A comparative analysis of automated MRI brain segmentation in a large longitudinal dataset: Freesurfer vs. FSL

The study of brain volumetry and morphology of the different brain structures can determine the diagnosis of an existing disease, quantify its prognosis or even help to identify an early detection of dementia. Manual segmentation is an extremely time consuming task and automated methods are thus, gaining importance as clinical tool for diagnosis. In the last few years, AI-based segmentation has delivered, in some cases, superior results than manual segmentation, in both time and accuracy. In this study we aim at performing a comparative analysis of automated brain segmentation. In order to test the performance of automated segmentation methods, the two most commonly used software libraries for brain segmentation Freesurfer and FSL, were put to work in each of the 4028 MRIs available in the study. We find a lack of linear correlation between the segmentation results obtained from Freesurfer and FSL. On the other hand. Freesurfer volume estimates of subcortical brain structures tends to be larger than FSL estimates of same areas. The study builds on an uniquely large, longitudinal dataset of over 4,000 MRIs, all performed with identical equipment to help researchers understand what to expect from fully automated segmentation procedures.