A Novel Automatic Segmentation Method for ROI-based Functional Analysis

We present an automatic segmentation method for the extraction of first Heschl’s gyrus (HG), the morphological landmark for human primary auditory cortex. Extracted HG regions can be used to test functional hypotheses about this area using a region-of-interest (ROI)based approach. The proposed technique consists of a coarse segmentation phase using a statistical deformation-based atlas, followed by a finer segmentation using a Laplacian level set. Eighteen subjects participated in an auditory fMRI study, with structural MR images also acquired. In each subject’s structural MRI volume, the first HG was manually identified and labeled by four independent observers. The performance of the segmentation procedure was assessed by calculating the overlap between the automatically extracted and the manually labeled HG regions. The overlaps were more than 83% in both hemispheres. In spite of high variability among subjects, the ROI-based functional analyses yielded similar results for both the automatic and manually segmented HG, across a variety of auditory and speech-related functional contrasts.