Gyral parcellation of cortical surfaces via coupled flow field tracking

This paper presents a novel method for parcellation of the cortical surface of human brain into gyral based regions via coupled flow field tracking. The proposed method consists of two major steps. First, the cortical surface is automatically parcellated into sulcal based regions using several procedures: estimating principal curvatures and principal directions; applying the hidden Markov random field and the Expectation-Maximization (HMRF-EM) framework for sulcal region segmentation based on the maximum principal curvature; diffusing the maximum principal direction field in order to propagate reliable and informative principal directions at gyral crests and sulcal bottoms to other flat cortical regions with noisy principal directions by minimization of an energy function; tracking the flow field towards sulcal bottoms to parcellate the cortical surfaces into sulcal basins. The sulcal parcellation provides a very good initialization for the following steps of gyral parcellation on cortical surfaces. Second, based on the sulcal parcellation results, the cortical surface is further parcellated into gyral based regions using the following procedures: extracting gyral crest segments; dilating gyral crest segments; inverting the principal direction flow field and tracking the flow field towards gyral crests in order to partition the cortical surface into a collection of gyral patches; merging gyral patches to obtain gyral parcellation of the cortical surface. The proposed algorithm pipeline is applied to nine randomly selected cortical surfaces of normal brains and promising results are obtained. The accuracy of the semi-automatic gyral parcellation is comparable to that labeled manually by experts.