An automated pipeline for processing and analysis of clinical fetal diffusion MRI
Résumé
In this work, we present an automated pipeline for processing diffusion MRI (dMRI) data acquired in fetuses in clinical routine. In-vivo fetal dMRI provides critical insight into brain development but is challenging due to motion, low SNR ratio, and artifacts. These challenges are prominent in the clinical routine, with additional constraints on the duration of the scanning session. Existing state-of-the-art pipelines are designed for high-quality research acquisition protocols that are not feasible in routine clinical practice. We introduce a comprehensive framework that integrates robust opensource tools to handle clinical fetal dMRI, performing motion correction, anatomical standardization, and diffusion tensor estimation. The pipeline is complemented by an extended set of anatomical parcellations for the dHCP fetal atlas, enabling detailed regional analysis. Applied to a large clinical cohort (N=314 sessions), the pipeline successfully processed all datasets without manual intervention. Following quality control, the resulting diffusion metrics from the 152 datasets with the best quality score revealed developmental trajectories consistent with established literature, demonstrating the pipeline's validity and robustness. The code is available at https://github.com/MecaLab/Fetal_DWI.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |