Connectivity-based segmentation of the thalamic motor region for deep brain stimulation in essential tremor: A comparison of deterministic and probabilistic tractography

Abstract

Objective: Deep brain stimulation (DBS) studies have shown that stimulation of the motor segment of the thalamus based on probabilistic tractography is predictive of improvement in essential tremor (ET). However, probabilistic methods are computationally demanding, requiring the need for alternative tractography methods for use in the clinical setting. The purpose of this study was to compare probabilistic vs deterministic tractography methods for connectivity-based targeting in patients with ET.

Methods: Probabilistic and deterministic tractography methods were retrospectively applied to diffusion-weighted data sets in 36 patients with refractory ET. The thalamus and precentral gyrus were selected as regions of interest and fiber tracking was performed between these regions to produce connectivity-based thalamic segmentations, per prior methods. The resultant deterministic target maps were compared with those of thresholded probabilistic maps. The center of gravity (CG) of each connectivity map was determined and the differences in spatial distribution between the tractography methods were characterized. Furthermore, the intersection between the connectivity maps and CGs with the therapeutic volume of tissue activated (VTA) was calculated. A mixed linear model was then used to assess clinical improvement in tremor with volume of overlap.

Results: Both tractography methods delineated the region of the thalamus with connectivity to the precentral gyrus to be within the posterolateral aspect of the thalamus. The average CG of deterministic maps was more medial-posterior in both the left (3.7 ± 1.3 mm³) and the right (3.5 ± 2.2 mm³) hemispheres when compared to 30 %-thresholded probabilistic maps. Mixed linear model showed that the volume of overlap between CGs of deterministic and probabilistic targeting maps and therapeutic VTAs were significant predictors of clinical improvement.

Conclusions: Deterministic tractography can reconstruct DBS thalamic target maps in approximately 5 min comparable to those produced by probabilistic methods that require > 12 h to generate. Despite differences in CG between the methods, both deterministic-based and probabilistic targeting were predictive of clinical improvement in ET.

Keywords: Deep brain stimulation; Deterministic; Diffusion tractography; Essential tremor; Probabilistic; Thalamus.

Fig. 1

Tractography Methods. Three-Dimensional tractography results from a single patient in T1-weighted space demonstrating streamlines (green) between the thalamus (red) and precentral gyrus (blue) using the probabilistic (A) and deterministic methods (B). Three-dimensional reconstructions of probabilistic and deterministic fiber tracts were created using MRtrix (https://www.mrtrix.org) and Brainlab, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 2

Qualitative comparison between deterministic and probabilistic thalamic segmentation maps. Coronal cross-sections of probabilistic (red) and deterministic (blue) thalamocortical segmentation maps and their intersection (yellow) in T1-weighted space for 36 patients with essential tremor. Both methods delineated the motor region in the posterolateral aspect of the thalamus (green). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 3

The average center of gravity for deterministic and probabilistic thalamic segmentation maps. A variability between 1 and 3 mm was observed between the deterministic maps (D) and variability of 1 mm was found between the probabilistic maps at 30 % (P30), 40 % (P40) and 50 % (P50) threshold. The average center of gravity of the deterministic maps was found bilaterally more medial posterior compared to the probabilistic maps. (Blue color corresponds to the average center of gravity of D maps. Red color corresponds to average center of gravity of P30 and P40 maps and green to P50 maps). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 4

Prediction of clinical improvement. Empirical ROC curves to examine the sensitivity and specificity of the volume of overlaps that were found significant predictors of the improvement in CRST scores. Both methods could reliably predict patients that had superior clinical improvement postoperatively. The AUC was 0.75 (CI:0.56–0.94) for the volume of overlap between the VTA and the deterministically determined CG maps (green) and 0.77 (CI: 0.60–0.94) for the volume overlap between the VTA and the 30% probabilistic CG maps (red). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)