Exploring Local Diffusion MRI Properties for Vim Localization: Evaluation in Clinical Cases

The lack of possibility to directly visualize the Ventro-intermediate thalamic nucleus (Vim) on current routine MRI lead to application of indirect targeting methods in stereotactic neurosurgical procedures, such as Gamma Knife Surgery (GKS) and Deep Brain Stimulation (DBS), for treatment of drug-resistant tremor.

Recently, our group proposed an automated robust and reproducible method across healthy subjects and tremor patients for parceling the thalamus in seven main groups of nuclei while exploring the local diffusion information from the spherical harmonics representation of the orientation distribution functions (ODFs) in k-means clustering framework. One among the resulting parcel is the Ventral-Lateral-Ventral group (VLV) enclosing all motor-related nuclei including the Vim. We aim at further automatic subdivision the VLV cluster. To this end, within the VLV, we first build a k-nn graph with edges corresponding to the respective ODFs distances and then, we perform a partition in 3 sub-clusters using the Normalized Cut algorithm.

The outcome of the proposed subdivision was compared with the radiological response in the follow-up images of 17 patients treated for tremor with GKS unilaterally. The MRI protocol included pre-operative diffusion-weighted images with 64 (or 72) gradient direction, b = 1000s/mm2 and voxel size of 2.2x2.2x2.2mm3, pre-operative and post-operative MPRAGEs with isotropic resolution of ∼1mm3, all of them acquired at field of 3T. In general, the sub-partition followed a spatial-distribution pattern and for 12 out of 17 cases we observed that one specific sub-cluster encloses entirely or the major part of the contrast enhancement corresponding to the GKS target that appeared on the follow-up images.

The evaluation of the sub-partition outcome was further extended in one additional tremor patient treated with DBS, bilaterally. Similar pre-operative MRI data as for the GKS-treated patients was acquired. We observed that in each hemisphere both initial targeting point and the final position of the electrode are inside the anticipated sub-cluster.

This study shows that the diffusion-MRI-based sub-clustering of the VLV thalamic nuclei could potentially allow, in an automated manner, to narrow the area of Vim’s localization. Our preliminary results will be further investigated in a larger cohort of patients (treated by either GKS or DBS) and on diffusion images with higher spatial resolution.