Data is the keystone of deep learning models, directly influencing their performance and capabilities. Acquiring large and diverse annotated datasets in the medical field, however, poses significant challenges. These challenges mainly arise from privacy concerns, the need for domain expertise, and the time-intensive nature of medical data annotation. 

Synthetic data offers a promising solution to address data scarcity. One effective technique involves using text- and mask-conditioned diffusion models. In recent years, research focused on medical image inpainting and editing using this technique has seen growth [1-4]. However, there remains a noticeable gap when compared to developments in the general domain.

Become familiar with text-conditioned diffusion models for image generation and editing, particularly those designed for chest X-rays (CXR). Subsequently, improve our current approach, focusing on realism and mitigating artifact generation.Your qualifications:
We are looking for a highly motivated Master’s student in Informatics, Mathematics, Physics, or Engineering, with
 

• Advanced Python programming skills and experience with common deep learning frameworks, such as PyTorch.
• Strong abilities to work independently and as part of a team, with an interest in interdisciplinary research.
• Interest in diffusion models and multimodal learning, particularly image and text modalities.
• Background in deep learning is essential. The project may involve model training, therefore, experience in training is highly desirable.

• Join an exciting, relevant, and cutting-edge medical ongoing project aimed at publishing in a top-tier conference.
• Opportunity to bring in your ideas.
• Close supervision and contact with an interdisciplinary network of experts in computer vision and medical imaging.
• Access to computational resources and datasets.

Start date: June 1, 2025 (flexible)

How to apply:
Send your CV and transcript of records to Andrea Posada Cardenas (andrea.posada-cardenas@tum.de). Please include links to any previous (e.g., GitHub profile and papers) if available.

[1] Hashmi, A.U.R., Almakky, I., Qazi, M.A., et al.: XReal: Realistic Anatomy and Pathology-Aware X-ray Generation via Controllable Diffusion Model. arXiv:2403.09240 (2024)
[2] Jin, Q., Gerych, W., Ghassemi, M.: MaskMedPaint: Masked Medical Image Inpainting with Diffusion Models for Mitigation of Spurious Correlations. arXiv:2411.10686 (2024)
[3] Pérez-García, F., Bond-Taylor, S., Sanchez, P.P., et al.: RadEdit: Stress-Testing Biomedical Vision Models via Diffusion Image Editing. In: ECCV. p. 358–376. Springer-Verlag (2024). https://doi.org/10.1007/978-3-031-73254-6_21
[4] Rouzrokh, P., Khosravi, B., Faghani, S., et al.: Multitask Brain Tumor Inpainting with Diffusion Models: A Methodological Report. arXiv:2210.12113 (2023)