A 2-Year Experience of Robot-Assisted Epilepsy Surgery Using a New Cranial Robotic System With Optical Tracking: Variables Influencing Accuracy and Complications

Abstract

BACKGROUND AND OBJECTIVES:

Our institution began using the Globus ExcelsiusGPS cranial robot in 2023 for placement guidance in epilepsy surgeries including depth electrodes for stereoelectroencephalography (SEEG), laser interstitial thermal therapy (LITT), NeuroPace responsive neurostimulator, and Medtronic deep brain stimulator. In our first 2 years, we performed 58 epilepsy surgeries for 37 patients using the robot for 466 trajectories. Surgery time, accuracy of trajectory, and hemorrhage rate were analyzed.

METHODS:

We performed a retrospective chart review of epilepsy surgeries using the Globus ExcelciusGPS cranial robot over our first 2 years of experience. Potential influencing factors analyzed included trajectory length, skull thickness, angle of trajectory with the skull, imaging used for registration, head holder, stabilization of the head, drill used, brain location, type of surgery, the number of each case performed, and the number of trajectories placed. We also analyzed hemorrhage rate and surgery time. Factorial analysis of variance, multiple regression, and Pearson correlation statistics were used.

RESULTS:

Accuracy of electrode or laser placement was significantly affected by imaging modality used for robot registration (P = .018), head holder used (P = .014), and the trajectory plan's angle from skull perpendicular (P < .001). The time required for placement per trajectory varied, with SEEG being the fastest and LITT being the slowest (P < .001). There were no symptomatic hemorrhages in this series, with 3 asymptomatic SEEG surgery subarachnoid hemorrhages (9.7% per surgery, 0.7% per electrode) and 2 asymptomatic LITT surgery parenchymal hemorrhages (15% per surgery, 10% per laser). There were no responsive neurostimulator or deep brain stimulator hemorrhages.

CONCLUSION:

The accuracy of surgery using the Globus ExcelsiusGPS robot was influenced by factors including the trajectory plan and equipment used. The type of surgery performed influenced the amount of surgical time. These findings may improve the efficiency and accuracy of epilepsy surgery using this cranial robot.

ABBREVIATIONS:

CRW

Cosman-Roberts-Wells

LITT

laser interstitial thermal therapy

OR

operating room

PSS

patient stabilization stand

RNS

responsive neurostimulator

SEEG

stereoelectroencephalography.

Assistive technologies, including the use of surgical robots, have risen in use over the past decades to reflect their potential to perform repetitive tasks in a manner that minimizes operative time and patient risk.^1,2 One task that fits these criteria and has been facilitated by robotic aids is the placement of brain electrodes. Robotic depth electrode placement is thought to have similar accuracy to frame-based placement of depth electrodes but to be much faster than using a stereotactic frame.^3-7 Cranial robot technologies have also been used for the placement of electrodes for stimulation devices and lasers for laser ablation of brain regions.^6,8 Intracranial electrode monitoring for epilepsy surgery has been performed for decades, but more recent techniques and technologies have made depth electrodes (stereoelectroencephalography [SEEG]) the preferred method for intracranial electrode monitoring.^4,7,9 Several studies have examined various cranial robotic platforms and their accuracy in the placement of intracranial electrodes to assess the efficacy of robotic assistants in neurosurgery.^10-14 Electrode length¹⁵ and angle of electrode trajectory¹⁶ have been studied for their impact on accuracy with robotic placement.

This study examines the accuracy of the Globus ExcelsiusGPS robot—a relatively new surgical technology—which makes use of optical imaging to eliminate the need for attaching the patient's head to the robot. The Globus ExcelsiusGPS robot (Globus Medical Inc.) was approved by the Food and Drug Administration for cranial use in 2021, and our institution began using the device in 2023. This study offers a retrospective analysis of the robot's use in epilepsy surgery for the placement of NeuroPace responsive neurostimulator electrodes (RNS, NeuroPace Inc.), the placement of depth electrodes for SEEG, the placement of lasers for laser interstitial thermal therapy (LITT), and the placement of Medtronic deep brain stimulator electrodes (DBS; Medtronic PLC). Our institution used the Globus ExcelsiusGPS robot in the guidance of 466 trajectories for epilepsy surgery—part of 58 cranial surgeries across 37 patients—in our first 2 years of its implementation. We analyzed factors that have been shown in other studies to affect the accuracy of electrode placement, including the region of electrode implantation, the entry angle with the skull, the length of the electrode placed intracranially, and the thickness of the skull at the entry location.^17-19 We also analyzed factors specific to this robotic system that have a potential impact on accuracy: the use of the patient stabilization stand (PSS; Globus Medical Inc.); a floor stand that was designed to work with this robotic system to better stabilize the head during surgery; the use of the Mayfield head holder (Integra LifeSciences Corporation) or the Cosman-Roberts-Wells (CRW) frame base (Integra LifeSciences Corporation) attached to the Globus fiducials; the use of the drill designed by Globus for drilling the entry hole with this robotic system compared with use of the Midas Rex drill (Medtronic PLC) or other manufacturers' drill bits; and imaging modality used to register the patient's head for the robot. This system is unique in that it has the option for C-arm registration and, because the patient is not directly attached to the robot, the patient can be moved into the intraoperative computed tomography (CT) during surgery. We evaluated the time required for surgery including room time, surgery time, setup time, and average time per electrode or laser placement to determine which factors influenced the time necessary for surgery.

METHODS

Surgery

Thirty-seven patients with medically refractory epilepsy, defined by 2 or more medications attempted with failure to achieve seizure control, underwent 58 epilepsy surgeries using the Globus ExcelsiusGPS robot for surgical guidance over a 2-year period in an attempt to control their seizures (Table). This included 31 placements of depth electrodes for SEEG procedures, 13 LITT ablations of the epilepsy focus, 12 NeuroPace RNS placements, and 2 Medtronic DBS placements. All of the surgeries were performed by the same surgeon at one institution, Inova Fairfax Hospital.

TABLE.

Patient Characteristics

Variable	No. (%)
	Surgery				Registration			Head holder		Stabilization		Drill
	DBS	RNS	LITT	SEEG	C-arm	O-arm	CT	Mayfield	CRW	PSS	Table	Midas	Globus	Other
Age, y, mean (SD)	22 (4)	41 (12)	42 (13)	41 (14)	40 (20)	39 (12)	42 (14)	41 (13)	40 (14)	41 (14)	40 (13)	37 (14)	41 (15)	42 (13)
Sex
Female	2 (100)	3 (25)	3 (23)	9 (29)	1 (25)	6 (26)	10 (32)	9 (32)	8 (27)	5 (26)	12 (31)	4 (33)	6 (29)	7 (28)
Male	0 (0)	9 (75)	10 (77)	22 (71)	3 (75)	17 (74)	21 (68)	19 (68)	22 (73)	14 (74)	27 (69)	8 (67)	15 (71)	18 (72)
Race/ethnicity
Asian	0 (0)	0 (0)	2 (15)	4 (13)	1 (25)	1 (4)	4 (13)	2 (7)	4 (13)	1 (5)	5 (13)	0 (0)	2 (10)	4 (16)
Black/African American	0 (0)	2 (17)	0 (0)	1 (3)	0 (0)	1 (4)	2 (6)	2 (7)	1 (3)	1 (5)	2 (5)	2 (17)	0 (0)	1 (4)
Hispanic/Latino	0 (0)	2 (17)	5 (38)	8 (26)	0 (0)	7 (30)	8 (26)	8 (29)	7 (23)	4 (21)	11 (28)	1 (8)	9 (43)	5 (20)
Other	0 (0)	0 (0)	1 (8)	1 (3)	0 (0)	2 (9)	0 (0)	1 (4)	1 (3)	1 (5)	1 (3)	0 (0)	2 (10)	0 (0)
White/Caucasian	2 (100)	8 (67)	5 (38)	17 (55)	3 (75)	12 (52)	17 (55)	15 (54)	17 (57)	12 (63)	20 (51)	9 (75)	8 (38)	15 (60)
Seizure type
Focal aware	0 (0)	1 (8)	0 (0)	2 (6)	1 (25)	1 (4)	1 (3)	1 (4)	2 (7)	1 (5)	2 (5)	0 (0)	2 (10)	1 (4)
Focal unaware	0 (0)	6 (50)	5 (38)	15 (48)	0 (0)	11 (48)	15 (48)	12 (43)	14 (47)	10 (53)	16 (41)	5 (42)	6 (29)	15 (60)
Focal unaware with secondary generalization	1 (50)	4 (33)	8 (62)	14 (45)	3 (75)	11 (48)	13 (42)	13 (46)	14 (47)	8 (42)	19 (49)	5 (42)	13 (62)	9 (36)
Primary generalized	1 (50)	1 (8)	0 (0)	0 (0)	0 (0)	0 (0)	2 (6)	2 (7)	0 (0)	0 (0)	2 (5)	2 (17)	0 (0)	0 (0)
MRI findings
None	1 (50)	8 (67)	6 (46)	12 (39)	2 (50)	13 (57)	12 (39)	15 (54)	12 (40)	11 (58)	16 (41)	8 (67)	10 (48)	9 (36)
Hippocampal sclerosis	0 (0)	0 (0)	6 (46)	8 (26)	0 (0)	2 (9)	12 (39)	6 (21)	8 (27)	2 (11)	12 (31)	0 (0)	4 (19)	10 (40)
Other	1 (50)	4 (33)	1 (8)	11 (35)	2 (50)	8 (35)	7 (23)	7 (25)	10 (33)	6 (32)	11 (28)	4 (33)	7 (33)	6 (24)
Seizure focus
Amygdala	0 (0)	1 (8)	0 (0)	1 (3)	1 (25)	0 (0)	1 (3)	1 (4)	1 (3)	0 (0)	2 (5)	1 (8)	1 (5)	0 (0)
Generalized	1 (50)	1 (8)	0 (0)	0 (0)	0(0)	0 (0)	2 (6)	2 (7)	0 (0)	0 (0)	2 (5)	2 (17)	0 (0)	0 (0)
Hippocampus	0 (0)	5 (42)	11 (85)	17 (55)	1 (25)	14 (61)	18 (58)	17 (61)	16 (53)	12 (63)	21 (54)	4 (33)	13 (62)	16 (64)
Neocortex	1 (50)	5 (42)	2 (15)	13 (42)	2 (50)	9 (39)	10 (32)	8 (29)	13 (43)	7 (37)	14 (36)	5 (42)	7 (33)	9 (36)
Side of seizure onset
Bilateral	1 (50)	3 (25)	0 (0)	1 (3)	0 (0)	1 (4)	4 (13)	4 (14)	1 (3)	1 (5)	4 (10)	3 (25)	1 (5)	1 (4)
Left	1 (50)	6 (50)	7 (54)	18 (58)	2 (50)	13 (57)	17 (55)	14 (50)	18 (60)	11 (57)	21 (54)	6 (50)	11 (52)	15 (60)
Right	0 (0)	3 (25)	6 (46)	12 (39)	2 (50)	9 (39)	10 (32)	10 (36)	11 (37)	7 (37)	14 (36)	3 (25)	9 (43)	9 (36)
Target variance, mean mm (SD)	0.9 (0.5)	2.0 (1.3)	2.1 (1.2)	2.5 (1.6)	1.7 (0.9)	2.4 (1.6)	2.8 (1.8)	2.0 (1.4)	2.5 (1.6)	2.2 (1.5)	2.7 (1.7)	1.7 (1.2)	2.1 (1.3)	2.8 (1.8)

CRW, Cosman-Roberts-Wells; DBS, deep brain stimulator; LITT, laser interstitial thermal therapy; PSS, patient stabilization stand; RNS, responsive neurostimulator; SEEG, stereoelectroencephalography; Table, operating room table.

Percentages may not sum to 100% due to rounding.

The surgeries were performed as follows. Before surgery, each patient had a stereotactic MRI with gadolinium performed. The gadolinium allowed for visualization of the brain vasculature, which could be avoided with the surgical plan. The MRI was loaded into the Globus planning software program, which was then used to plan each trajectory for the surgery. The number of trajectories per surgery varied from 1 to 20, with a total of 466 trajectories across the 58 surgeries. All electrodes and lasers were placed under general anesthesia. Each patient was placed in either the CRW frame base or the Mayfield head holder, which was attached to the operating room (OR) table or the PSS. Intraoperative CT, O-arm, or C-arm was used to register the patient's head to the stereotactic MRI plan on the robotic system. The robot was used to guide the placement of each electrode or laser. The power twist drill—either the Globus drill or another twist drill type—was used to drill the hole for placement of the anchor bolt and electrode or laser for SEEG and LITT surgery. The Midas drill was used to drill a burr hole for placement of the electrode in RNS and DBS surgery. After surgery, stereotactic CT imaging was performed to verify the location of the electrodes or laser. This procedure was done in the OR for LITT, RNS, and DBS using the intraoperative CT or O-arm and in the postoperative period for SEEG. These CT studies were fused with the MRI plan in the Globus planning software program to assess the accuracy of placement compared with the plan. After assessment of the placement, 3 lasers were repositioned, but none of the SEEG, RNS, or DBS electrodes required repositioning.

Data Collection

This study protocol was reviewed and approved by the institutional review board with exemption from written informed consent because it is a retrospective study of deidentified clinical data. To determine the robot's accuracy and factors influencing accuracy, we collected numerous variables from the imaging, including skull thickness, angle with the skull perpendicular at entry location, brain location, and trajectory length. This was done by fusing postoperative or intraoperative CT with the MRI plan to determine accuracy using the axial deviation of the electrode's center or laser's tip from the center of the planned trajectory's tip. We also collected surgical data on factors that could influence accuracy, including type of surgery, imaging modality for registration, drill type, head holder, and stabilization of the head with the Globus PSS or attachment to the OR table. The chart review was used to collect additional data for analysis, including number of each case performed, number of trajectories, room time, surgical time, setup time, and calculated time for placement of each electrode or laser. Average time for each electrode placed was calculated by dividing total surgical time by the number of electrodes placed, so this value included the time required for placement of the generator in NeuroPace RNS surgery and the time required for repositioning in the few stimulator or LITT surgeries in which repositioning was necessary. We also collected data on hemorrhage rate.

Data Analysis

The data were analyzed using IBM SPSS software (IBM) on a Dell DESKTOP-7NIJA7V computer (Dell Inc.). The statistics performed to determine which factors influenced electrode placement accuracy were factorial analysis of variance and multiple regression. The statistics used for the analysis of OR time included factorial analysis of variance, multiple regression, and Pearson correlation. P < .05 was used for statistical significance.

RESULTS

The Globus ExcelsiusGPS robot's overall mean radial variance from trajectory was 2.4 mm. The factors that influenced the accuracy of electrode or laser placement included the modality of imaging for registration with the robotic system (P = .018, Figure 1) and head holder used for head stabilization (P = .014, Figure 2, CRW variance from Mayfield 0.481 mm, 95% CI: 0.108-0.853). The type of drill used, the location in the brain of the electrode or laser placement, the base used for stabilization of the head, and the type of surgery performed did not influence the accuracy of electrode or laser placement (P > .10 for all). Post hoc analysis demonstrated a statistically significant difference in accuracy of the imaging modality used for registration (P < .05 for all) with the C-arm registration being the most accurate, followed by O-arm registration (variance from C-arm 0.632 mm, 95% CI: 0.126-1.139), then intraoperative CT registration as the least accurate (variance from C-arm 1.034 mm, 95% CI: 0.510-1.559). For the head holder, the Mayfield provided greater accuracy of placement than the CRW frame base with attached Globus fiducials.

FIGURE 1.

Axial variation of electrode or laser from planned trajectory based on imaging modality used to register the patient to the robotic system. There were statistically significant differences between C-arm and O-arm (P = .009), C-arm and CT (P < .001), and O-arm and CT (P = .025). CT, computed tomography.

FIGURE 2.

Difference in electrode or laser variance from plan for the different head holders used in surgery (P = .014).

Concerning anatomic factors, angle from perpendicular with the skull was correlated with accuracy of placement (P < .001, Figure 3), where the accuracy of the electrode or laser placement decreased as the angle of entry became less perpendicular to the skull. Skull thickness and trajectory length of electrode or laser were not correlated with accuracy (P > .10).

FIGURE 3.

Positive correlation between increased angle from perpendicular with the skull and increased electrode or laser variance from the plan, showing an increase in inaccuracy with increased angle with the skull (P < .001).

Surgical factors influenced OR time as well, with type of surgery influencing average time per electrode or laser placed (P < .001, Figure 4), but not setup time, surgery time, or overall room time (P > .05). SEEG had the lowest time per electrode placed at 18 minutes per electrode, followed by DBS (variance from SEEG 63.8 minutes, 95% CI: 34.9-92.7), then RNS (variance from SEEG 102.4 minutes, 95% CI: 88.9-115.8), and finally LITT (variance from SEEG 176.9 minutes, 95% CI: 163.8-190.0). Imaging modality used for registration and drill used for surgery did not influence any of the measures of time in the OR (P > .10). Surgery time and room time were not correlated with any of the variables measured, but setup time was inversely correlated with case number (P = .005, Figure 5)—with setup time decreasing as more surgeries were performed—and average electrode time was inversely correlated with number of trajectories (P < .001, Figure 6)—surgeries with more electrodes requiring less time per electrode.

FIGURE 4.

Difference in time per electrode or laser placed based on type of procedure performed (P < .001). DBS, deep brain stimulator; LITT, laser interstitial thermal therapy; RNS, responsive neurostimulator; SEEG, stereoelectroencephalography.

FIGURE 5.

Negative correlation between surgery case number and the time required for surgical setup, indicating that the setup and registration of the robotic system became faster with increased surgical experience with the system (P = .005).

FIGURE 6.

Average time required per electrode or laser placement based on the number of electrodes or lasers placed at the time of surgery, showing a negative correlation, which indicates that the time required for placement of each electrode or laser is less for surgeries involving more trajectories (P < .001).

Complication rate was low, with a total of 6 hemorrhages, 3 of which were minimal subarachnoid hemorrhage and none of which were symptomatic. Three hemorrhages occurred in SEEG surgeries, putting the incidence at 9.7% per surgery and 0.7% per electrode. Two hemorrhages took place during LITT surgery, with an incidence of 15% per surgery and 10% per laser. There were no hemorrhages during placement of RNS or DBS electrodes.

DISCUSSION

Depending on the technique used, stereotactic frame electrode placement—the gold standard for accuracy in electrode placement—has a reported variance from plan ranging from 1.77 to 2.69 mm.²⁰ This range places the accuracy of the Globus ExcelsiusGPS robot, with a variance from plan of 2.4 mm, within the range of the stereotactic frame accuracy from published studies. Furthermore, the overall mean variance from trajectory was similar to the 2.13-mm reported accuracy error of other robotic systems.²¹ This variance was slightly higher than the 1.9-mm radial error reported for the Globus ExcelsiusGPS robot in an earlier study of SEEG electrode placement.²² To measure the axial variance from the plan, we fused the CT containing the implanted electrodes or lasers with the MRI plan. The error in fusing CT to MRI is reported to be 1.25 mm,²³ adding an unavoidable error as part of the assessment.

It is not surprising that an increased angle of trajectory with the skull decreased electrode accuracy as greater angles produced more dramatic skiving effects: skiving refers to the drill's horizontal movement upon contact with the skull before burrowing.²⁴ This is consistent with other studies that have found that the angle of entry with the skull affects the accuracy of electrode placement.^17,18 However, the findings of these studies that factors including electrode length, region of implantation, and bone thickness affected electrode accuracy were not replicated in our study. We expected to find that the accuracy of electrode placement differed with varying drill types because the Midas drill produced a larger burr hole, preventing the skull and dura from deflecting the trajectory; additional features of the Globus drill increase accuracy, including a drill guide, a thicker drill shaft to prevent bending, and spikes at the end of the drill guide to prevent it from deviation with the skull, but these factors did not have an influence on the accuracy of placement. We also did not find that location influenced the accuracy of placement, which is inconsistent with the findings of a study on factors influencing accuracy in the placement of DBS electrodes.¹⁹

Factors that we found to influence accuracy were the imaging modality used for registering the patient to the robotic system and the head frame used for stabilizing the head during surgery. These factors have been discussed as potential influences on accuracy in a literature review of robotic stereotactic placement of electrodes.²⁴ A novel feature of the Globus ExcelsiusGPS robot is its ability to be registered using a C-arm rather than axial imaging like an O-arm or intraoperative CT. This feature comes from its adaptation from the Globus ExcelsiusGPS spine robot, which predates the cranial robot. The potential advantages of C-arm registration are a decrease in registration time, a decrease in radiation exposure, and potentially lower barriers for use of the technology in facilities without an intraoperative CT or O-arm available for registration. Our findings indicate that C-arm registration is at least as good as, if not superior to, registration with axial imaging modalities. Both the Mayfield head holder and the CRW frame base with attached Globus fiducials are rigidly affixed to the skull with pins, so a difference in accuracy between these 2 devices cannot be from head movement in relation to the head holder. We have considered that the difference in accuracy that we observed may be related to the fiducials used for each type of head holder: a close set of 3 fiducials for the Mayfield and a more widely placed set of 5 fiducials for the Globus fiducial system attached to the CRW frame, which are perpendicular to the plane of the robot end effector for placement of lateral electrodes.

Limitations

We expected a difference in accuracy between DBS, RNS, LITT, and SEEG because DBS and RNS electrodes are placed through wider burr holes that allow the elimination of the skull and dura's influence. Nevertheless, we found no statistically significant difference in accuracy. We did find a difference in average time per electrode or laser placement, in which SEEG had the lowest time per electrode placement, followed by DBS, then RNS, and finally LITT. The time per SEEG electrode is consistent with the time per electrode reported in the literature²¹ and may be much lower than the other procedures because surgical time was divided by number of electrodes, leading to lower times with more electrodes. RNS likely has a higher time per electrode than DBS because the generator was placed at the time of electrode placement. The highest time per trajectory for LITT may be a result of the need to reposition some of the lasers, mostly due to one surgery in which, unbeknownst to the surgeons, the patient moved in the Mayfield, requiring complete reregistration after a repositioning attempt.

Compared with other reports on electrode placement in the literature, symptomatic complication rate was comparably low.^20,21,25 We found a 0% symptomatic hemorrhage rate, with a 9.7% minimal subarachnoid hemorrhage rate per SEEG surgery and 15% rate of small parenchymal hemorrhage per LITT surgery. There were no hemorrhages with DBS or RNS. This contrasts the reported symptomatic complication rate of 1.15%, 2.4%, and 5.6% for SEEG. The 15% LITT asymptomatic hemorrhage rate is higher than the 2.8% clinically significant LITT hemorrhage rate reported in the literature,²⁶ which may be due to the use of the non-Globus drill bit without the drill guide, which was more likely to skive on the skull, potentially changing entry point from the planned entry point and involving cortical vessels.

CONCLUSION

This study has demonstrated the efficacy of the Globus ExcelsiusGPS robotic system in the placement of electrodes for SEEG, DBS, and RNS, as well as the placement of lasers for LITT. It is important to note that these findings can be generalized, and that the use of robotic systems such as the Globus ExcelsiusGPS has the potential to enhance the speed and proficiency of electrode and laser placement. This study, therefore, might be used by other institutions and surgeons to increase performance and accuracy with the implementation of the Globus ExcelsiusGPS. Furthermore, the accuracy afforded by the varying of robotic implements—the differing imaging systems used for registration, for example—might be used to enhance the design of current and future robotic systems. Nevertheless, the differences in electrode and laser placement accuracy seen with the different surgical factors evaluated in this study ranged from half a millimeter to one millimeter. Given such a small difference in accuracy, factors such as the ability to move the bed during surgery may be important enough to accept such a small increase in inaccuracy. The findings of this study may help guide surgeons in finding their own personal workflow for using cranial robotics to assist in the placement of electrodes and lasers during surgery.

Funding

This study did not receive any funding or financial support.

Disclosures

The authors have no personal, financial, or institutional interest in any of the drugs, materials, or devices described in this article.