The FreeSurfer subcortical training set is derived from manually annotated brain MRI scans. Expert radiologists segment subcortical structures, such as the hippocampus and amygdala, to create high-quality labels. These annotations form the ground truth for training models.
FreeSurfer uses these labeled datasets to train its algorithms, which automate the segmentation of subcortical regions in new MRI scans. This involves statistical modeling and machine learning techniques that generalize across diverse brain anatomies.
The training set is continually refined to improve accuracy, ensuring that FreeSurfer provides reliable and reproducible results for neuroimaging research and clinical applications.