Validation of semiautomated methods for quantifying cingulate cortical metrics in schizophrenia

This paper validates semiautomated methods for reconstructing cortical surfaces of the cingulate gyrus from high-resolution magnetic resonance (MR) images. Bayesian segmentation was used to delineate the image voxels into five tissue types: cerebrospinal fluid (CSF), gray matter (GM), white matter (WM), and partial volumes of CSF/GM and GM/WM; the tissues were then recalibrated as CSF, GM, and WM via the Neyman-Pearson Likelihood Ratio Test. To generate cortical surfaces at the interface of GM and WM, the thresholds between the tissue types were first used to reassign partial volume voxels to CSF, GM, and WM with minimum error (that varied from 0.06 to 0.15 for the 10 subjects). Next, topology-correct cortical surfaces were generated and validated with almost all surface vertices lying within one voxel (0.5 mm) of hand contours. Dynamic programming was used to delineate and extract the cingulate gyrus from the cortical surfaces based on its gyral and sulcal boundaries. The intraclass correlation coefficient for surface area obtained by two raters for all 10 surfaces was 0.82. In addition, by repeating the entire procedure three times in one subject, we obtained a coefficient of variation of 0.0438 for surface area.