Abstract
To develop a method for segmenting cerebrospinal fluid (CSF) regions with complex, inhomogeneous pulsatile patterns in phase contrast magnetic resonance imaging (PC-MRI) sequences. Our approach used various temporal features of flow behavior as input attributes in an unsupervised k-means classification algorithm. CSF flow parameters for the cervical subarachnoid spaces and the pontine cistern were calculated in 26 healthy volunteers. Background and aliasing corrections were applied automatically. The algorithm's reproducibility was determined by calculating two parameters (area and stroke volume) while varying the initially selected seed point. The influence of background correction on these parameters was also assessed. The method was highly reproducible, with coefficients of variation of 3 and 4% for the cervical stroke volume and area, respectively. In an analysis of variance, background correction did not have a statistically significant effect on either the stroke volume (p = 0.32) or the CSF net mean flow (p = 0.69) at the C2C3 level. The method presented here enables rapid, reproducible, quantitative analysis of CSF flow in complex regions such as the C2C3 subarachnoid spaces and the pontine cistern. © 2011 Wiley Periodicals, Inc.
