Automatic segmentation of thalamic nuclei from diffusion tensor magnetic resonance imaging Mette R. Wiegell,a,b,* David S. Tuch,a,c Henrik B.W. Larsson,b and Van J. Wedeena a Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, 149 13th Street, Room 2301, Charlestown, MA 02129, USA b Danish Research Center for Magnetic Resonance, Hvidovre Hospital, afd. 340, Ketteg��rds alle 30, 2650 Hvidovre, Denmark c Harvard���MIT Division of Health Sciences and Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA Received 21 February 2002 revised 2 December 2002 accepted 10 December 2002 Abstract The nuclei of the thalamus have traditionally been delineated by their distinct cyto/myeloarchitectural appearance on histology. Here, we show that diffusion tensor magnetic resonance imaging (DTI) can noninvasively resolve the major thalamic nuclei based on the characteristic fiber orientation of the corticothalamic/thalamocortical striations within each nucleus. Using an automatic clustering algorithm, we extracted the Talairach coordinates for the individual thalamic nuclei. The center-of-mass coordinates for the segmented nuclei were found to agree strongly with those obtained from a histological atlas. The ability to resolve thalamic nuclei with DTI will allow for morphometric analysis of specific nuclei and improved anatomical localization of functional activation in the thalamus. �� 2003 Elsevier Science (USA). All rights reserved. Introduction As the central relay station for the brain, the thalamus mediates communication among sensory, motor, and asso- ciative brain regions. The multiple functional pathways which relay through the thalamus form the thalamic cyto- architecture. The thalamic cytoarchitecture is divided into functionally specific clusters referred to as nuclei. The thalamic nuclei have traditionally been delineated by their distinct cyto- and myeloarchitectural appearance on histology (Smeets et al., 1999 Morel et al., 1997 Scannell et al., 1999 Van Buren and Borke, 1972). The number of thalamic nuclei reported with histological methods varies with the method employed, although most cyto/myeloarchi- tecture stains identify 14 major nuclei, with several subdi- visions of the individual nuclei, some established by addi- tional chemoarchitectural stains. Thalamic changes have been implicated in a large num- ber of diseases, including schizophrenia (Portas et al., 1998 Staal et al., 1998 Buchsbaum et al., 1999), Parkinson���s disease (Giroux et al., 1998 de la Monte et al., 1989 McNeill et al., 1988 Samra et al., 1971 Xuereb et al., 1991), chronic pain syndrome (Davis et al., 1998), multiple sclerosis (Combarros et al., 1994), and wallerian degenera- tion (Ogawa et al., 1997). Parkinson���s disease, multiple sclerosis, and chronic pain syndrome can also be treated by surgical ablation or electric stimulation (Tornqvist, 2001) of the involved nucleus. Presurgical planning of these cases often uses generic thalamic atlases to target the pertinent nucleus (Hardy et al., 1992 Nowinski, 1998 Otsuki et al., 1994 Tasker et al., 1991). Given the large degree of inter- subject variability (Van Buren and Borke, 1972) in the location and size of the thalamic nuclei, such generic atlases may be highly inaccurate. Functional studies (fMRI, PET, SPECT) have also documented disease-related changes in functional activation of the thalamus (Blinkenberg et al., 2000 Heckers et al., 2000 Rauch et al., 2001 Rubia et al., 2001 Volz et al., 1999 Vuilleumier et al., 2001). However, due to the lack of a precise anatomical reference, these studies are generally not able to localize the activation to a specific nucleus within the thalamus. The ability to resolve thalamic nuclei by noninvasive imaging would enable quantitative morphometric analysis of thalamic changes in the above-mentioned diseases, pro- vide more accurate neurosurgical planning, and offer im- * Corresponding author. Fax: 1-617-726-7422. E-mail address: mette@nmr.mgh.harvard.edu (M.R. Wiegell). NeuroImage 19 (2003) 391��� 401 www.elsevier.com/locate/ynimg 1053-8119/03/$ ��� see front matter �� 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S1053-8119(03)00044-2

Fig. 1. Mid-thalamic diffusion tensor images of four subjects (a, b, c, d). Images a��� c are axial slices and image d is a coronal slice at approximately mid-thalamic level. The region-of-interest (yellow box) for each subject is shown in the fractional anisotropy image (Pierpaoli and Basser, 1996) in the top left corner of the diffusion tensor image. In the diffusion tensor images, the cylinders depict the diffusion tensor within each voxel. The axes of each cylinder are oriented in the direction of the principal eigenvector of the local diffusion tensor. The length of the axes is scaled by the product of the corresponding eigenvalue and the square-root of the fractional anisotropy metric (Pierpaoli and Basser, 1996). The cylinders are colored by the direction of the principal eigenvector according to the red��� green��� blue sphere shown at bottom right with red indicating mediolateral, green anteroposterior, and blue superoinferior direction. The background slice is colored by the direction of the third eigenvector, an indicator of the sheet-normal direction (Wiegell et al., 2000a). Note how the clusters if mean fiber direction do not completely coincide with the clusters of mean sheet-normal direction. Fig. 2. Histological comparison. (a) Diffusion tensor image (Fig. 1a) compared with a histological slice (b) from Van Buren and Borke (Van Buren and Borke, 1972) at a similar anatomical level. The region-of-interest (yellow box) is shown in the fractional anisotropy image (Pierpaoli and Basser, 1996) in the top left corner of the diffusion tensor image. Note the correspondence between diffusion orientation clusters and histologically defined nuclei borders. 392 M.R. Wiegell et al. / NeuroImage 19 (2003) 391��� 401