An optimized deep learning-based method for cerebellum segmentation

This project aims to develop a tool that will help standardize and automate the analysis of images of the cerebellum obtained in MRI studies in FA patients. A critical gap in FRDA therapeutic development is the lack of sensitive clinical or biological outcome measures for treatment monitoring, pharmacodynamic tracking, and patient stratification. In this sense, neuroimaging-based parameters have emerged as potential candidates, although further studies are necessary to validate them. In particular, the cerebellum is a major site of neurodegeneration in FRDA, with a recognized progressive pattern of damage in the cerebellar cortex and pathways. Unfortunately, the segmentation of the cerebellum and its structures is a challenging task due to the intricately folded cerebellar cortical tissue and its proximity to the cerebral cortex. Efforts towards creating accurate tools to address these challenges are underway. Nevertheless, these tools suffer from accuracy, reproducibility, and generalizability limitations, particularly when applied to patients with FRDA. The investigators, therefore, propose to leverage the availability and maximize the utility of the high-quality data resources that are being collected and aggregated through significant funding investments in TRACK-FA, ENIGMA-Ataxia, and other in-house projects to deliver updated and optimized analytical tools. In particular, they will develop a comprehensive deep learning-based model for cerebellum segmentation to achieve automated and standardized tools that can be readily deployed for use in observational, natural history, and treatment trials. An accurate and sensitive quantitative assessment of cerebellum structure is essential to maximize the likelihood of successful neuroimaging biomarker discovery.