Magnetic resonance images were acquired with a Philips NT scanner (Philips Medical Systems, Best, the Netherlands) operating at 1.5 T. A T1-weighted 3-dimensional fast field echo (FFE) scan (echo time [TE], 4.6 milliseconds; repetition time [TR], 30 milliseconds; flip angle, 30°; and field of view, 256 mm) with 160 to 180 contiguous coronal 1.2-mm slices and a T2-weighted dual-echo turbo spin-echo scan (TE1, 14 milliseconds; TE2, 80 milliseconds; TR, 6350 milliseconds; flip angle, 90°; and field of view, 256 mm) with 120 contiguous coronal 1.6-mm slices of the whole head were used for the quantitative measurements. In addition, a T2-weighted dual-echo turbo spin-echo scan (TE1, 9 milliseconds; TE2, 100 milliseconds; TR, 2200 milliseconds; flip angle, 90°; and field of view, 250 mm) with 17 axial 5-mm slices and a 1.2-mm gap of the whole head was acquired for clinical neurodiagnostic evaluation. Processing was done on the neuroimaging computer network of the Department of Psychiatry, University Medical Center Utrecht, which includes Hewlett Packard(Palo Alto, Calif) Unix 9000 workstations, a server, and Pentium III–equipped personal computers. Prior to quantitative assessments, 10 images were randomly chosen and cloned for interrater reliability purposes determined by the intraclass correlation coefficient. All images were coded to ensure blindness for subject identification and diagnosis; scans were put into Talairach frame (no scaling)49 and corrected for inhomogeneities in the magnetic field.50 Intracranial, total brain, cerebral gray and white matter, lateral ventricle, third ventricle, and cerebellum volumes were then automatically measured with histogram analysis algorithms and a series of mathematical morphological operators to connect all voxels of interest.51- 53 Intracranial volume was segmented on the dual-contrast turbo spin-echo scans (second echo). Using mathematical morphological operators the intracranial volumes were segmented automatically, with the foramen magnum being used as the inferior boundary. The dual-contrast turbo spin-echo scans and the intracranial region segmentations on them were registered to the 3-dimensional FFE images using a mutual information maximization algorithm.51 The 3-dimensional FFE images were masked with the binary intracranial volumes. This was done to remove tissues such as extracranial fat. Total brain volumes were segmented on the T1-weighted scans and contained gray and white matter tissue only. Gray and white matter segmentation was done based on intensity histogram shape analysis of the 3-dimensional FFE images and had been calibrated and validated on a set of 80 human brain images.52 In lateral ventricle segmentation, automatic decision rules bridged connections that were not detectable and prevented "leaking" into cisterns. The third ventricle was limited by coronal slices that clearly showed the anterior and posterior commissures; the upper boundary was a plane through the plexus choroideus ventriculi tertii in the midsagittal slice perpendicular to this slice. The cerebellum was limited by the tentorium cerebelli and the brainstem. All images were checked after the measurements and corrected manually, if necessary. The interrater reliability of the measurements, determined by the intraclass correlation coefficient and based on 10 brains, was 0.95 and higher.