Introduction

Deep brain stimulation (DBS) is an effective method for treating the most disabling symptoms of certain movement disorders [15, 17, 23, 39, 40]. Such treatments have involved modulation of several different nuclei, including the subthalamic nucleus (STN) and nucleus ventralis intermedius (VIM) of the thalamus [2, 7, 39, 40, 59].

Targets have traditionally been selected through the anterior commissure-posterior commissure (AC-PC) reference system, which is further refined by neurophysiologic microelectrode recording (MER) and macroelectrode stimulation to ensure electrode localization at the optimal site for clinical improvement. In up to 50 % of cases, the radiologically defined first trajectory for the STN is deemed unreliable based on neurophysiologic results and has to be modified, lengthening the procedure’s duration and thus increasing patient’s fatigue (awake surgery), and possibly the risk of morbidity [66]. Also, some authors have found a discrepancy greater than 1.5 mm (corresponding to the radius of the sphere of the effective electric field generated by the stimulator) between the radiologically defined target and the one obtained by the microrecordings [31]. Some brain structures can be easily identified using current brain imaging methods, and it is possible to choose targets directly based on such images [2, 62, 71]. However, the optimal electrode position for DBS is at a functional “sub-structure” within target nuclei (herein referred to as an optimized target (OT)); for example, the precise stimulation site for motor symptoms of Parkinson’s disease (PD) is the motor subdivision of the STN located at its dorsolateral region [34, 40, 50, 53, 64].

Tractography is a technique based on diffusion tensor imaging (DTI), which has been used to non-invasively reconstruct white matter pathways in the brain [8, 45, 68]. DTI deterministic tractography (DTI-DT) estimates the neural connections by designating at least two regions of interest (ROI) in the 3-dimensional (3D) space [68]. The DTI software StealthViz® (Medtronic, MN, USA) is based on the fiber assignment by continuous tracking (FACT) algorithm [37, 45] for white matter tract reconstruction and is widely used for surgical planning in clinical practice. Here, we present a method using DTI-DT, MRI sequences available in clinical practice, and StealthStation® navigation software that allows the localization of different DBS target subdivisions through basal ganglia circuit segmentation to provide an OT, although this method is not intended to replace the intraoperative MER and macrostimulation to further refine the final stimulation site. In this paper, we describe this method and compare the accuracy of the OT obtained by tractography with the atlas-based targets.

Material and methods

Patients

We used the imaging studies of nine patients who were operated for DBS in our center from 2011 to 2014. We selected those patients who had undergone diffusion MR imaging according to the protocol specified subsequently, prior to the DBS system implantation. Clinical and demographic data including the preoperative clinical status, stimulation parameters, postoperative clinical outcome, and follow-up were recorded (Table 1). All patients provided consent for analysis and publication of their data.

Table 1 Clinical and demographic data of the patients, stimulation features, clinical outcome, and follow-up
Full size table

Data acquisition

All MRI studies were performed using a 3 T clinical imager (Signa HDXt GE Healthcare) with an eight-channel head coil. The imaging protocol was the same for all patients. The T2-weighted fast spin-echo sequence was acquired with the following parameters: repetition time (TR), 6000 ms; echo time (TE), 95 ms; field of view (FOV), 220 mm; interpolated matrix, 512 × 512; and slice thickness, 1 mm. The T1-weighted 3-dimensional Fast SPGR IRprep sequence was acquired with the following parameters: TR, 8 ms; TE, 2 ms; flip angle, 12; FOV, 240 mm; matrix, 256 × 256; slices, 160; slice thickness, 1 mm; inversion time (TI), 450 ms; and isotropic voxels, 1 mm. The FLAIR-FSE 3D sequence was acquired with the following parameters: TR, 6600 ms; TE, 110 ms; TI, 2200 ms; FOV, 240 mm; matrix, 256 × 256; slices, 160; and slice thickness, 1 mm. These parameters enabled reconstruction with a 1-mm isotropic voxel size. Diffusion weighting was encoded along 55 independent orientations using a single-shot multi-slice 2D spin-echo diffusion-sensitized and fat-suppressed echo planar imaging (EPI) sequence, with b values of 0–1000 mm2/s, TR/TE of 9600/82 ms, FOV of 250 × 250 mm, matrix of 96 × 96, and slice thickness of 2.6 mm with no inter-slice gap, resulting in isotropic voxels of 2.6 mm. FUNCTOOL software (General Electric HC) was used to improve geometric distortion. Preoperative and postoperative computed tomography (CT) scans were acquired on a multi-slice Philips ® Brilliance 64 with spiral pitch of 0.891, rotation of 0.75 s, no gantry tilt, matrix of 512 × 512, slice thickness of 1 mm, tube voltage of 120 kV, and tube current of 75 mA.

Regions of interest

Several cortical and subcortical structures involved in extrapyramidal circuits were defined as ROIs for DTI generation and were classified as projection areas and subcortical targets. The following were included as projection areas: primary motor cortex (M1, Brodmann area 4); supplementary motor area (SMA, part of Brodmann area 6) [30, 46]; pre-supplementary motor area (pre-SMA, Brodmann area 6 and part of 8) [30, 46, 69]; red nucleus (RN); and dentate nucleus (DN). The subcortical targets included the thalamus (Th) and the subthalamic nucleus (STN) (Fig. 1).

Fig. 1
figure 1

Anatomical boundaries of the regions of interest: projection areas (M1 motor cortex, SMA supplementary motor area, pre-SMA pre-supplementary motor area, DN dentate nucleus, RN red nucleus) and subcortical targets (Th thalamus, STN subthalamic nucleus) segmented manually in FLAIR 3D and T1 sequences. These are sequences commonly used in clinical practice that we found adequate to define the boundaries of the subcortical ROIs. Other sequences are used for this aim, but our approach takes advantage of readily available studies performed in the clinical setting. The Th is difficult to define in MRI due to the poor between-tissue contrast at the thalamic gray-white matter interface. By adjusting the contrast in T1-weighted images, we were able to obtain the lateral limit of the thalamus defined by the posterior limb of the internal capsule. Finally, we used the 3D FLAIR sequence to obtain the maximal resolution of the voxels for structures such as STN and SN. The substantia nigra (SN) was also segmented to obtain the antero-inferior boundary of the STN

Full size image

Delineation of projection areas and subcortical targets (manual segmentation)

This process was performed with a StealthStation® (Medtronic, MN, USA) and the StealthViz® software package (Medtronic), using the “segmentation section” of the software. Projection areas segmentation was performed by manually tracing the cortical segment boundaries based on Brodmann areas. Subcortical target segmentation was performed manually on the MR sequence in which the structure was most clearly defined (Fig. 1). We used several neuroanatomical sources to guide the delineation of neuroanatomical ROIs on MRI images [24, 25, 30, 38, 42, 46, 48, 63, 70].

Deterministic tractography processing and analysis

Tractography was processed using the FACT algorithm [37] with a StealthStation® and the StealthViz® software package. The fractional anisotropy (FA) “start value” and “stop value” were not less than 0.1, and the maximal directional change of fibers was set as 45 to 80°. These values were chosen after a thorough analysis of the fiber tracking and plausibility of the results according to previous reports [4, 6, 7, 9, 11, 19, 20, 26, 32, 38, 41, 42, 70].

Parcellation process with projection areas and subcortical targets

Figure 2 shows a flowchart of the method. Following segmentation of the ROIs, the subcortical targets were exported and binarized as “3D objects”. Then, the subcortical targets, processed as 3D objects, were imported back and coregistered into the working session as independent MRI series. In the “view section” of the software, we used the suggested range of DTI values and generated tracts from the projection areas under a neuroanatomical appraisal. The generated tracts were subjected to multiplanar reconstruction (MPR) and then used in the segmentation section in which we selected the subcortical target that we wanted to parcellate, which had been previously imported as a MRI series. With the subcortical target in the background and with the “selection tool”, we selected the voxels located in the intersection between the outlined tract and the subcortical target, and thus, we were able to generate this region as an independent 3D object (Figs. 3 and 4).

Fig. 2
figure 2

Flowchart of the method used for localization of deep brain stimulation targets. DN dentate nucleus, DTI diffusion tensor image, DWI diffusion-weighted image, FACT fiber assignment by continuous tracking, M1 primary motor cortex, pre-SMA pre-supplementary motor area, RN red nucleus, SMA supplementary motor area, STN subthalamic nucleus, Th thalamus

Full size image
Fig. 3
figure 3

Identification of the optimized target of the thalamus (Th) for patients with essential tremor. a The thalamus (white region) was fused with the structural magnetic resonance image (MRI). b The tracts were generated through projections from the dentate nucleus (DN) (green) and red nucleus (RN) (red); the dentatorubrothalamic tract is also identified (arrow). c The “3D lines” (fibers in green) obtained were superimposed with the fusion. d Coronal view, multiplanar reconstruction of the tract (green) in the segmentation section of the software. In this way, we were able to choose the voxels in the intersection between the thalamus and the reconstructed tract (red voxels). e Axial view of d. f Patient 8. The 3D renderization of the thalamus and the region obtained by our method is shown in green (arrow). The projections from this area replicate the known connections of the VIM nucleus (RN, DN, primary motor cortex, and supplementary motor area) [3, 11, 20, 33, 36]

Full size image
Fig. 4
figure 4

Identification of the optimized target of the subthalamic nucleus (STN) for patients with Parkinson’s disease. a Coronal view of the 3D tracking from the M1 and SMA-pre-SMA (red) showing the pyramidal tract (green). b The 3D lines (fibers in green) obtained were superimposed with the fusion of the segmented STN (white) and the structural MRI in axial view. c Axial view, multiplanar reconstruction of the tract (green) in the segmentation section of the software. In this way, we were able to choose the voxels in the intersection between the STN and the reconstructed tract (red voxels). d Patient 6. The 3D renderization of the STN and the region obtained by our method (green). The projections from this area replicate the corticosubthalamic connections, mainly with M1, and also with SMA and pre-SMA [16, 41]

Full size image

Frameless stereotactic surgery

One day before surgery, seven fiducial markers of the frameless stereotactic system (Nexframe, Medtronic) were fixed onto the patient’s skull under local anesthesia. Then, a CT scan was performed and the image data were transferred to the operating room and fused with the preoperative MRI studies using the Framelink® software 5.1 (Medtronic, Iberica, Spain). The initial target was determined using the AC-PC reference system according to the Schaltenbrand and Wahren atlas [54]. The structures targeted were STN for PD and VIM for essential tremor (ET) (coordinates are shown in Table 2). The optimal trajectory was chosen to avoid the vessels, sulci, and ventricles. Duraseal® (Integra LifeSciences Services, France) was used to plug the burr holes to avoid CSF egress. The final electrode position was determined by microelectrode recording (MER), micro and macrostimulation. The sites showing the most typical STN spike activity in the MER were tested with micro and macrostimulation for neurological symptom improvement and possible side effects. The trajectory was modified until MER and stimulation effects were satisfactory. The electrodes implanted were the reference 3389 from Medtronic. Finally, the patient underwent implantation of the pulse generator. Table 1 summarizes the stimulation parameters and clinical outcomes of the patients and the number of trajectories performed.

Table 2 Relationship between atlas-based coordinates used for DBS with the coordinates of the optimized target (OT) obtained by our method and the clinically effective electrode contact. The coordinates are referenced from the AC-PC mid-commissural point
Full size table

Comparison of the tractography-based OT and the atlas-based target with the clinically effective electrode contact

After surgery, a CT scan was performed and the images were fused with the MR preoperative planning. In the early postoperative period, a neurologist expert in movement disorders evaluated the patients and the most clinically effective and best-tolerated contact combination was selected. DBS was programmed with a constant current, which permits to avoid the effect of the changes in the impedance taking place after surgery [47]. After a clinical follow-up period (Table 1), the coordinates of the center of the cathodic pole of the clinically effective electrode contacts (CEEC) were determined measuring their distance to the AC-PC line and midline sagittal plane. The coordinates of the OT were determined as the geometrical center of the parcellation measuring their distance to the AC-PC line and midline sagittal plane. These groups of coordinates obtained (CEEC coordinates and OT coordinates) and the planned coordinates, based on Schaltenbrand-Wahren atlas, are summarized in Table 2. The Euclidean distance between the CEEC coordinates and those of the atlas-based target was measured as follows: atlas-based √(XA − XE)2 + (YA − YE)2 + (ZA − ZE)2 where A is the atlas-based target coordinates and E is the CEEC coordinates (A-E distance). Similarly, the Euclidean distance between the CEEC coordinates and those of the tractography-based target (OT coordinates) was measured as follows: √(XT − XE)2 + (YT − YE)2 + (ZT − ZE)2 where T is the tractography-based target coordinates and E is the CEEC coordinates (T-E distance). Table 3 shows both the T-E and A-E distances and their difference for each case by side.

Table 3 Euclidean distance transformation of the coordinates
Full size table

Statistical analysis

Statistical data analysis was done using STATA software package version 12.0 (TS, USA). Wilcoxon signed-rank test was used to compare each distance and the median of the distances, with p < 0.05 considered as statistically significant.

Results

The patients were six males and three females, with a median age of 60 years. The surgical indications were medically resistant PD in six cases and medically resistant essential tremor (ET) in three cases. All patients showed significant clinical improvement with the stimulation, and there were no complications in the DBS procedures (Table 1).

The motor division of the STN was parcellated using the projections from M1, SMA, and pre-SMA. The regions obtained were mainly encountered in the dorsolateral part of the STN (Figs. 4 and 5). The projections from the DN, using the RN as “mid region” for the tracking, were used to reconstruct the dentate-rubro-thalamic (DRT) tract. This tract was found to be passing the Th through its infero-lateral border and was used for parcellation of the VIM nucleus (Figs. 3 and 5).

Fig. 5
figure 5

Optimized target of the STN and Th of patients with Parkinson’s disease and essential tremor. a Patient 4. Axial view, the subthalamic nucleus (STN) is shown (orange + green), the green area is the segment obtained by projections from M1, SMA, and pre-SMA. b Coronal view of a. It is important to note that the projections are mainly located at the dorsolateral region of the STN coincident with the motor part of the STN. c Patient 2. Axial view, the thalamus (Th) is shown (orange + green), and the projections from the dentate nucleus and the red nucleus are depicted in green. d Coronal view of c

Full size image

In eight patients and 14 electrodes (patient 2 right side, patient 3 left side), the T-E (tractography-based target-CEEC) distance was shorter than the A-E (atlas-based target-CEEC) distance (Figs. 6 and 7). In the left side of patient 2 and the right side of patient 3, the A-E distance was shorter than the T-E distance; and in patient 7, the A-E distance was shorter than the T-E distance bilaterally. The median T-E distance (IQ range) was 2.72 mm (2.23–4.00 mm), and the median A-E distance (IQ range) was 4.12 mm (2.82–6.00 mm) (Fig. 8). Wilcoxon’s test showed that the T-E distance was significantly shorter (p = 0.003) than the A-E distance in the majority of the patients with a positive difference, suggesting that the areas obtained by our method are more closely related with the most clinically accurate location than the first location planned at the beginning of the surgery (Fig. 8) (The raw data are summarized in Tables 3 and 4).

Fig. 6
figure 6

Patient 6 (Parkinson’s disease and target in the subthalamic nucleus (STN)). a Structural MRI showing the optimized target (OT) of the STN (green) obtained by our method. b Coronal view of a. c A zoom of the subthalamic area of the patient in a (white inset) with the electrodes superimposed; the green areas are the OT of the STN obtained by our method. We found that the most clinically effective electrode contacts are in close relation with the OT

Full size image
Fig. 7
figure 7

Patient 9 (essential tremor and ventral intermediate nucleus (VIM)). a Structural MRI showing the optimized target (OT) of the thalamus (green) obtained by our method. b A zoom of the thalamic area of the patient in a (white inset) with the electrodes superimposed. This patient was operated for medically refractory essential tremor with bilateral VIM nucleus DBS. Post-surgical CT scan was fused with MRI. The clinically effective electrode contacts (CEEC) (L with contact 2 and R with contact 11) are coincident with the OT obtained by our method. c 3D surface plot with cartesian coordinates (x, y, z) to illustrate the study design with the electrode of the right side of the patient in b: Clinically effective electrode contact (E) is shown in green, tractography-based target (T) is shown in red, and atlas-based target (A) in blue. The distance between T and E is the red line, and the distance between A and E is the blue line

Full size image
Fig. 8
figure 8

Statistical comparisons of the Euclidean distance T-E vs A-E. Dot plots and box plots overlaid. The boxplots show the median distance corresponding to T-E and A-E and the difference of the distances between A-E and T-E. The dots show the individual values of all observations. TE dist optimized target and clinically effective electrode contact distance, AE dist atlas-based target and clinically effective electrode contact distance. AE dist-TE dist difference of the distance between AE and TE

Full size image
Table 4 Descriptive statistics of the distances
Full size table

Discussion

Our present study describes a method for parcellation of DBS targets using DTI-DT implemented by a widely used navigation system (StealthStation® and StealthViz® software packages). With this method, we were able to identify an OT within the DBS target nuclei, such as the motor part of the STN and the VIM nucleus of the Th. An advantage of this method is that it could be entirely performed by a functional neurosurgeon using a commercially available surgical navigation system and clinically available MR sequences. There are other surgical planning workstations in the market such as Surgiplan® by Leksell stereotactic system. In this software, it is not possible to generate tracts from DTI information. However, this software offers image fusion tools, which enable to import the OT (obtained by tractography) as a digital imaging and communications in medicine (DICOM) file. This allows to coregister the OT with the structural MRI of the patient in order to be used during DBS surgery [44, 61].

The current techniques for targeting in DBS are based on atlases with a reported accuracy similar to the direct targeting in 3 T machines [58]. In the case of the STN, the contour of this structure in high-field MR machines, or the boundary of the red nucleus, is used by several authors to improve this accuracy [1, 5]. Direct visualization of the STN is a possibility to select the target [52]. Division of the directly visualized STN into four quadrants could be used to identify the dorsolateral part of the STN, where stimulation obtains the best clinical outcomes [67]. However, Coenen and coworkers [21], using an anatomo-radiologic analysis in a cadaver brain, showed that the sensoriomotor part of the nucleus is located at both the dorsolateral and anterior STN. Thus, a geometrical parcellation of the STN would not be sufficient to define the optimal target [57]. We would propose our method as an improvement of the direct visualization method since it uses the MR-defined STN to determine its intersection with the projection of the hyperdirect pathway defined by DTI-DT.

To interpret our results, we considered the best clinically effective electrode contact (CEEC) as the reference and compared its location with the atlas-based target and with the target defined by our method (OT). Our results show that, in most patients (eight patients and 14 electrodes), the OT obtained by our method is more closely related with the CEEC than the atlas-based target (Fig. 8). Also, we found that there is more variance in the distribution of the A-E distances; this result is explained by the need of performing multiple trajectories during DBS surgery to obtain the clinically and neurophysiologically reliable site for stimulation. Finally, we found that the difference between the A-E distance and the T-E distance is mainly positive, suggesting that the target obtained by our method is more accurate as long as we consider the CEEC as the reference measure for effective targeting.

Many methods have been described to parcellate subcortical structures in humans using different algorithms [4, 11, 18, 25, 26, 41]. Behrens et al. described a fully probabilistic algorithm method for human thalamus parcellation, and others have used this method to parcellate other subcortical structures and even for targeting [11, 41, 43, 51]. Here, we used deterministic tractography under the “knowledge-based approach” described by Mori et al. [45], applying neuroanatomical knowledge to assess the generated tracts [29]. DTI is a newly available resource to optimize DBS targeting preoperatively in an individualized fashion [19, 20, 22]. Colored FA maps have been used to identify certain DBS targets by recognizing the major tracts connected to these targets, yielding the maximum anatomical information from this available MRI sequence that is scarcely utilized in clinical practice [55, 56]. Connectivity-based approaches for DBS targeting using probabilistic tractography have recently been assessed and validated in clinical practice and promise to be superior to indirect methods [27, 51, 60]. The efficient use of connectivity-based approaches will depend on the armamentarium for high-order imaging acquisition and software-related resources to achieve individualized targeting. The recent development of these techniques has changed the paradigm of DBS surgery, providing new insight into the rationale of DBS targeting and the understanding of the stimulation mechanism on subcortical networks.

To optimize targeting, functional subdivisions in the STN have been explored [9, 16, 41, 66]; for example, impulse control disorders are related to stimulation of a more ventromedial location of the electrode (the limbic STN) [34, 53]. With our novel method, we were able to identify the motor STN, as well as the limbic STN, which has been proposed as a target in OCD (data not shown) [13, 14, 17, 35, 39]. There is some debate over the optimal targeting in using DBS for tremor [49, 65]. The subdivisions of the thalamus cannot be directly visualized on 1.5 and 3 T MR machines. The location of the interface between the VIM nucleus and the ventralis caudalis (VC) nucleus—which is the primary somatosensory integration center—is critical because stimulation close to the interface may cause intolerable paresthesias [49]. We were able to identify the region coincident with the VIM nucleus using the projection from DN and RN, which could enable the avoidance of stimulation paresthesias via direction of the electrode to a site far from the posterior border of the parcellation. We could also identify the location of the DRT tract, thus enabling stimulation optimization by placement of the electrode in a close relation with this tract [19, 20, 22, 36].

The main limitation of this work was the small number of patients involved, which made it difficult to correlate clinical results. However, the purpose of this paper is not to establish DTI as a standard procedure to determine targets but to describe the method and show its feasibility. Another limitation of this technique can be the interindividual variability and the reproducibility of the method, since the determination of the structures is subjective. Also, the manual segmentation process is time consuming and requires expertise in the use of planning station software; however, the software includes many tools and functions that could reduce the time required to perform this method. Some software allows automated parcellation that requires further adjustment of the cortical masks generated to deal with interindividual variability of the cortical gyri [24, 25]. A main advantage of this method is its simplicity, which lowers the need of human and computational resources.

Also, DTI technology must be used with caution. A single-tensor model (even with multiple directions) cannot describe the reality of huge voxels (2.6 mm) that have multiple fiber populations. So, the DTI values of the voxels must not be taken exactly as the real white matter populations that are intended to represent. Also, this low resolution of DTI could lead to interpret fibers belonging to nearby structures (e.g., the internal capsule) as fibers specific from the target structure (for example, the STN). However, our tractography-based target is defined by the intersection of the DTI fibers (from the projection areas) and the anatomical STN or Th (subcortical targets). Therefore, any voxel belonging to outside the STN or the Th (i.e., the internal capsule) would be avoided during the segmentation process. Distortion could shadow the anatomical accuracy; however, we used Functool® software to deal with the geometric distortion [8, 12, 45]. Additionally, the diffusively overlapping nature of the basal ganglia connections can limit the tracing of segregated loops throughout the cortico-subcortical circuits [26, 43]. Furthermore, it is not possible to determine the polarity of the fibers using DTI [11, 26]. Another limitation appears when an image voxel contains fiber populations with more than one dominant orientation [28]. However, neuroanatomical knowledge can enable the rejection of a misled group of generated fibers [29, 45]. Accumulated uncertainties in fiber orientation have clear a potential for leading to erroneous pathway reconstructions [12]. Probabilistic methods might adequately deal with some of these limitations [8, 10]. Moreover, the precision of the technique is still limited by the MR voxel size (around 1 mm) and the precision of the stereotactic instrument (about 0.5 mm). Brain shift due to the outflow of CSF could also distort the target location especially in the second treatment side. However, this would affect both the atlas-based target and the optimized target. Measures to minimize CSF egress must be taken such as plugging the burr holes with glue. These considerations also suggest that neurophysiological confirmation cannot be replaced by the procedure presented here.

The presently described method is not intended as a substitute for neurophysiological confirmation of the target but as a means of starting the targeting with a more individualized initial estimate. It could also be particularly helpful in patients who do not tolerate an awake surgery. Further investigation in a larger population is needed to determine if this method is more accurate than the standard atlas or plain MR image-based targeting methods.

Conclusions

Our results show that identification of the OT within the DBS target nuclei is feasible with our novel method. Based on neural circuits, we obtained plausible results that were consistent with clinical data in a group of patients, although prospective and controlled studies are necessary to demonstrate its beneficial role. The newly described method is straightforward and is entirely performed using available navigation software, with the possibility to use these regions during surgery as additional landmarks or to post-surgically explore the results of stimulation.