Comparison of Two Methods for Automatic Brain Morphometry Analysis

Abstract

The methods of computational neuroanatomy are widely used; the data on their individual strengths and limitations from direct comparisons are, however, scarce. The aim of the present study was direct comparison of DBM based on high-resolution spatial transforms with widely used VBM analysis based on segmented high-resolution images. We performed DBM and VBM analyses on simulated volume changes in a set of 20 3-D MR images, compared to 30 MR images, where only random spatial transforms were introduced. The ability of the two methods to detect regions with the simulated volume changes was determined using overlay index together with the ground truth regions of the simulations; the precision of the detection in space was determined using the distance measures between the centers of detected and simulated regions. DBM was able to detect all the regions with simulated local volume changes with high spatial precision. On the other hand, VBM detected only changes in vicinity of the largest simulated change, with a poor overlap of the detected changes and the ground truth. Taken together we suggest that the analysis of high-resolution deformation fields is more convenient, sensitive, and precise than voxel-wise analysis of tissue-segmented images.