NemoImage 11, Number 5, 2000, Part 2 of 2 Parts 10 E Ir;[@ METHODS - ACQUISITION Development of a template-based quantitative volumetric method: application to structural MRI data from schizophrenic patients, their siblings and normal controls S. Japee, A. Meyer-Lindenberg, B. Verchinski, P. Kohn, M. Egan, L. Bigelow, J. Callicott, A. Bertolino, V. Mattay, D. Weinberger, K. Berman Clinical Brain Disorders Branch, National Institutes of Health, Bethesda, Maryland, USA Introduction Quantitative volumetric assessment is a popular tool used in the analysis of structural magnetic resonance (MR) images. Current methods of volumetric analysis, however, lack automation and require the manual delineation of regions of interest by the operator. These methods are thus tedious and susceptible to a user’s subjective bias and other operator-related sources of error. These drawbacks become especially critical in population studies involving large sample sizes and subtle inter-group differences. We describe here a highly automated, computationally efficient and objective method to analyze large samples of structural MR images to extract volumetric information. Our approach is based on the application of a-priori templates derived from the Human Brain Project [l] to affine-normalized segmentation maps. As a test, we applied the method to measure ventricular volumes in a large group of schizophrenic patients, their siblings and normal controls, and compared the results with previously published values. Methods Tl-weighted structural MR images of the head were obtained from 54 patients diagnosed with DSM-IIIR schizophrenia, 106 of their first-degree relatives and 41 normal controls as part of an ongoing research protocol. MR images comprising 124 sagittal slices (voxel size 1.5mm x .9375mm x .9375mm) were obtained using a 1.5 Tesla MRI scanner (GE Signa). Each volume was intensity-normalized and then passed through a skull-stripping routine (using the software package Freesurfer [2]) to retain only brain regions for further analysis. The stripped intensity-normalized volumes were then segmented into gray matter, white matter and cerebrospinal fluid (CSF) maps based on an intensity-driven Bayesian approach within SPM99 [3]. Extracerebral tissue was then removed from the resulting maps and visual quality control was performed by human operators. The segmented maps were normalized to a standard Talairach template using SPM99 by applying a 12-parameter affine transform. In order to determine the ventricular volumes, over-inclusive masks of the lateral and third ventricles were created from the probabilistic brain atlas resource [ 11. These masks were applied to the normalized CSF maps to exclude extra-ventricular CSF and to count the number of voxels enclosed within the corresponding ventricular regions. The resulting volumes were divided by the determinant of the affine transformation matrix of the mapping used (which gives the volume change resulting from normalization) to obtain the corresponding volumes in the original MR images. Results An analysis of variance of ventricular volumes between schizophrenic patients, their siblings and normal controls (see Figure I I. showed that the three groups differed significantly in the volumes of the left and right lateral and third ventricles. A post-hoc analysis revealed that patients with schizophrenia exhibited highly significant left and right lateral and third ventricular enlarge- ments when compared to normal controls (p < 0.00001). The same finding was evident in the comparison of ventricular volumes between schizophrenic patients and their siblings (p < 0.00005). In all cases, the values for siblings were intermediate between those of schizophrenics and controls, and when compared with the latter, siblings tended to have larger right lateral and third ventricles (p < 0.05). A complimentary analysis performed on groups of schizophrenics, their siblings and normal controls matched for gender, handedness and age distribution, showed similar results for paired comparisons of patients with normal controls and siblings. Siblings of schizophrenic patients however, differed significantly from normals only in the size of their third ventricle (p < 0.06). Discussion We describe an automated 2nd compuratir~nally rf!itiznt method to oht-1!. a’ 1 \ drmelric mea~uremr‘nt~ from y!ruimr:li MR ;rn;i~:~~ When applied to ventricular morphometry in a large sample populatiou stud), this method pro\ i&a ~r>ult\ that are in a<c,lrdanci with previous findings of ventricular enlargements associated with schizophrenic patients [4] and their non-qchi7ophrenic sibling< (51. By creating masks of other anatomical regions of the brain (using the probabilistic brain atlas resource 111). thi\ method can be further applied to volumetric analyses of other cortical and sub-cortical regions. that are known to play an important role 111 schizophrenia. References 1. L. Collins, C. Holmes, T. M. Peters, A. C. Evan, Human Brain Mapping, 3, 190.208 (1996). 2. A. M. Dale, B. Fischl and M. I. Sereno, Neurolmage, 9, 179-194 (1999). 3. J. Ashburner, K. Friston, Neurolmage, 6, 209-217, (1997). 4. D. R. Weinberger, E. F. Torrey, A. N. Neophytides, R. J. Wyatt, Arch Gen Psychiatry, 36, 735-739 (1979). 5. D. R. Weinberger, L. E. DeLisi, A. N. Neophytides, R. J. Wyatt, Psychiatry Res . 4, 65-71 (1981). s545