Professional Documents
Culture Documents
2014 - Evaluating Indirect Subthalamic Nucleus
2014 - Evaluating Indirect Subthalamic Nucleus
a
Surgical Therapies Improving Movement Program and Departments of b Neurosurgery, c Neurology and d Biomedical
E-Mail karger@karger.com
1500 East Medical Center Drive, SPC 5338, Ann Arbor, MI 48109-5338 (USA)
www.karger.com/sfn
E-Mail pgpatil @ umich.edu
outside of the motor STN is thought to produce side ef- Table 1. Indirect targeting of the STN for DBS surgery
fects such as cognitive decline and behavioral changes
reported in nearly half of STN DBS cases [11–13]. How- MCP-based targeting RN-based targeting
ever, targeting the STN can be very challenging. The STN x 12 mm lateral 3 mm lateral of lateral border
is small (158 mm3) and poorly visualized on convention- y 3–4 mm posterior y-coordinate of anterior border
al MRI (1.5 T). Furthermore, it is often indistinguishable z 3–5 mm inferior 2 mm inferior of superior border
from the adjacent substantia nigra (SN) without the use
of high field strength MRI (3–9.4 T), which is not widely
available around the world [14].
To circumvent the limitations to direct STN visualiza- known precisely, it is hypothesized that the active con-
tion on conventional 1.5-tesla MRI, neurosurgeons have tact of the DBS electrode should be located in the dorso-
utilized a combination of presurgical indirect targeting lateral (motor) region of the STN [5–8]. The challenge
techniques and intraoperative microelectrode recordings of these studies is that the location of the DBS electrode,
to target the STN for DBS surgery [15]. These presurgical and therefore the location of the most efficacious con-
indirect techniques are based on the position of the STN tact, is itself constrained by and dependent upon the tar-
relative to internal fiducial markers as imaged on MRI, geting method used during surgery. With the develop-
such as the anterior and posterior commissures, accord- ment of a validated 3-tesla MRI protocol to visualize the
ing to neuroanatomical atlases [6–8, 16, 17]. Among these STN [1], the opportunity arises to evaluate targeting
indirect techniques, the most commonly used are based methods without this limitation. Determining which
upon the location of the midcommissural point (MCP) targeting method most reliably predicts the location of
[18–20] or the borders of the red nucleus (RN) [15, 17] the midpoint of the STN may allow neurosurgeons to
(table 1), although other targeting approaches have been target the dorsolateral region of the STN more reliably
proposed [21, 22]. Optimization of presurgical indirect and efficiently.
targeting methods can reduce the number of surgical In this study, we located the midpoint of the MR-visu-
passes, thus reducing the risk of bleeding and the length alized STN (fig. 1). We then compared the accuracy and
of surgery [10]. Further, although microelectrode record- precision of MCP- and RN-based targeting methods rela-
ing can assist STN targeting with the availability of a tive to this location. Since we were able to visualize the
trained electrophysiologist, it has been reported to in- midpoint of the STN directly, we then evaluated the im-
crease the risk of bleeding [10, 23]. These factors motivate pact of neuroanatomical variability, such as changes in
the continued use of indirect targeting techniques during third-ventricle size, on STN location. We utilized regres-
STN DBS surgery. sion analysis to optimize the indirect targeting model and
Determining the relative quality of indirect STN tar- to compare this optimization with traditional MCP- and
geting techniques is of broad interest to neurosurgeons, RN-based targeting of the STN. Finally, we evaluated RN-
particularly at international centers or in intraoperative based targeting utilizing images at 1.5-tesla field strength
MR settings where the absence of high field strength MRI compared to RN-based targeting utilizing images at
(>1.5 T) makes direct STN targeting difficult. Two mea- 3-tesla field strength. Taken together, our findings sug-
sures of targeting quality are of interest: accuracy and pre- gest that neurosurgeons employing RN-based indirect
cision. Accuracy is the measure of the degree of closeness targeting may expect more consistent results than those
between the prediction of a quantity and its actual value. employing MCP-based indirect targeting.
For example, the prediction of the STN midpoint will be
close to its actual location in an accurate targeting tech-
nique. Precision, by contrast, is the measure of the degree
Methods
of variability in repeated applications of the targeting
technique. For example, if the distance from the predicted Patient Selection
STN location to its actual location is highly variable across We evaluated the 3-tesla MR images of 58 patients who had un-
patients, the targeting technique has low precision. dergone bilateral STN DBS for advanced idiopathic PD (42 men,
Previous studies have evaluated targeting techniques 16 women, mean age 63 ± 7 years, range 46–78 years). Selection
criteria for DBS included a diagnosis of idiopathic PD, with the
by comparing their predictions to the coordinates of the presence of motor fluctuations not optimally managed with medi-
most efficacious, active DBS electrode contact [17]. cations, or severe levodopa-unresponsive tremor. Patients with
While the ideal stimulation site in the STN region is not contraindications to 3-tesla MRI scanning, structural abnormali-
128.122.253.228 - 5/17/2015 4:45:11 PM
NYU Medical Center Library
Midpoint
ties on preoperative MRI or dementia by neuropsychological test- en, mean age 63 ± 7 years, range 46–78 years; 9 patients in the orig-
ing were excluded from the study. We performed the study in ac- inal group were excluded due to incomplete whole-brain imaging
cordance with the policies of the Medical Institutional Review data): brain length was measured as the maximal anteroposterior
Board of the University of Michigan. dimension in the intercommissural plane on either side of the mid-
line; brain width was measured as the distance between the insular
MRI and Analysis cortical pia at the anterior commissure and the temporoparietal
Patients under evaluation for DBS surgery received preoperative cortical pia at the posterior commissure; frontal horn width was
volumetric 3-tesla MRI scans (Philips Achieva; Philips, Amsterdam, measured as the maximal lateral ventricular width on axial slices;
The Netherlands). The 3-tesla MRI protocol was developed by our third-ventricle width was measured in the intercommissural plane
group and optimized for STN visualization, and has been validated at the MCP, and brain height was measured as the maximal vertical
against cadaveric anatomical studies and intraoperative electro- length in coronal images at the anterior commissure. All anatomical
physiology [1]. The images were oriented in Talairach coordinate measurements were performed by the neurosurgeon performing
space along the intercommissural and midsagittal planes (Analyze the DBS surgeries (P.G.P.) to minimize interobserver variability.
9.0; AnalyzeDirect Inc., Overland Park, Kans., USA). Consistent To allow comparison between imaging field strengths, RN-
with neurosurgical convention, the MCP was designated as the ori- based indirect targeting coordinates were calculated for 10 patients
gin. Positive x, y and z directions were defined as left, anterior and who had received 3-tesla imaging with our validated protocol as
inferior, consistent with targeting using the Leksell Stereotactic well as 1.5-tesla T2-weighted imaging (TR 5550, TE 82, 2-mm slice
Frame (Elekta AB, Stockholm, Sweden). Windowing of the 3-tesla thickness). Images at 1.5 T were oriented in Talairach coordinate
MR data for visualization was uniformly centered at the level that space through coregistration (Analyze 9.0; AnalyzeDirect Inc.).
maximized contrast between the STN and the SN and a width twice Coordinates of the RN borders were measured in identical and
that amount to minimize image measurement variability. blinded fashion to images acquired at 3 T.
Since the shape of the STN is not precisely ellipsoidal, we de-
fined the midpoint of the STN as the location midway between the Statistical Methods
oral and caudal poles of the STN, which were selected on coronal Statistical analysis was performed using commercially available
MRI for closest correspondence to a well-established atlas of the software: SAS (SAS Institute Inc., Cary, N.C., USA), Excel (Micro-
human brain (Atlas of the Human Brain, plates 31–38) [24] (fig. 1). soft Inc., Redmond, Wash., USA) and MATLAB (MathWorks,
Our method is consistent with previously published methodology Natick, Mass., USA). For all comparisons, statistical significance is
on determining the midpoint of the STN [9]. Coordinates of the defined at the p < 0.05 level. Except where otherwise noted, the co-
RN borders were measured as defined in prior studies [15, 17]. ordinate data are reported as mean ± 95% confidence interval (CI).
Briefly, the anterior, lateral and superior borders were selected as Comparisons of coordinates were performed with two-tailed paired
the most anterior (axial plane), lateral (axial plane) and superior t tests for nonindependent data. The Bonferroni correction was uti-
(coronal plane) points on volumetric 3-tesla MRI in Talairach ori- lized for multiple comparisons. Comparisons of the precision of
entation. standard RN-based targeting and novel targeting methods were per-
In addition to the STN midpoint location, whole-brain neuro- formed with right-tailed F tests. We compared the variances in each
anatomical parameters were measured as follows (35 men, 14 wom- coordinate on each side of the brain under the null hypothesis that
128.122.253.228 - 5/17/2015 4:45:11 PM
NYU Medical Center Library
Evaluating Indirect STN Targeting with Stereotact Funct Neurosurg 2014;92:337–345 339
3-Tesla MRI DOI: 10.1159/000366286
Downloaded by:
Table 2. Coordinates of the STN midpoint determined by validated 3-tesla MRI (n = 58)
x y z
Measurements are in millimeters and reported as means ± 95% CI. RSTN = Right STN; LSTN = left STN.
Table 3. Pearson’s correlation coefficients for STN coordinates and neuroanatomical parameters (n = 49)
Coordinate R brain length L brain length Brain width Brain width Frontal horn Third-ventricle Brain
at AC at PC width width at MCP height
Empty cells indicate the absence of statistically significant correlation. RSTN = Right STN; LSTN = left STN; AC = anterior commis-
sure; PC = posterior commissure.
the variance of two normal populations are the same and the alter- Correlation of STN Midpoint Coordinates and
native hypothesis that the variance of the standard RN-targeting Neuroanatomical Parameters
population is greater than the variance of the model RN-targeting
population. The best simple linear regression model for each coor-
Table 3 reports the correlation between the right and
dinate was determined as the model with the highest R2 value that left STN midpoint coordinates and major neuroanatom-
had statistically significant parameter estimates (p < 0.05). For each ical parameters, as defined in Methods. Significant cor-
linear regression model, the data were tested for homoscedasticity relation for the laterality (x-coordinate) of the STN is
with the Kolmogorov-Smirnov test. Sources of statistical variability present for CSF spaces in the brain. Correlation between
in our data include quantization effects due to voxel representation
intrinsic to MRI and human error in measurement [1].
laterality and maximal frontal horn width is weaker,
though statistically significant (right: r = –0.33, p = 0.02;
left: r = 0.37, p < 0.01). By comparison, we found stronger
correlation between STN laterality and third-ventricle
Results width (right: r = –0.54, p < 0.01; left: r = 0.51, p < 0.01).
Linear regression of bilateral STN midpoint separation
STN Midpoint Coordinates and third-ventricle width yielded strong correlation
Table 2 reports the x-, y- and z-coordinate locations of (fig. 2), with third-ventricle width accounting for 40% of
the right and left STN in our study (n = 58). The STN mid- the variance in STN separation (R2 = 0.4). Juxtaposed
points are symmetrical within 95% CI limits for individual with this strong overall explanatory value, the individual
patients. The MCP is designated as the origin and left, an- lateral coordinates of the right and left STNs were only
terior and superior represent increasing x, y and z coordi- weakly explained by the width of the third ventricle (right:
nates, respectively. With this convention, the right STN R2 = 0.29; left: R2 = 0.26). In the rostrocaudal dimension,
midpoint was located at x = –10.90 ± 0.31 mm, y = –0.66 ± there were no statistically significant correlations ob-
0.35 mm and z = –3.63 ± 0.29 mm. The left STN midpoint served between the y-coordinate of the STN midpoint
was located at x = 11.41 ± 0.28, y = –1.23 ± 0.34 and z = and any of the neuroanatomical parameters that we ex-
–3.88 ± 0.31. These values compare well with traditional amined. In the dorsoventral dimension, there were sig-
targeting coordinates for the dorsolateral STN (table 1). nificant correlations bilaterally between brain length
128.122.253.228 - 5/17/2015 4:45:11 PM
NYU Medical Center Library
0.8
Creating an Optimized Targeting Model 0.6
To create an optimized targeting model, we evaluated 0.4
the targeting parameters (location relative to MCP, RN
0.2
borders and neuroanatomical parameters) with regres-
0
sion analysis. Table 4 reports the final optimized model MCP RN Optimized RN
parameters. Regression analysis eliminated the MCP co-
ordinates and all other measured neuroanatomical pa-
Fig. 3. Comparison of MCP- and RN-based targeting accuracy. Ab-
rameters as statistically insignificant. By contrast, the solute errors in targeting (mean ± SD) are plotted for MCP, RN and
borders of the RN correlated strongly with STN midpoint optimized RN models. Grayscale indicates x-, y- and z-coordinates.
coordinates, with slopes for all regression models approx-
imately equal to 1 (range 0.91–1.05). The best-optimized
predictors for the y-coordinate of the right STN and all of
Table 4. Results of STN-RN regression modeling (n = 58)
the left STN coordinates were the same as those used dur-
ing standard RN-based targeting (lateral, anterior, and Regression line R2
superior RN borders for x, y and z STN midpoint coordi-
nates, respectively). For the right STN midpoint, howev- RSTNx = 1.05 × RSx – 5.77 0.56
er, the x-coordinate was better predicted by the superior RSTNy = 0.94 × RAy + 1.42 0.45
border of the RN than the lateral border (R2 = 0.56 vs. RSTNz = 0.96 × RAz + 1.39 0.70
LSTNx = 0.91 × LLx + 3.27 0.58
0.48), while the z-coordinate was better predicted by the LSTNy = 0.98 × LAy + 1.05 0.55
anterior border of the RN than the superior border (R2 = LSTNz = 1.04 × LSz – 1.75 0.65
0.70 vs. 0.65). These small improvements in predictive
power resulted in a small but statistically significant im- RSTN = Right STN; LSTN = left STN; RL/LL = right and left
lateral; RA/LA = right and left anterior; RS/LS = right and left
provement in accuracy for the optimized regression mod-
superior; x, y, z indicate corresponding coordinate directions.
el, without improvement in precision (fig. 3).
128.122.253.228 - 5/17/2015 4:45:11 PM
NYU Medical Center Library
Evaluating Indirect STN Targeting with Stereotact Funct Neurosurg 2014;92:337–345 341
3-Tesla MRI DOI: 10.1159/000366286
Downloaded by:
a b
Fig. 4. Comparison of coronal 3-tesla (a) and 1.5-tesla (b) MRI images for RN visualization in the same patient.
The RN is well visualized at both 3 and 1.5 T. By contrast, the STN and SN are only well visualized at 3 T.
MCP MCPx – 10.90 MCPy – 0.66 MCPz – 3.63 MCPx + 11.41 MCPy – 1.23 MCPz – 3.88
RN RLx – 2.75 RAy + 1.55 RSz – 1.68 LLx + 2.55 LAy + 1.08 LSz – 1.84
Optimized RN 1.05 × RSx – 5.77 0.94 × RAy + 1.42 0.96 × RAz + 1.39 0.91 × LLx + 3.27 0.98 × LAy + 1.05 1.04 × LSz – 1.75
Comparison of Targeting Parameters 3-tesla images. As expected with imaging at lower field
Table 5 presents the final optimized STN midpoint strength, tissue contrast and delineation of structural
targeting parameters for MCP-based, RN-based and borders are less distinct at 1.5 T than at 3 T.
regression optimized targeting methods. MCP-based To evaluate the application of our 3-tesla RN-based tar-
midpoint targeting parameters were equivalent to values geting methodology to 1.5 T, we performed a blinded
found in STN midpoint measurements. Optimized tar- comparison of RN boundary coordinates in 10 patients
geting parameters were determined through linear re- who had imaging performed at both field strengths. Max-
gression. Both RN-based and optimized targeting strate- imal RN-based targeting coordinate differences between
gies were significantly more accurate for individual 3- and 1.5-tesla images were less than 1 voxel, ranging
patients (p < 0.05) and more precise (p < 0.002) than from –0.41 ± 0.17 mm (mean ± SEM) for the right supe-
MCP-based targeting. RN-based and regression opti- rior z-coordinate to 0.27 ± 0.22 mm for the left anterior
mized RN-based methods were not statistically distin- y-coordinate. None of the indirect targeting differences
guishable. RN-based coordinates for the STN midpoint between images at 3 and 1.5 T were statistically significant.
compared well with traditional targeting coordinates for
the dorsolateral STN (table 1).
Discussion
Evaluation of RN-Based Targeting Results at 1.5 T
Figure 4 illustrates 3-tesla (fig. 4a) and 1.5-tesla DBS of the STN region is the predominant surgical
(fig. 4b) coregistered coronal MRI images for the same therapy for PD. Accurate surgical targeting is essential for
patient. The RN is visible in both 1.5- and 3-tesla images optimal patient outcomes. With advances in magnetic
(white arrows), while the STN and SN are only visible in field strength, direct visualization of the STN on MRI has
128.122.253.228 - 5/17/2015 4:45:11 PM
NYU Medical Center Library
Evaluating Indirect STN Targeting with Stereotact Funct Neurosurg 2014;92:337–345 343
3-Tesla MRI DOI: 10.1159/000366286
Downloaded by:
Evaluation of RN-Based Targeting Results at 1.5 T Danish et al. [29] reported that 1.5-tesla MRI is inadequate
In this study, we utilized a validated 3-tesla MRI pro- to delineate the relationship between the STN and RN
tocol to evaluate MCP- and RN-based indirect targeting borders. However, at the voxel sizes in each dimension
of the STN. Our aim was to determine and to compare the utilized in that study (>2 mm), the image resolution at the
accuracy and precision of these commonly utilized meth- time of the study may have been inadequate compared to
ods of STN targeting for neurosurgeons lacking high- currently available technology. Confounding results may
field MRI capabilities at their centers. We found that RN- also be MR protocol specific, as the RN may be difficult to
based targeting was both more accurate and more precise visualize at 1.5 T without optimization of scanning param-
than MCP-based targeting at 3 T. An important addition- eters. In our 1.5-tesla MRI protocol, the RN is well visual-
al step was the validation of these results at 1.5 T. RN vi- ized, while both the RN and STN are clearly visualized in
sualization is protocol dependent. We found that the RN our validated 3-tesla imaging protocol. Our comparison
(but not the STN) could be well visualized at 1.5 T (fig. 4) between 1.5- and 3-tesla-based RN border measurements
and that differences in targeting parameters calculated demonstrates that 1.5-tesla MRI is equivalent to 3-tesla
from 3- and 1.5-tesla images were clinically and statisti- MRI in terms of measuring RN borders, and therefore is
cally insignificant. RN-based targeting at 1.5 T produced appropriate for indirect STN targeting. Another historical
the same target coordinates for use during STN DBS sur- criticism of RN-based targeting has been that the param-
gery as targeting at 3 T. eters for targeting were derived from the Schaltenbrand-
Wahren Atlas [17]. The atlas represents one representa-
Comparison with Previous Work on STN Localization tive brain specimen and thus does not capture variability
Our results compare well with STN midpoint coordi- among PD patients [9]. In the present study, validated
nates (11.8 mm lateral, 1.3 mm posterior, 4.0 mm inferior) MRI in a large population of PD patients allowed the
in the Schaltenbrand-Wahren atlas as well as with a previ- quantitative evaluation of targeting uncertainty.
ous study at 1.5-tesla magnetic field strength, which pooled
right and left STN coordinates to reduce statistical vari-
ability [9]. Several previous studies have reported the infe- Conclusions
riority of direct targeting alone with 1.5-tesla MRI com-
pared to a combination of indirect targeting methods [17, RN-based indirect STN targeting is significantly more
25, 26]. However, higher field strength appears to allow for precise than MCP-based STN targeting and accommo-
greater localization accuracy. Several recent studies have dates broad changes in anatomic parameters such as third-
reported that 3-tesla MRI is a more accurate method for ventricle width. Our analysis suggests that neurosurgeons
direct targeting of the STN for DBS surgery [27, 28]. Our employing RN-based indirect targeting may expect more
study extends these findings at 1.5 and 3 T to measure the consistent results than those employing MCP-based indi-
location of the STN midpoint directly at 3-tesla MR field rect targeting at 1.5 T. Future work is required to determine
strength. Significantly, when pooled, our measurements the optimal site of DBS stimulation relative to the STN
for the center of the right and left STNs agree well with the midpoint and to incorporate this information into RN-
earlier measurements of Daniluk et al. [9]. based indirect targeting parameters during DBS surgery.
Evaluating Indirect STN Targeting with Stereotact Funct Neurosurg 2014;92:337–345 345
3-Tesla MRI DOI: 10.1159/000366286
Downloaded by: