[13] It improves anatomical correspondence and decreases ambiguity, which often occurs in intensity-based registration. In addition, HAMMER has been demonstrated to be only able to capitalize on the excellent anatomical resolution of a single-subject reference template, compared with the other three normalization methods used in SPM software packages, which used DARTEL, cosine basis functions and unified segmentation.[16] Given these merits, HAMMER was selected as the registration method in the following DBM analysis. The procedure was performed using HAMMER software (http://www.rad.upenn.edu/sbia/rsoftware.html). Considering that HAMMER is a

feature-based registration method, the template should be a single SC subject rather than an average atlas template because it has sharper and easily distinguishable features. Before the registration, an appropriate template image was selected by an experienced neurosurgeon. Compared with the study of Leporé and colleagues, the images of all subjects were not first aligned with the ICBM-53 brain template. Instead, they were normalized to the selected SC subject by affine transformation with 12 parameters, which reduced

their variability in orientation, position, and general size. After affine transformation, the images were warped to the template image using a multiresolution strategy. Deformation fields were generated, which described the transformation map from the template to the subject. In principle, the Jacobian matrices of the deformations are more reliable for indicating local brain shape and for reflecting the shape variations between the two groups.[17] The Jacobian matrix contains information not only on local stretching but also on shearing and rotation. The determinant of the Jacobian matrix represents the pointwise volume change induced by the transformation. That is, if the infinitesimal area

A around point in the template is mapped to an area B around point in the subject, the Oxymatrine Jacobian value quantifies the local expansion or contraction resulting from this mapping. Values above 1 indicate tissue expansion, values below 1 indicate tissue contraction, negative values indicate folding, and infinite values indicate tearing. To improve the delineation of shape CAL 101 difference patterns, the images of the Jacobian matrix were smoothened with a Gaussian filter of full-width half-maximum equal to 12 mm. The third step was statistical analysis. Given that the Jacobian value for each subject was derived with respect to the same template, all data can be compared in a voxel-by-voxel manner regardless of their shapes. A means test between the EB and SC groups was performed under the null hypothesis that the means are equal, with the assumption that the variances of the Jacobian value within the two groups are equal.