Reproducibility of nonparametric feature map segmentation for determination of normal human intracranial volumes with MR imaging data

Semiautomated segmentation of dual‐contrast magnetic resonance images was used to determine volumes of total brain, gray matter, white matter, and cerebrospinal fluid (CSF) in healthy volunteers. Reproducibility of the technique was evaluated in terms of intraobserver, interobserver, and study‐to‐study variations. Intraobserver coefficients of variation ranged from 0.4% to 6.0%, while interobserver values ranged from 0.8% to 9.9%. In both cases, the maximum variations were obtained in volume measurements of tissues with maximum complexity (ie, CSF), and the minimum variation was obtained in determining total brain volume. This was also true in the case of study‐to‐study variations in volume measurements, for which the coefficients of variation ranged from 0.5% to 8.7%. The use of appropriate preprocessing techniques, which are crucial to the accuracy and reproducibility of the segmentation technique, are described in detail.