited data remains a challenge. Recent breakthroughs in generative artificial intelligence, particularly diffusion models, provide a promising pathway to synthesize high-quality medical images [1]. Additionally, synthesizing realistic longitudinal medical data can facilitate studies on disease progression and enhance privacy research by enabling the creation of synthetic datasets that preserve patient confidentiality.

In this thesis, your goal will be to investigate state-of-the-art diffusion architectures and innovative training methodologies to generate realistic 3D brain tumor images from tumor concentration data. You will explore how diffusion models can be conditioned effectively on abstract tumor representations to produce clinically meaningful MRI-like images. The exact direction of the thesis is flexible and will be collaboratively defined, incorporating your interests and strengths.

Your qualifications:
• Motivated by developing innovative AI solutions with real-world clinical impact.
• Currently enrolled in computer science, mathematics, physics, biomedical engineering, or a related discipline.
• Strong Python programming skills, preferably experienced with frameworks like PyTorch
• Previous experience with diffusion models, generative AI, or medical imaging is beneficial but not mandatory.

What we offer:
• The opportunity to contribute to cutting-edge research at the intersection of AI and medical imaging.
• Close mentorship and access to advanced computational resources and hardware.
• Collaborative environment with interdisciplinary experts in artificial intelligence, medical imaging, and computational science.


[1] Dorjsembe, Zolnamar, et al. "Conditional diffusion models for semantic 3d brain mri synthesis." IEEE Journal of Biomedical and Health Informatics (2024).


How to apply: Send your CV and transcript via email to j.weidner@tum.de and niklas.bubeck@tum.de

Kontakt: j.weidner@tum.de and niklas.bubeck@tum.de