Unsupervised Segmentation of Extraventricular Brain Regions in PC-MRI Using CSF Flow Dynamics

Cerebrospinal fluid (CSF) plays a crucial role in maintaining health, and its pulsatile movement driven primarily by the cardiac cycle and respiration serves as a valuable physiological signal, with abnormalities increasingly associated with neurological disorders. Phase-contrast magnetic resonance imaging (PC-MRI) offers a non-invasive means of capturing both anatomical structure and flow characteristics. In this study, we propose an unsupervised method to segment extraventricular brain regions using CSF flow dynamics derived from PC-MRI data acquired along the anterior commissure–posterior commissure (AC–PC) line. PC-MRI data were collected from eight healthy volunteers (VENC = 10 cm/s) with 23–30 images per cardiac cycle. Voxel-wise features — extracted from time-, frequency-, and wavelet-domain representations —were clustered using the K-means algorithm. An ablation study confirmed that frequency-domain features are the most discriminative single domain. The resulting segments, validated by expert neuroradiologists, showed strong alignment with anatomical regions- grey matter, white matter, vasculature, and CSF spaces- beyond the ventricles. Additionally, the CSF flow characteristics within the segmented regions were also examined and presented for comparative analysis. By focusing on extraventricular regions, this method enables a more comprehensive analysis of CSF dynamics across the entire brain, offering new insights into how CSF interacts with surrounding tissues. This brain-wide, label-free approach opens the door to earlier detection and improved monitoring of neurological conditions where CSF flow is disrupted and lays the groundwork for future research into the role of glymphatic transport, neuroinflammation, and fluid-tissue interactions in both health and disease.