Motion artifacts remain one of the most persistent causes of image degradation in medical imaging. In MRI, patient motion, respiration, cardiac pulsation, and involuntary motion can produce ghosting, blurring, ringing, and local structural corruption because data are acquired over relatively long time intervals in k-space. Classical reviews emphasize that MRI is particularly sensitive to motion and that no single correction strategy solves all motion scenarios (Zaitsev et al, 2015). In clinical routine, these artifacts may reduce diagnostic confidence, force repeat scans, and negatively affect downstream tasks such as segmentation, registration, and quantitative analysis.
Recent deep-learning methods have substantially improved retrospective motion correction, especially in brain MRI. In particular, generative approaches are becoming a strong research direction because they can model more realistic image priors than earlier CNN-only restoration methods. Recent work includes diffusion-based MRI correction frameworks such as MoCo-Diff (Li et al., 2024), which combines a first-stage correction network with diffusion-based restoration for 3D MRI volumes, and new diffusion-based CT motion correction methods such as HeadMotion-EDM (Chen et al., 2024), which show that conditional diffusion can also suppress motion artifacts in portable head CT. At the same time, latent diffusion frameworks such as MAISI and MedLoRD (Guo et al., 2024) demonstrate that high-resolution medical image generation can be made substantially more scalable by operating in compressed latent spaces rather than in full image space. These developments suggest that conditioned latent diffusion is a promising foundation for motion artifact suppression, because it combines strong generative priors with better computational feasibility for high-resolution 3D medical imaging.

The main objective is to develop and validate a generative AI method for motion artifact suppression in medical images based on conditioned latent diffusion.
Specific objectives are to: (1) design a conditioning strategy that combines the motion-corrupted with motion-aware priors (2) train and compare different latent diffusion variants for motion artifact suppression (3) benchmark the proposed method against classical and modern motion-correction baseline

The developments will be performed in close collaboration with the university hospital UZ Brussel. The research will mainly address the exploration of novel generative AI methodologies for motion artifact suppression from 3D medical volumes, with primary focus on MRI and optional validation on CT, depending on dataset availability. This thesis will expose the student to a broad set of tasks and competences, covering experimental work and algorithm development and testing. The following tasks can be distinguished:
•
Literature review on motion artifacts in MRI and CT, retrospective motion correction, diffusion models, latent diffusion, and weakly supervised medical image restoration.
•
Implementation of preprocessing, motion simulation and baseline correction methods, and the proposed conditioned latent diffusion model in Python (PyTorch and MONAI).
•
Design of motion-aware conditioning inputs with code deployment on GPU clusters (VSC).
•
Training and validation on simulated and real motion-corrupted data, including ablation studies on different conditioning strategies and latent-space configurations.
•
Comparison with state-of-the-art baselines using quantitative metrics and targeted visual analysis in motion-sensitive anatomical regions.
•
Interpretation of failure cases, discussion of clinical relevance, and full thesis writing.