Anatomical brain structures normalization for deep brain stimulation in movement disorders

Abstract

Deep brain stimulation (DBS) therapy requires extensive patient-specific planning prior to implantation to achieve optimal clinical outcomes. Collective analysis of patient's brain images is promising in order to provide more systematic planning assistance. In this paper the design of a normalization pipeline using a group specific multi-modality iterative template creation process is presented. The focus was to compare the performance of a selection of freely available registration tools and select the best combination. The workflow was applied on 19 DBS patients with T1 and WAIR modality images available. Non-linear registrations were computed with ANTS, FNIRT and DRAMMS, using several settings from the literature. Registration accuracy was measured using single-expert labels of thalamic and subthalamic structures and their agreement across the group. The best performance was provided by ANTS using the High Variance settings published elsewhere. Neither FNIRT nor DRAMMS reached the level of performance of ANTS. The resulting normalized definition of anatomical structures were used to propose an atlas of the diencephalon region defining 58 structures using data from 19 patients.

Keywords: Atlas; Deep brain stimulation (DBS); Group analysis; Image registration; Patient normalization; Template; Thalamus.

Fig. 1

Result of the planning for a DBS implantation. A slice of the WAIR anatomical image is presented together with the 3D representation of the deep brain structures manually segmented by the neurosurgeons used to decide on the trajectories for the stereotactic implantation (blue). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2

Top-level workflow description of the group normalization procedure. After preparation of the surgery on the planning station by the neurosurgeon, the data is exported and prepared (A). Data is first aligned on the MNI152 template with linear registration based on the T1 contrast (B). Aligned images are then non-linearly warped with three different tools (C) in order to compare the results achieved in the deep brain region by each tool (D).

Fig. 3

Linear patient registration as a prealignment for the non-linear registrations. The skullstripped T1 image from each patient is registered to the skullstripped MNI152 template using FLIRT with 12 degrees of freedom (dof). The resulting transformation matrix is then applied to the WAIR image and each anatomical structure label. Transformed anatomical images are then averaged for each modality to create a T1 and a WAIR linear atlas, while the labels are summed for each structure to create cumulative definitions of the labels.

Fig. 4

Non-linear atlas creation workflow: A. The template is created in a two-step iterative registration process. In the first step, the T1 images lead the registration (active image set) while the WAIR and label images are passively transformed in the new reference system. After this step, a volume of interest centered on the deep brain region is selected. In the second step, the WAIR images are used as active set and the T1 and label images are passively transformed. B. Each of these two iterative template creations consists in registering the active anatomical images from each patient to a reference. The first reference for the T1-led non-linear normalization is the Linear template (Fig. 3). C. Each registration iteration uses the selected active image to register it to the current reference. D. The resulting warped active images are averaged to create a temporary template. The latter is warped with the inverse average affine transform composed with the inverse average non-linear warp to create the updated template. All the transforms are composed to create a transform for each patient from the original images to the updated template that allows transforming the passive images to the updated template space.

Fig. 5

Presentation of the results of the normalization using ANTS, FNIRT and DRAMMS. Registration accuracy was measured using the Dice coefficient (DC) (A) and Mean Surface Distance (MSD) (B) between the definition of each structure for each patient and the cumulative definition of the structure in the template. Kernel density estimation bandwidth: DC:0.05, MSD:0.1. p-values for the one-way Anova test results between selected iterations are presented above each plot.

Fig. 6

Results of the normalization with ANTS and the High Variance settings after the first WAIR-based registration. Results for each anatomical structure are presented for both the Dice coefficient (A) and the Mean Surface Distance (B).

Fig. 7

Anatomical normalization results. An image from an individual patient is presented for comparison purposes together with the ANTS template with Ewert et al. (2019) – high variance settings and one step of WAIR refinement, the FNIRT template created with the default settings and a single WAIR refinement and the atlas created with DRAMMS using the default settings and one WAIR refinement step. Each template slice is presented with the heatmap of the left subthalamic nucleus overlaid. Slices were taken at the largest cross-section of STN.

Fig. 8

Two WAIR slices together with anatomical structures from a single patient, transformed in the space of the best template created with ANTS are presented in A and C. The corresponding slices in the WAIR template of the group, with the cumulative definition of the structures with a 50% confidence level are presented in B and D. The Fields of Forel (FF), sub-thalamic nucleus (STN), substrancia nigra (SN), antero-lateral nucleus (AL), Mammillary body (MB) and Mammillothalamic tract are marked for anatomical reference.