Variations in Thalamic Anatomy Affect Targeting in Deep Brain Stimulation for Epilepsy

Abstract

Background

Thalamic size and shape vary significantly across patients – with changes specific to the anterior thalamus occurring with age and in the setting of chronic epilepsy. Such ambiguity raises concerns regarding electrode position and potential implications for seizure outcomes.

Methods

MRIs from six patients from a single center underwent quantitative analysis. In addition to direct measurements from post-implantation MRIs, the CRAnialVault Explorer suite was used to normalize electrode position to a common reference system. Relationships between thalamic dimensions, electrode location, and seizure outcome were analyzed.

Results

Although this study group was too small to sufficiently power statistical analysis, general trends were identified. There was a trend towards smaller thalamic volumes in non-responders. Electrode locations demonstrated more variation after normalization. There was a trend towards a more lateral, posterior, and inferior electrode position in non-responders.

Conclusions

Variations in thalamic shape and volume necessitate direct targeting. Given changes that occur to thalamic anatomy with age and in the setting of epilepsy, improved methods for visualizing and targeting the anterior nucleus are necessary. Pronounced thalamic atrophy may preclude proper electrode placement and serve as a poor prognostic indicator. A greater understanding of thalamic anatomy and connectivity are necessary to optimize DBS for epilepsy.

Background

In the 1980s, Cooper and Upton were the first to describe chronic stimulation of the anterior nucleus of the thalamus for the treatment of epilepsy in six patients [1–4]. Although their results were positive, further reports did not emerge until approximately two decades later. This revival culminated in the landmark bilateral stimulation of the anterior nuclei of the thalamus for epilepsy (SANTE) trial; a multicenter, double-blinded, randomized study for which initial results were published in 2010 [5]. Four months after implantation, patients with active stimulation demonstrated a 29% greater reduction in seizure frequency when compared with the control group. Long-term follow-up showed a 41% median percent reduction in seizure frequency at one year, 56% at two years, and 69% at five years after implantation [6].

Unlike other targets for DBS, the ANT cannot be well defined by microelectrode recordings alone; and therefore necessitates clear visualization and direct targeting.[7] In the SANTE trial, as well as in prior reports, DBS electrodes were placed into the anterior nucleus of the thalamus (ANT) using direct targeting [5,8]. Furthermore, precise targeting relies on visualization of the mammilothalamic tract, external medullary lamina, and internal medullary lamina, which bound the ANT [7,8]. While recent studies have illustrated these structures clearly in high-channel MRI acquisitions and Short-TI Inversion Recovery (STIR) sequences [7,9], they are not consistently well demarcated on all clinical MRI platforms. As a result, electrode placement depends solely on targeting based on relative location within the thalamus and comparisons to the Schaltenbrand and Wahren atlas. Such an approach is suboptimal because of the high degree of variation in thalamic anatomy between patients and the lack of consistency in its location in the AC-PC coordinate system [7].

Variations in thalamic size and shape have been well documented. First of all, thalamic volume is known to decrease with age; but more importantly, this atrophy disproportionately affects the medial and anterior portions of the thalamus [10]. Given known thalamocortical connections, concurrent atrophy of the frontal lobes, temporal lobes, and region of the ANT is understandable. More importantly for the patient population of interest, a number of studies have demonstrated that patients with epilepsy experience thalamic atrophy [11] and that this atrophy is more pronounced on the side ipsilateral to seizure onset in patients with temporal lobe epilepsy [12] and occurs in a time-dependent fashion relative to the duration of disease. Pathway-specific deafferentation also occurs and results in atrophy of frontocentral and limbic cortices, which is significantly greater than that seen with aging alone [13,14]; and atrophy specifically of the ipsilateral anterior thalamus has been described in patients with chronic refractory mesial temporal lobe epilepsy [15].

Such variability and ambiguity raises concerns regarding electrode position and potential implications on clinical outcomes [7]. We therefore intended to critically evaluate electrode positions in patients implanted at our center and assess the impact of this variable on seizure reduction.

Methods

Under approval of the institutional review board at Thomas Jefferson University, charts and perioperative MRIs from the eight patients involved in the SANTE trial were reviewed. SANTE study records were also reviewed to collect 5-year outcome data with regards to seizure reduction. Volumetric (1 mm³ voxel) T1-weighted MRI acquired both preoperatively and postoperatively was required for quantitative analysis. As such, two patients were excluded from the analysis because their preoperative imaging was no longer available on the picture archiving communication system (PACS). Both direct and normalized measurements were taken for the remaining six patients. [Figure 1]

Figure 1.

Workflow of image processing and analysis. Thalamic measurements were taken from multiplanar reconstructions of the preoperative MRI and overlays of thalamic segmentation were used as guidance when uncertainty arose. Thalamic volumes were generated directly from the automated segmentation of the preoperative MRI. Relative contact positions were taken from multiplanar reconstructions of the postoperative MRI; and raw contact locations were taken directly from the postoperative MRI in which AC-PC has been defined. Normalized contact locations were generated by normalizing both pre- and post-operative MRIs to the CranialVault atlas. (In this figure, the green outline represents the ANT and the red dot represents the location of the active contact.)

Direct Measurements

Preoperative images underwent automated segmentation (FreeSurfer, Boston, MA) to calculate thalamic volumes [16]. Multiplanar reconstructions of the preoperative MRIs were generated (OsiriX, Pixmeo, Geneva, CH) to allow measurement of the maximal anterior-posterior and medio-lateral diameters. In cases when the thalamic border was not well-defined, an overlay of the thalamic segmentation was used as guidance for these thalamic measurements.

For each patient, the locations of the active contacts were recorded relative to the midcommissural point (MCP) and to the borders of the thalamus. The raw contact location was recorded as the distance in each direction (x, y, z) from the center of the active contact to the MCP in the AC-PC coordinate system. The WayPoint planning software (FHC, Bowdoin, ME) was used to define AC, PC, and a midline point on the postoperative MRI and identify the raw contact location of the active contact on each side. To maintain consistency between patients, all postoperative MRIs underwent multiplanar reconstruction such that the axial images were parallel with the AC-PC line and orthogonal to the plane formed by the AC-PC line and a midline point. At the level of the active contact of the DBS electrode, the relative contact position within the thalamus was measured as the distance from its center to the medial and anterior borders of the thalamus.

Normalized Measurements

Image sets were further processed by leveraging the CranialVault central repository and its accompanying CRAnialVault Explorer (CRAVE) suite to normalize electrode position to a common reference system based on 7T MRIs. This platform performs both a global rigid registration and a non-rigid coregistration focused on the deep structures of the basal ganglia and thalamus. These algorithms have been well validated for typical DBS targets and are currently being used for clinical purposes [17]. We were therefore able to use this platform to normalize subcortical structures and electrode positions of all six SANTE patients into a common reference atlas. The normalized contact location was recorded as the distance in each direction (x, y, z) from the center of the active contact to the MCP in the atlas.

Statistical Analysis

All statistical analysis was performed in MATLAB within the Statistics and Machine Learning Toolbox (MathWorks, Natick, MA). Relationships between all measurements and contact positions were analyzed via a correlation matrix. Measurements and contact positions were also compared between responders and non-responders with an unpaired t-test. Given the small sample size of six patients, statistical analysis of the results was largely underpowered. Trends in such relationships were otherwise noted.

Results

Clinical Outcomes

Demographic information and 5-year seizure outcomes for the six patients included in the analysis are shown in Table 1. At the end of five years, three responders had worthwhile improvement in seizures (60-90% seizure reduction) and three non-responders had no worthwhile improvement (no significant seizure reduction). One of the three non-responders had her system explanted secondary to infection within two years of implantation, but had no worthwhile improvement prior to that time.

Table 1.

Demographic information and 5 year seizure outcomes (Engel Class) for the six SANTE patients included in the analysis of DBS electrode position within the anterior nucleus of the thalamus. Responders (Patients A-C) demonstrated 60-90% seizure reduction; while non-responders (Patients D-F) demonstrated no seizure reduction.

Patient	Age	Sex	Seizures		Seizure Outcome
			Type	Frequency (per month)	3mo	6mo	1yr	2yr	3yr	4yr	5yr
A	24	Male	CPS & SPS	9	III	III	III	III	III	III	III
B	22	Female	CPS & GTC	60	IV	IV	IV	IV	IV	IV	III
C	60	Male	SPS	67	IV	III	III	III	III	III	III
D	18	Male	CPS	75	IV	IV	IV	IV	IV	IV	IV
E	47	Female	CPS & GTC	20	IV	IV	IV	IV	IV	IV	IV
F	21	Female	CPS & SPS	>100	IV	IV	IV	infection -> explant

With regards to seizure type, 2/5 (40%) of patients with complex partial seizures demonstrated seizure reduction; as did 2/3 (66%) of patients with simple partial seizures and ½ (50%) of patients with generalized tonic-clonic seizures. The size of this study group was too small to perform a multivariate regression analysis to determine the significance of seizure type on seizure outcome from ANT DBS.

Relationships between Measurements and Contact Positions

Larger thalamic volumes were associated with longer thalamic measurements (r = 0.63, p = 0.028) and a more lateral relative contact position (r = 0.66, p = 0.021). A more lateral relative contact position was also associated with a more lateral normalized contact location (r = 0.72, p = 0.009), but was not significantly associated with the raw x coordinate. No other relationship was found to be statistically significant.

Thalamic Dimensions

Responders were found to have shorter but wider thalamic measurements; and overall volumes that were lower. These differences were not statistically significant [Table 2].

Table 2.

Thalamic Dimensions for the six SANTE patients included in the analysis.

Patient	Thalamic Length (mm)		Thalamic Width (mm)		Thalamic Volume (cc)
	Right	Left	Right	Left	Right	Left
Responders
A	28.6	25.7	10.2	12.4	9.05	7.43
B	21.2	18.9	8.6	10.6	6.02	6.74
C	26.7	21.2	11.7	9.5	8.04	8.50
x ± σ	23.7 ± 3.5		10.5 ± 1.3		7.63 ± 1.03
Non-Responders
D	26	27.9	6.8	11.9	8.61	10.16
E	26	25.5	9.5	7.9	7.98	6.52
F	22.9	24.7	8.5	11.2	7.37	7.93
x ± σ	25.5 ± 1.5		9.3 ± 1.8		8.09 ± 1.12

Raw Location of Active Contacts

There was no significant difference in relative contact position between the two groups. Similarly, although responders had a slightly more lateral (x coordinate) and anterior (y coordinate) raw contact location on average, this difference was not statistically significant. In the z-axis, however, responders were found to have a more inferiorly located active contact than their counterparts (p = 0.009). [Table 3] [Figure 2]

Table 3.

Locations of active contact for the six SANTE patients included in the analysis. Relative contact position was measured from the medial border of the thalamus (lateral measurement) and from the anterior border of the thalamus (posterior measurement). Both raw and normalized contact locations are expressed relative to the midcommissural point (MCP). As such, locations left of, in front of, and above the MCP have positive values; while locations right of, behind, and below the MCP have negative values. Averages and standard deviations were taken from the absolute values of the x coordinate, but from the actual value of the y and z coordinates.

Patient		Relative Contact Position		Raw Contact Location			Normalized Contact Location
		Lateral	Posterior	x	y	z	x	y	z
Responders
A	R	2.9	4.0	-4.1	1.2	10.2	-5.1	3.4	11.6
	L	5.2	5.2	4.2	0.7	11.8	10.0	3.3	13.1
B	R	3.8	4.2	-5.2	-0.5	10.7	-9.5	3.3	13.3
	L	3.0	4.2	3.9	-0.3	11.9	9.8	3.4	14.5
C	R	3.9	2.2	-3.9	1.0	9.6	-10.8	5.6	10.0
	L	5.8	4.7	5.8	0.6	10.6	12.1	6.3	12.1
x ± σ		4.1 ± 1.2	4.1 ± 1.0	4.5 ± 0.8	0.5 ± 0.7	10.8 ± 0.9	9.5 ± 2.4	4.2 ± 1.4	12.4 ± 1.6
Non-Responders
D	R	4.9	3.2	-4.2	-2.7	11.7	-12.1	1.2	11.5
	L	4.4	3.6	4.4	-1.2	11.7	13.6	3.9	12.0
E	R	3.4	3.7	-4.7	0.9	12.3	-8.4	1.9	12.2
	L	4.4	4.7	4.3	0.8	12.6	11.9	4.6	13.1
F	R	3.9	2.1	-4.5	0.7	11.9	-8.0	4.9	10.7
	L	2.5	2.5	3.7	1.4	13.6	7.2	6.4	13.3
x ± σ		3.9 ± 0.9	3.3 ± 0.9	4.3 ± 0.3	0.0 ± 1.6	12.3 ± 0.7	10.2 ± 2.7	3.8 ± 2.0	12.1 ± 1.0
Entire Cohort
x ± σ		4.0 ± 1.0	3.7 ± 1.0	4.4 ± 0.6	0.2 ± 1.2	11.5 ± 1.1	9.9 ± 2.4	4.0 ± 1.6	12.3 ± 1.2

Figure 2.

Raw location of active contact for responders (top row) and non-responders (bottom row).

Normalized Location of Active Contacts

After normalization of all patient images, responders on average had active contacts located medial, anterior, and superior to that of their counterparts; although these differences were not statistically significant. [Table 3] [Figure 3]

Figure 3.

Normalized locations of the active contact for responders (green) and non-responders (red) overlaid onto the 7T CranialVault atlas normalized using non-rigid image registration. Points have been projected onto a representative slice and do not all lie on a single plane. Although not statistically significant, responders tended to have active contacts that were more medial, anterior, and superior than non-responders. The anteroventral (AV) subnucleus as defined by normalized 7T MRI is highlighted in green and is represented as a three dimensional volume in the bottom right pane.

Discussion

In the SANTE trial, electrode position was validated under visual inspection by an independent third party [5]. As such, it follows that all electrodes appeared to reside in the region of the ANT. In our cohort, there was little difference in the raw location of the active contacts; with a standard deviation no greater than 1.2mm in all three planes. Despite this precision, we still observed differences in clinical outcome. While likely multifactorial in nature, the effect of DBS on epilepsy has been shown to vary based on the location of stimulation within the ANT [18–20]. It is therefore important for us to better understand thalamic anatomy in an effort to optimize trajectory planning to the ANT and electrode position within the thalamus.

In line with prior reports [7,18], we too found significant variability in thalamic dimensions and volumes between patients. Even more concerning is that changes to thalamic anatomy are specific to the region of the ANT and are of a greater magnitude in patients suffering from long-term epilepsy [10–14,21]. This anatomical inconsistency necessitates direct targeting of thalamic structures based on visualization of the target on preoperative MRI. While recent studies have illustrated the utility of high-channel MRI acquisitions and STIR sequences in imaging the ANT[7,9], these details were not defined when the SANTE trial was started over ten years ago. A post hoc analysis of study patients has recently revealed that nearly 10% of electrodes were not within the limits of the ANT [22]. Interestingly, we found that in our cohort only two active contacts were within the limits of the ANT and five active contacts were close enough such that the average activation volume of 71 mm³ [23] would involve the ANT [Figure 3].

In order to properly assess the effect of electrode location across patients, perioperative image sets must be normalized in order to account for differences in patient anatomy [24]. Although a recent analysis of electrode positions in the SANTE trial demonstrated that there was no difference between responders and non-responders[22], the analysis used raw electrode locations and did not account for individual patient anatomy. Lehtimäki et al have proposed an ANT-normalized coordinate system by which electrode location can be assessed [18]. While their normalization scheme requires someone to manually define the borders of the ANT, our analysis leverages a previously validated, automated method for normalizing deep brain structures across patients [17,25,26].

After normalization with CRAVE, responders tended to have active contacts that were more medial, anterior, and superior than non-responders. Although this relationship was not statistically significant, it is interesting to note that Lehtimäki et al reported similar findings. Specifically, they found that after normalization to an ANT-normalized coordinate system, contacts associated with clinical response were located significantly more superior and anterior (p < 0.01) [18].

Anatomical Correlates

When one considers the location of the ANT in the medial, anterior, and superior portion of the thalamus, this difference between responders and non-responders is logical. Furthermore, it has been suggested that stimulation specifically of the anteromedial (AM) and anteroventral (AV) subnuclei of the ANT is important given their connections to the frontal cortex, cingulate gyrus, and amgydalohippocampal complex [18,19]. The AV subnucleus in particular has the most extensive connections with the temporal circuit and is involved in generating theta rhythm activity felt to promote synaptic plasticity in the hippocampal circuit [19]. The AM subnucleus is more intimately connected with orbitofrontal circuits and is felt to be more involved in cognitive, emotional, and executive functions [19]. Meanwhile, the smaller anterodorsal (AD) subnucleus, located posteriorly along the dorsal aspect of the ANT, is part of a separate network involving the lateral mammillary nucleus and felt to be responsible for spatial navigation and memory [19].

A more anterior location of stimulation within not only the thalamus, but also within the ANT would affect both the AV and AM subnuclei; while a more superior location of stimulation within the ANT would affect the AV subnucleus primarily. [Figure 4] It has also been hypothesized that more posterior and inferior stimulation may be unable to impact enough ANT neurons because the intervening internal medullary lamina prevents sufficient anterior spread of current [18]. This theory has been supported in the rodent model, where stimulation applied posteriorly and inferiorly did not impart any antiepileptic effect [20]. A more anterior and superior active contact location is therefore supported by ANT anatomy, with activation specifically of the AV and AM subnuclei.

Figure 4.

Anatomical relationship of anteroventral (AV, red), anteromedial (AM, blue), and anterodorsal (AD, yellow) subnuclei of the anterior nucleus of the thalamus (ANT). The AD subnucleus is the largest of the three and occupies the anterior and dorsal portions of the ANT. The AM subnucleus sits below the AD subnucleus anteriorly. The AM subnucleus is the smallest of the three and sits posteromedially within the ANT. The number below each slice indicates the distance in millimeters from the midcommissural point. (Adapted from Schaltenbrand-Wahren atlas)

Clinical Implications

Electrode location in DBS for movement disorders has been shown to affect clinical outcomes [27–29]. Similar findings have emerged for DBS of the ANT for epilepsy as well. Lehtimäki et al reported that six patients in their series experienced increased seizure reduction only after reprogramming to activate the most superior contacts [18]. Other studies have also demonstrated that only more anteriorly placed contacts within the ANT are associated with hippocampal evoked potentials in humans [30] and more superiorly located contacts were associated with suppression of hippocampal activity [31]. As with other applications of DBS, contact location appears to play a critical role in clinical outcome.

As outlined above, ANT anatomy, our findings, and those of Lehtimäki et al all suggest that more anterior and superior stimulation, which predominately affects the AV subnucleus, may result in better seizure control. The role of the AV subnucleus in the thalamic circuit may explain why a recent subgroup analysis of SANTE trial patients demonstrated greater improvement in patients with temporal lobe epilepsy than in those with frontal lobe epilepsy [22]. In that same vein, stimulation of the AM subnucleus could theoretically result in better seizure control for patients with frontal lobe epilepsy. While such claims associating specific thalamic subnuclei to seizure type are premature, what is clear is the importance of accurate electrode placement and the need for direct targeting based on individual patient anatomy. Given the variations in thalamic anatomy between patients, direct targeting of the ANT is imperative and has been shown to improve the accuracy of electrode placement [18].

Study Limitations

The main limitations of this study are the retrospective nature and the small sample size. While we benefited from rigorous data collection performed as part of the SANTE trial, we were limited with regards to perioperative imaging. Image quality, in particular, was not consistent in our series; and unlike prospective studies we could not optimize acquisition parameters to improve visualization of the ANT. Instead, for our analysis, we relied on a previously validated method for non-rigid normalization of deep brain structures to an atlas based on multiple 7T MRIs. This process requires a high resolution preoperative MRI for the purpose of normalization to the atlas, and a postoperative MRI for the purpose of electrode localization. Unfortunately, the combination of the 1.5T MRI and metal-induced artifact from the implant prohibit use of the postoperative scan alone for both anatomical normalization and electrode localization. It is for this reason that we had to exclude two of our own patient from the analysis; and could not extend the analysis beyond our center. Unfortunately, only postoperative MRIs were archived for the SANTE trial (G Molnar, Medtronic, personal communication, June 26, 2015). As such, we were only able to identify trends and were unable to achieve statistical significance in our data analysis. Although our findings do agree with prior work, our capacity to draw definitive conclusions is certainly limited. Instead of serving as a definitive solution, we believe that the importance of this work is to highlight the problem at hand and present a method for critically analyzing electrode position across patients in ANT DBS.

Future Directions

In order for us to better understand these variables, larger clinical studies further investigating electrode position and its effects on clinical outcomes in epilepsy must be performed. Specifically, based on our analysis presented here, a recruitment of 50 patients (25 in each cohort) would be required to sufficiently power an analysis of electrode locations to detect a difference of 2mm between responders and non-responders. We intend to apply the methods presented here to analyze data from an existing European registry of epilepsy patients undergoing ANT DBS to determine the effects of active contact position on seizure control.

While recent work has demonstrated the ability of customized sequences on clinical scanners to visualize the ANT for the purpose of direct targeting, borders of ANT subnuclei still cannot be discerned. Even as imaging capability improves, incorporation of functional connectivity based on blood oxygen level dependent contrast imaging and anatomical connectivity based on diffusion tensor imaging, may help to better define a target within the ANT for DBS for epilepsy [32–37]. Such a multimodal imaging approach may be necessary in order to cater an implant to each patient – especially given the changes that occur to thalamic volume, shape, and connectivity in the setting of epilepsy [11–15]. Pronounced atrophy or pathway-specific deafferentation may even preclude proper electrode placement and therefore aid in patient selection for ANT DBS.

Conclusions

Variations in thalamic shape and volume necessitate direct targeting. Preliminary findings associating improved seizure control with a more anterior and superior active contact location may be associated with stimulation of the anteroventral subnucleus of the anterior nucleus of the thalamus. Given the anatomical variation and changes that occur to thalamic anatomy over time and in the setting of epilepsy, improved methods for visualizing and targeting the anterior nucleus of the thalamus are necessary. Pronounced thalamic atrophy may preclude proper electrode placement and serve as a poor prognostic indicator. Ultimately, a greater understanding of thalamic anatomy and connectivity along with a method for accurate thalamic targeting are necessary to optimize DBS for epilepsy.