Segmentation of thalamic nuclei using a modified k-means clustering algorithm and high-resolution quantitative magnetic resonance imaging at 1.5 T.

Patient outcome in minimally invasive stereotactic neurosurgical procedures depends on the ability to accurately locate the desired functional region within the deep brain while avoiding the surrounding anatomy. Due to the lack of sufficient contrast within this region in pre-operatively acquired MR images, electrophysiological exploration and histological atlases are currently required to define the surgical target within the thalamus in the treatment of many motor-control disorders. In this paper we introduce a method for segmenting the individual thalamic nuclei based on high-resolution quantitative magnetic resonance images, providing improved target visualization. The method was tested using whole-brain T1 and T2 data acquired from four healthy individuals. Accuracy of the segmentation results was assessed by comparing the center-of-mass coordinates of the segmented nuclei, with coordinates obtained from a classic histological atlas registered to these images. Strong agreement was found, with an average Euclidean distance difference of less than 4.5 mm averaged across all nuclei and all individuals. Reproducibility of the method, determined by calculating the percent similarity of segmentation results derived from data acquired from repeated scan sessions, was greater than 85%. These results illustrate the ability to accurately and reliably segment the primary nuclei of the thalamus and suggest that the method may have utility in the study of individual nuclear regions in disease state as well as for planning deep-brain surgical procedures.