Reliability of brain volumes from multicenter MRI acquisition: a calibration study

Multicenter studies can provide additional information over single center studies because of their increased statistical power. Because similar acquisition protocols are being used internationally for structural magnetic resonance imaging (MRI) studies of the human brain, volumetric MRI data studies seem suitable for this purpose. Possible systematic differences between sites should be avoided, however, particularly when subtle differences in tissue volume are being searched for, such as in neuropsychiatric diseases. In this calibration study, the brains of six healthy volunteers were (re)scanned with MR scanners from four different manufacturers at five different sites, using the local acquisition protocols. The images were segmented at a central reference site. The intraclass correlation coefficient (ICC) was determined for the whole brain, gray and white matter, cerebellum, and lateral and third ventricle volumes. When required, the processing algorithms were calibrated for each site. Calibration of the histogram analysis was needed for segmentation of total brain volume at one site and for gray and white matter volume at all sites. No (additional) calibration was needed for cerebellum and ventricle volumes. The ICCs were > or = 0.96 for total brain, > or = 0.92 for cerebellum, > or = 0.96 for lateral ventricle, > or = 0.21 for third ventricle, > or = 0.84 for gray matter, and > or = 0.78 for white matter volume. Calibration of segmentation procedures allows morphologic MRI data acquired at different research sites to be combined reliably in multicenter studies.