Reconstruction haute-résolution de données IRM pour l’étude du développement cérébral

Fetal MRI is commonly used for early diagnosis of developmental brain abnormalities. However, fetal movement during acquisition induces artifacts that degrade image quality. To mitigate these artifacts, ultrafast acquisition sequences are employed, acquiring multiple stacks of orthogonal slices (axial, sagittal, coronal). However, inter-slice motion persists. If left uncorrected, this leads to artifacts in the 3D reconstructed volume. Traditional motion correction methods rely on a two-step iterative process (registration followed by reconstruction) and often discard an excessive number of misaligned slices, thus deteriorating the quality of the reconstructed image. In this thesis, we propose a novel reconstruction-independent motion correction approach. Our method exploits the intersection of orthogonal slices to estimate the motion of each slice by minimizing the differences in intensity profiles along these intersections. We have also developed a machine learning-based classifier to identify misaligned slices. The parameters of these slices are then refined using a multi-start optimization approach. The results on simulated and real data demonstrate the interest of the method compared with the state of the art.