Magnetic Resonance Imaging (MRI) has become a reference modality for diagnosing and monitoring neurological disorders. However, acquiring high-resolution (HR) brain images remains challenging in clinical practice due to limited scan time, patient comfort constraints, and image degradation caused by patient motion. The increased signal enabled by higher magnetic field strengths can be invested to achieve higher spatial resolution within the same acquisition time. This project aims at taking advantage of the unprecedented spatial resolution achievable with the 11.7T Iseult MRI scanner, currently the most powerful MR scanner in the world, to train a machine learning-based super-resolution (SR) model that enhances the spatial resolution of 3T MRI images acquired in clinical practice. Current SR approaches are typically trained on public datasets, using pairs of high- and low-resolution images, with the low-resolution data synthetically generated from the high-resolution images. In this project we will use a real dataset consisting of 3T and 11.7T images acquired from the same cohort, ensuring higher anatomical fidelity and enabling a rigorous assessment of hallucination risks, i.e. of generating anatomically incorrect details that could be misinterpreted by the radiologists. The project will involve the following steps: improving the quality of 11.7T images (through motion correction and artifact reduction), acquiring pairs of images at 3T and 11.7T, developing and validating SR models, and finally assessing their generalizability on public datasets. This work supports the integration of reliable SR methods into clinical practice, allowing conventional MRI scanners to benefit indirectly from Iseult's unique capabilities.