Single-subject functional parcellation of the human brainstem

Functional MRI data are being used to parcellate the human brain and brainstem into clusters of related functional connectivity. Prior work has computed parcellations strictly at the multi-subject level. Pooling data across subjects in this way requires, however, a very strong prior hypothesis that all humans have very similar patterns of functional connectivity. We suggest that, especially in cases of pathologies, this assumption may not be justified. To address the problem that inter-individual variations will preclude analysis of grouped data, we are developing a single-subject parcellation model by 1) obtaining high SNR and BOLD sensitivity with MRI at ultra-high-field (7 T), and, 2) development of a novel and sensitive parcellation algorithm that incorporates spatial priors and nonlinear manifold learning. Ten minutes of eyes-closed resting state data were gathered from 2 subjects in a Siemens MAGNETOM 7 T scanner using a multiband EPI pulse sequence (TE = 16 ms, TR = 750 ms, 1.5 mm isotropic). The time series of each voxel in the medulla and pons was correlated to the time series of each voxel in the cortex. Our new manifold learning algorithm was then used to cluster together voxels with similar patterns of functional connectivity. Two preliminary observations are available: 1) The entire cortex is functionally connected to some, but not all, regions of the brainstem. 2) Distinct clusters in the ventral and caudal ventrolateral medulla, and posterior-superior medulla, displayed connectivity to known cortical autonomic regions in the medial prefrontal cortex, insula and anterior cingulate. The parcellations were similar but not identical between subjects.