Nine healthy individuals, with no history of neurological or psychiatric diseases were included in the study (mean age: 28.3±7.09, male: 3). Furthermore, two patients were also included, who underwent stereotactic thalamotomy.

The study was approved by the Ethics Committee of University of Szeged (authority number: 87/2009), and all subjects provided written consent.

Imaging was carried out using a 1.5 T GE Signa Excite scanner. High resolution T1 weighted images (3D IR-FSPGR: TR/TE/TI: 10.3/4.2/450ms, flip angle: 15°, ASSET: 2, FOV: 25*25 cm, matrix: 256*256, slice thickness: 1mm) and diffusion-weighted images (DTI: TR/TE: 16000/93.8ms, flip angle: 90°, FOV: 23*23cm, matrix: 96*96, slice thickness: 2.4mm, ASSET: 2, NEX: 2) were acquired. Diffusion weighting was performed along 60 independent directions, with a b-value of 1000 s/mm². Six reference images, with no diffusion weighting were also obtained. The preoperative scans of the patients were obtained using identical acquisition sequences to those used for control subjects. Postoperative scans were carried out three months after the surgery. High-resolution-T1 weighted images (with parameters identical to the preoperative ones) and sagittal 3D FLAIR images (3D FLAIR: TR/TE/TI: 6000/134.6/1839ms, flip angle: 90°, FOV: 23*23cm, matrix: 256*256, slice thickness: 2mm) were acquired to localise the thalamotomic lesion.

Data were processed using the tools from the FMRIB Software Library (FSL, version 5.0; Oxford Centre for Functional MRI of the Brain (FMRIB), UK; www.fmrib.ox.ac.uk/fsl^[17]) according to the method reported by Behrens and colleagues ^[14]. Initially the raw diffusion data was corrected for eddy-currents and motion artefacts with FMRIB's Diffusion Toolbox (FDT ^[18]). The images were then skull stripped using the Brain Extraction Tool (BET ^[19]) and the diffusion-weighted images were registered to the high-resolution T1-weighted image with a 6 degree-of-freedom linear registration using the FMRIB Linear Image Registration Tool (FLIRT ^[20]). Probability distributions of fiber orientation were estimated for each brain voxel in the acquired diffusion space using a multi-fiber extension ^[21] of the probabilistic tractography available in FDT ^[18].

Probabilistic diffusion tractography can estimate the pathways that originate at any given seed voxel, as well as the probability that such a pathway will pass through any other voxel in the brain ^[18]. Binary masks of the thalamus and the cortical targets were drawn manually for each subject. Probabilistic multi-fiber diffusion tractography was initiated from every voxel inside the thalamic mask: 5000 times for each voxel. Counters were increased every time an individual streamline reached the cortical target region. Hence, the values stored in thalamic voxels in one of the resulting images represent the probability of those voxels being connected to the particular target cortical mask assigned to that image ^[14].

To investigate the inter-subject variability of the target nuclei, a specific distance reserving registration method was used: the high-resolution T1-weighted images were registered to the standard MNI brain with a 6 degree-of-freedom transformation (rotations and translations only) that was based on landmark points in a way that kept the position of the anterior commissure fixed, while aligning the ACPC line and the mid-sagittal plane between images.

The Euclidean distance, given by the equation:

was calculated for each subject between points representing the thalamic voxel with peak connection probability to the premotor cortex (for Vop) and to the motor cortex (for Vim) and the standard space voxel with peak connectivity as indicated by the Oxford Thalamic Connectivity Map ^[13]. The mean pair-wise distance was also calculated: the distance between the voxel with peak connection probability to the premotor (for Vop) and to the motor (for Vim) cortex was calculated between all possible pair of subjects and this inter-subject distance was averaged for Vop and Vim separately.

The connectivity images for the thalamus were thresholded at 10% of individual connectivity maximum and binarised in order to create masks of the thalamic regions connected to the premotor or motor cortex. To assess the similarity of positions the overlap between each pair of masks (between subjects) were calculated according to the method proposed by Crum and colleagues ^[22]. Overlap was measured by the Tanimoto Coefficient (TC), which is defined as the ratio of the number of voxels in the intersection of the two regions to the number of voxels in the union:

We also took further steps to exclude possible registration bias between the DTI and T1-weighted images that might originate from the EPI distortions. We manually compared the high-resolution T1-weighted image and the first volume of the diffusion data (no diffusion gradient applied) registered to the T1-weighted image for each subjects. The positions of the following landmarks were compared: AC, PC, the maximal width of the third ventricle at the level AC and the highest point of the corpus callosum in the midsagittal line.

As found with the manual comparison the shape and the position of the thalamus was not affected by EPI distortions. The largest misregistrations were found along the anterior-posterior axis, but even that was minimal (for AC and for PC 0.6±0.5mm). The width of the third ventricle was not different, and the position of the highest point of the corpus callosum differed only in case of a single subject with 1 mm.

The premotor thalamus was consistently localised in all subjects, however, since the individual brain shapes differed, the exact location of the Vop and Vim nucleus varied substantially across subjects (Fig 1). The mean distance of the peak connection probability of Vop from the peak probability indicated by the Oxford Thalamic Connectivity map was 5.08mm. The equivalent distance was 6.26mm for the Vim. The mean pair-wise inter-subject distance of the peak connectivity voxel of Vop was 7.33±3.37mm (range: 0–14.56mm) and 7.42±3.35 mm (range: 2–14.28mm) for Vim. The mean distance of the tractography defined Vop and Vim coordinates from the Stereotactic target point defined by Hyam's method ^[12] was 7.19±4.36mm (range: 2.45–14.89mm) for Vim and 9.58±4.82mm (range: 3.0–17.12mm) for Vop.

The mean pair-wise overlap for Vop as calculated by the Tannimoto Coefficient was 40.2% (range: 15.5%–66.2%) and 31.8% (range: 3.2%–66.2%) for Vim.

A left-handed, 32 year-old male, after having a motorbike accident that caused brainstem bleeding (at the level of the red nucleus; Fig 2), developed a tremor having resting, postural, as well as intention components, bradykinesia, rigidity and a slight paresis of the left arm. Seven years after the accident, because of ineffective tremor control by various medication regimes, functional neurosurgery was offered. Because the patient has not accepted deep-brain-stimulation implantation stereotactic thalamotomy was carried out on the right side in the region of Vop/Vim border with target coordinates according to Guiot's method ^[5], ^[23] that is, given the AC-PC length of 24.9mm, was 8mm anterior of PC and 12.7mm lateral from the mid-sagittal line on the AC-PC plane. During the standard surgical procedure the final position of the lesion along the selected trajectory was adjusted according to the tremor control of the intraoperative stimulation. After surgery the patient had an immediate relief from his left upper extremity tremor (his score on the part A of the Fahn-Tolosa-Marin tremor rating scale (FMT) reduced form 16 to 4 points). Neither the muscle strength, nor the bradykinesia improved. At the three-month checkup these findings had not changed for the left upper extremity, however a small amplitude flexion-extension tremor appeared in the left toes.

As shown by the connectivity-based segmentation of the thalamus, regions connected with highest probability to the prefrontal, premotor, motor and sensory, parietal, occipital and temporal cortices were clearly delineated (Fig 3). The operative lesion of the thalamus, identified on the postoperative FLAIR images (registered to the preoperative T1-weighted structural image with 6 DOF linear registration) was situated in the region of the thalamus connected with the highest probability to the premotor cortex (Fig 3).

Probabilistic multi-fiber tractography initiated from the thalamic lesion showed, on one hand, connection to the medial premotor - prefrontal cortex and, on the other hand, to the brain stem particularly to the rubral/perirubral mesencephalic posthaemorrhagic lesion and some fibres travelling further on to the cerebellum (Fig 4).

A right-handed, 50-year-old women had developed tremor ten years earlier. The tremor was most prominent in the hands but also appeared in the head, having postural and action components, and was also present when walking. Her family history was unremarkable. Medication regimes that had been tried over the years were ineffective, and the tremor severely affected the quality of life of the patient. Stereotactic Vim thalamotomy was carried out with target coordinates according to Guiot ^[5], ^[23] that correspond well with the coordinates defined by Hyam's method ^[12]. After surgery the tremor of the right hand improved significantly and on the three-month control examination this effect was found to be enduring (FMT score reduced from 17 to 11 points). No significant change was detected in the head-tremor or in the tremor of the left hand.

The postoperative T1-weighted image was registered to the preoperative T1-weighted image in order to identify the position of the lesion in relation to the diffusion–based connectivity results. The connectivity-based segmentation of the thalamus indicated that the surgical lesion was in the putative Vim nucleus, the region that showed the highest probability of connection to the primary motor cortex (Fig 5). Postoperative probabilistic tractography from the lesion showed that the lesioned thalamus was connected to the primary motor cortex, mainly to the medial structures, and downstream to the ipsilateral cerebellar hemisphere (Fig 6).

Atlas-based targeting for stereotactic functional neurosurgery is a frequently considered option for the treatment of movement disorders. While the overall outcome is reported to be positive, the failures and side effects might be related to the unaccounted inter-individual variability of the target localisation. To overcome such limitations we suggest incorporating information about individual anatomy from diffusion imaging, using connectivity-based segmentations of the thalamic nuclei. However, it has to be pointed out that in the two cases we reported in current analysis we did not plan the targeting with tractography based segmentation a priori of the surgery. Extension of this study to compare the result of the stereotaxic neurosurgery with and without tragtography based targeting is crucial. Further improvements might be expected if additional imaging modalities (e.g. relaxometry, fMRI) could also be utilised.