Automatic atlas-based volume estimation of human brain regions from MR images.
OBJECTIVE: MRI offers many opportunities for noninvasive in vivo measurement of structure-function relationships in the human brain. Although automated methods are now available for whole-brain measurements, an efficient and valid automatic method for volume estimation of subregions such as the frontal or temporal lobes is still needed. MATERIALS AND METHODS: We adapted the Talairach atlas to the study of brain subregions. We supplemented the atlas with additional boxes to include the cerebellum. We assigned all the boxes to 1 of 12 regions of interest (ROIs) (frontal, parietal, temporal, and occipital lobes, cerebellum, and subcortical regions on right and left sides of the brain). Using T1-weighted MR scans collected with an SPGR sequence (slice thickness = 1.5 mm), we manually traced these ROIs and produced volume estimates. We then transformed the scans into Talairach space and compared the volumes produced by the two methods ("traced" versus "automatic"). The traced measurements were considered to be the "gold standard" against which the automatic measurements were compared. RESULTS: The automatic method was found to produce measurements that were nearly identical to the traced method. We compared absolute measurements of volume produced by the two methods, as well as the sensitivity and specificity of the automatic method. We also compared the measurements of cerebral blood flow obtained through [15O]H2O PET studies in a sample of nine subjects. Absolute measurements of volume produced by the two methods were very similar, and the sensitivity and specificity of the automatic method were found to be high for all regions. The flow values were also found to be very similar by both methods. CONCLUSION: The automatic atlas-based method for measuring the volume of brain subregions produces results that are similar to manual techniques. The method is rapid, efficient, unbiased, and not subject to the problems of rater drift or potentially poor interrater reliability that plague manual methods. Consequently, this method may be very useful for the study of structure-function relationships in the human brain.
SciCrunch is a data sharing and display platform. Anyone can create a custom portal where they can select searchable subsets of hundreds of data sources, brand their web pages and create their community. SciCrunch will push data updates automatically to all portals on a weekly basis. User communities can also add their own data to scicrunch, however this is not currently a free service.