An Integrated Robotic System for MRI-Guided Neuroablation: Preclinical Evaluation

Abstract

Objective:

Treatment of brain tumors requires high precision in order to ensure sufficient treatment while minimizing damage to surrounding healthy tissue. Ablation of such tumors using needle-based therapeutic ultrasound (NBTU) under real-time magnetic resonance imaging (MRI) can fulfill this need. However, the constrained space and strong magnetic field in the MRI bore restricts patient access limiting precise placement of the NBTU ablation tool. A surgical robot compatible with use inside the bore of an MRI scanner can alleviate these challenges.

Methods:

We present preclinical trials of a robotic system for NBTU ablation of brain tumors under real-time MRI guidance. The system comprises of an updated robotic manipulator and corresponding control electronics, the NBTU ablation system and applications for planning, navigation and monitoring of the system.

Results:

The robotic system had a mean translational and rotational accuracy of 1.39±0.64mm and 1.27±0.56° in gelatin phantoms and 3.13±1.41mm and 5.58±3.59° in 10 porcine trials while causing a maximum reduction in signal to noise ratio (SNR) of 10.3%.

Conclusion:

The integrated robotic system can place NBTU ablator at a desired target location in porcine brain and monitor the ablation in realtime via magnetic resonance thermal imaging (MRTI).

Significance:

Further optimization of this system could result in a clinically viable system for use in human trials for various diagnostic or therapeutic neurosurgical interventions.

I. Introduction

An estimated 16,700 people in the United States died from primary brain and nervous system cancers in 2017 [1], a disease with a 5 year survival rate of 33.2% [2]. Additionally, 9%−17% of all cancers that metastasize result in secondary tumors in the brain [3]. Treatment options for tumors in the brain include craniotomy and/or radiation therapy; however, craniotomy is a highly invasive procedure leading to longer patient recovery period, while radiation therapy often causes damage to the healthy tissue around the tumor and results in deteriorated quality of life. In recent years, minimally invasive stereotactic neurosurgery approaches have been explored.

Precise placement of a needle or a probe for therapeutic or diagnostic purposes is essential for minimally invasive stereotactic neurosurgery procedures such as Deep Brain Stimulation (DBS) electrode placement and tumor ablation. These procedures are often performed under Computed Tomography (CT) guidance using a mechanical stereotactic frame such as the Leksell Stereotactic System ® (Elekta AB, Stockholm, Sweden). In a typical procedure, preoperative MR images acquired prior to the surgery are registered to the intraoperative CT images [4]. The surgical device is then adjusted manually by setting entry angles and insertion depth based on the aforementioned images. Such manual adjustments are time consuming and prone to human error, which could result in imprecise tool placement. A robotic system can be used in place of manual fixtures to reduce said error. NeuroArms [5], NeuroMate (Integrated Surgical Systems, CA) [6], PathFinder (Armstrong Healthcare Ltd., UK) [7], Renaissance (Mazor Robotics, Israel), and Rosa (Medtech Surgical Inc., France) [8] are some of the integrated robotic systems providing tool placement using preoperative CT/MR images. However, as these robotic systems rely on preoperative image guidance, they cannot account for shifts in the target location between preoperative imaging and surgery. Performing such procedures under intraoperative image guidance can address these issues.

MRI’s better soft tissue contrast and absence of ionizing radiation compared to CT make it well suited for neurosurgery procedures. MRI-guided robot-assisted minimally invasive procedures have been studied extensively, however the constrained bore and strong magnetic field of the MRI environment still poses challenges for performing an in-bore robot-assisted procedure. MRI-guided robotic systems have previously been demonstrated for prostate biopsy [9]–[17], breast biopsy [18], [19], shoulder arthrography [20], [21] and abdominal interventions [22], [23]. Masamune et al. presented the first needle insertion manipulator for stereotactic neurosurgery under MRI guidance [24]. Raoufi et al. and Comber et al. developed pneumatically actuated robotic systems for MRI-guided neurosurgery [25], [26]. Ho et al. developed an SMA-actuated MRI-compatible neurosurgical robot [27]. Monfaradi et al provides a review of MRI robots for needle based interventions [28]. In recent years, interventional MRI guided neurosurgery systems [29] and laser interstitial thermal therapy (LITT) for treatment of brain tumors [30] have been explored. Intraoperative MRI guidance and semi-actuated stereotactic devices such as the smart frame ClearPoint™ (MRI Interventions, United States) have shown potential for performing in-bore minimally invasive neurosurgery procedures. Laser interstitial ablation system NeuroBlate™ (Monteris Inc, United States) enables real-time MRI guided brain tumor ablation using a 2 degrees of freedom (DOF) cable actuated needle manipulation device which adjusts needle depth and rotation. However, a fully actuated robotic system which can control both the trajectory as well as insertion depth and rotation can lead to more accurate placement of the ablator. We have previously demonstrated prototypes of actuated stereotactic manipulators for neurosurgery [31], [32].

We present a robotic system integrated with NBTU ablation system designed for use in clinical environments to deliver closed-loop brain tumor ablation. Design aspects such as safety, sterility and minimization of changes to the clinical work-flow are considered. The robotic system is entirely made from materials suitable for MRI environment. The NBTU ablator integrated with the robotic system can provide the thermal dosage required to ablate a tumor. The contributions of the presented work include a pre-clinically tested robotic manipulator, integration of the robotic manipulator with the MRI system, usage of an NBTU ablation system, and demonstration of clinical viability of the robot through preliminary in-vivo porcine studies.

II. System Overview

The system presented herein is developed to deliver real-time MRI guided thermal dosage to targeted brain tumors. The robot provides a motorized embodiment of a stereotactic frame and provides additional DOF to precisely place the ablation device at the target location. The general concept and early prototype was originally described in [32]. The system comprises of the following modules: Robotic stereotactic manipulator, real-time Linux based robot controller, software subsystem providing robot control, surgical planning and navigation, NBTU ablation system and real-time scanner control and thermometry module. Fig. 1 shows typical system setup for an animal study.

Fig. 1:

Integrated system setup: In the control console room: (1) Ablation system, (2) MRI control console, (3) surgical navigation user interface, and robot control application, (4) patch panel, and (5) network interface box with fiber optic connector. Inside the scanner room: (6) patch panel through which scanner console and components inside the MR room are connected, (7) MRI compatible robot controller, (8) custom cable carrying motor and encoder signals, (9) robotic manipulator inside the scanner bore, (10) fiber optic cable for communication between robot control application and the embedded robot controller, (11) custom MR imaging coil, and (12) foot-switch to enable/disable robot motion.

A. Robotic Stereotactic Manipulator

The system is comprised of a 7-DOF robot with remote center of motion (RCM) mechanism. As shown in the Fig. 2, the robot has 3 modules providing: (1) three-DOF Cartesian stage for RCM position, (2) two DOF rotations for orienting the ablation probe to the desired trajectory, and (3) two DOF for probe rotation and insertion. In our previous prototypes [31], [32], we demonstrated the functional requirements of such a robotic device. The robotic system presented here is designed for clinical applications with improved mechanisms machined from polyetherimide plastics for better rigidity and accuracy. The robot base contains an embedded registration frame used to localize the robot base in the MRI coordinate system. Provisions for maintaining the sterile field are shown in Fig. 3. Components of the robot that operate within the sterile field were designed to be suitable for ethylene oxide gas and desiccation sterilization. Other robot components are separated from the sterile field by a drape.

Fig. 2:

CAD render of clinically optimized neurosurgery robot showing all three modules of the manipulator along with the robot registration frame and the coordinate system.

Fig. 3:

(a) CAD render of the needle driver assembly showing sterilizable parts and parts covered in a sterile drape during the procedure. During surgery, a sterile bag containing the probe assembly, drive gear, cannula depth stop, and cannula is provided to the surgeon. All sterile components are assembled and attached over the sterile drape and (b) actual ablation probe with (1) ablation system connector, (2) water circulation port, (3) case for the ablation probe, (4) two 360 degree ablation elements, (5) connector for tracking coils, and (6) tracking coils.

All three cartesian axes are driven by piezoelectric ultrasonic motors (USR30-NM, Fukoku Co., Ageo-shi, Japan). The joint positions are measured by quadrature encoders (EM1-0-500-N, US Digital, Vancouver, Washington) with a resolution of 500 lines per inch (0.013mm/count). The rotations are actuated by piezoelectric ultrasonic motors (USR45-NM, Fukoku Co. and USR30-S4N, Shinsei Corp., Tokyo, Japan). Both rotation angles are measured directly by quadrature encoders (EM1-0-1250-N, US Digital) with 1250 counts per rotation (CPR) resolution (0.072°/count) placed after their gear reductions to avoid measurement errors related to backlash. Probe insertion and rotation are implemented in the needle driver shown in Fig. 2. Both axes are driven by ultrasonic motors (USR30-S4N, Shinsei Corp.). Probe insertion is measured by a reflective linear encoder with 300 lines per inch resolution (0.021mm/count) and probe rotation is measure using a 1250 CPR rotary encoder (EM1-0-1250-N, US Digital). A detailed overview of the mechanics was presented in [33].

The manipulator has been evaluated in bench-top experiments using an optical tracking system, giving the probe placement accuracy of 1.37±0.06mm and 0.79° ± 0.41°, probe insertion accuracy of 0.06±0.07mm and probe rotation accuracy of 0.77° ± 1.31° [33]. It has also been evaluated for compatibility with the MRI environment giving no visual image distortion and SNR loss of less than 10.3% [33], to the best of our knowledge this is the lowest reported SNR loss using piezoelectric actuation in an MRI environment.

B. Robot Control System

MRI machines can be affected by radio frequency noise emitted from electronic devices. Shielding and filtering were implemented to minimize noise from the robot control system. The system presented here is inspired by our previous work, which reported SNR loss of less than 16% [34]. The new control system consists of a backplane, a low noise ultrasonic motor control module for each axis, an LED based axis status reporting module providing visual feedback, and safety interface for emergency stop. The full system can be observed in Fig. 1 item 7 and Fig. 4 with and without the motor control modules. The backplane is built around an sbRIO-9651 (National Instruments, United States) system on module (SOM) using a Xilinx Zynq chip and runs real-time Linux. FPGA based Serial Peripheral Interface, developed using National Instrument’s LabView^TM, is used for 1 KHz communication from the centralized real-time SOM controller and the motor control modules. TCP/IP based communication interface was developed to exchange data between the robot control application and the real-time robot controller. Physical connection between the robot control workstation and the robot controller is accomplished using a fiber optic ethernet to eliminate any noise that could pass through the patch panel. The control system when powered ON inside the MRI room does not cause a statistically different SNR when compared to the controller powered OFF (p=0.714) [35].

Fig. 4:

Top view of the control box with motor control modules (top) and without motor control modules (bottom). The power supplies are seen in (a), backplane (b), a power distribution board (c) and breakout board (d) which consolidates the outputs of each motor control module into a single cable.

The Control system is optimized for clinical usage by providing multi-level safety, improved user feedback and single cable connection to the robot for simpler system setup as shown in Fig. 1 item 8. Three level safety mechanism is implemented: (1) High-level robot control application provides graphical user interface to monitor system, (2) low-level firmware monitors a foot-switch interlock and blocks any motion unless it is engaged and (3) a hardware kill switch physically kills the motor power while maintaining logic power to preserve robot status. LED based robot status information is provided to the physician to convey the robot motion status. Custom developed 151 conductor cable with self aligning connectors (HD 38999, Amphenol, Connecticut, USA), with ferrous ratchet spring removed, are implemented for connection between the controller and the robot for repeatable and safe installation.

C. Needle-based Therapeutic Ultrasound Applicator

The robotic system is designed to hold, position and orient the NBTU applicator to a desired trajectory. The NBTU applicator can produce different heating patterns by employing ablation elements with varying size and ablation directionality. In the animal studies presented in this paper, a TheraVision^TM ultrasound ablation system and ACOUSTx^TM applicators with multiple, cylindrical, MRI-compatible transducers [36] are used. Aforementioned systems are developed by our collaborators at Acoustic Medsystems Inc (Savoy, Illinois, United States). As shown in Fig. 3(b), ACOUSTx^TM applicator has: (1) one or more ablation elements, (2) sheath made from bio-compatible plastic for water circulation which is required for acoustic coupling and cooling, (3) connectors for water circulation lines, (4) connector for electrical signals to ablation elements, and (5) connector for active tracking coils. The active tracking coils were incorporated for future use to increase the speed, accuracy, and robustness of determining the NBTU applicator’s position and orientation.

D. Surgical Planning and Navigation

Planning and navigation software provide an interface between clinicians and the surgical robot platform. The software allows for visualization of pre-operative and intra-operative scans, the surgical plan, and tracking of real-time changes during the procedure. Three applications work cooperatively in the presented surgical robot system: 3D Slicer [37], TheraVision^TM, and the robot control application.

1. 3D Slicer: An open source surgical planning and navigation application was used for preoperative planning and visualization. The robot workspace is overlayed on intraoperative images to convey the reachable workspace of the robot. The surgeon selects the desired target and entry points using 3 views as displayed in Fig. 5.

2. TheraVision^TM: A surgical planning and navigation application with FDA approval for use in soft tissue. It has four modules related to brain tumor ablation procedures:

   • NBTU ablation probe controller: Consists of embedded controller modules to power the NBTU ablator with required electrical signals.
   • Ablation control application: Allows execution of the desired ablation plan by controlling the produced acoustic power and duration of treatment.
   • Robot registration module: Uses MR images to calculate the transformation between the image space and the robot coordinate system using the line marker registration method [38]. The registration algorithm detects nine cylindrical markers configured in three planar Z-shapes for calculating the 6-DOF pose of the robot base in scanner coordinate system.
   • OpenIGTLink [39] based robot control module: Communicates with the robot control application to enforce the clinical workflow shown in the Fig. 7.

3. Robot Control Application: Provides joint space and task space control of the robotic manipulator. It implements the robot initialization and homing procedures which are required in order to move the robot into a known position after power up. The procedure includes moving each axis until reaching the optical limit switch installed on the respective axis and, automatically, saving the incremental encoder count at that position. The robot control application provides real-time status and 3D visualization of the robot for the clinician. It also provides an interface for communication with 3D Slicer to send the registration transform and current robot pose over OpenIGTLink.

Fig. 5:

3D Slicer user interface for brain tumor ablation planning: showing intraoperative MR images, overlay of robot workspaces, ablation probe and delivered thermal dosage during one of the porcine study.

Fig. 7:

Surgical workflow showing planning, targeting and ablation tasks

E. MRI Based Thermal Dose Monitoring

Thermal dosage monitoring in real-time helps in ensuring the shape and size of a desired ablation lesion. MR imaging is used for noninvasive temperature monitoring by measuring the shift in Proton Resonance Frequency (PRF) through the use of phase images [40]. Temperature change between two consecutive phase images is calculated by multiplying the phase difference by a scaling factor, as shown in Eq. 1.

$$\Delta T=\frac{\Delta \varphi }{\gamma \alpha B_0{T}_E}\mathit{\text{where}},$$ (1)

Δϕ = the phase difference between consecutive images,

ΔT = scaled temperature difference,

α = temperature dependent water resonance chemical shift − 0.0094ppm/°C,

γ = gyromagnetic ratio − 42.58 MHz/T,

B₀ = magnetic field strength,

T_E = echo time

Temperature difference between pairs of consecutive images is summed over the ablation procedure period and added to the starting temperature to yield the absolute temperature. Accumulated temperature over a time period determines cell survival. Sapereto et al. developed the concept of Thermal Dose (TD) and demonstrated that a temperature of 43°C caused a clearly observable tissue necrosis after 240 minutes [41]. Therefore, the TD at which the tissue necrosis occurs is defined as 240 cumulative equivalent minutes (CEM43). Using this method any time and temperature data could be converted to an equivalent minutes of heating at 43 °C. In Eq. 2, the resulting CEM43 value represents the effect of the time-temperature history on cell necrosis.

$$CEM43=\sum _{i=1}^n{t}_i{R}^{43-T_i}\mathit{\text{where}},$$ (2)

$$\begin{array}{l}\mathit{\text{CEM}}43\text{\hspace{0.17em}is the cumulative number of equivalent\hspace{0.17em}minutes at\hspace{0.17em}}43°C,\\ t_i\mathit{\text{is the }}{i}^{th}\text{\hspace{0.17em}time interval},\\ T_i\mathit{\text{is the average temperature during time interval }}{t}_i\text{\hspace{0.17em}and\hspace{0.17em}}\left(R\left(T<43°C\right)=0.25,R\left(T>43°C\right)=0.5\right)\end{array}$$ (3)

A MATLAB^TM (Mathworks Inc., Natick, MA, USA) application (MatMR-TD Monitor) is developed to perform real-time temperature monitoring and thermal dose calculations. The MATLAB interface for Philips scanners, MatMRI [42] is used to communicate with the MRI scanner to align the scan-plane of MR Thermometry (Table I) images to be perpendicular to the NBTU ablation elements and receive those images in real-time for thermal dose calculations.

TABLE I:

MR imaging protocols for the phantom and animal studies

Imaging Protocol	Scan Sequence	Flip Angle (deg)	TR (ms)	TE (ms)	Slice Thickness (mm)	Slice Spacing (mm)	Pixel Spacing (mm × mm)
Localizer	T2W-TSE/MS	90	3000	90	3	3	0.47 × 0.47
Plan/Confirmation	T1-FFE/3D	15	10	2	2	1	0.44 × 0.44
MR Thermometry	FFE-EPI	20	39	15	7	0	1.39 × 1.39
Post Ablation	DWI-SE	90	3700	76	4	0	0.90 × 0.90

F. System Integration

The system architecture and data flow between the robotic manipulator, NBTU ablation system and MR imaging system are depicted in Fig. 6. Fig. 1 shows the typical system setup for a brain tumor ablation procedure using the robotic system.

Fig. 6:

System integration diagram showing layout of system components and communication interfaces between them.

The robotic manipulator is registered to the scanner coordinate system using TheraVision. The transform is sent to the robot control application over OpenIGTLink. The surgical plan is prepared on 3D Slicer by the neurosurgeon using MR images taken after registration. The plan includes desired target and entry locations for the NBTU. TheraVision receives the target and entry points and superimposes them on the intraoperative MR images. It then computes the target pose (4×4 homogeneous transform) and sends it to the robot control application. The robot control application performs the inverse kinematics and sends the joint level commands to the robot controller. The neurosurgeon engages the foot-pedal to activate the robot motion and the robot aligns all axes except insertion to the desired target pose. Surgeon attaches a cannula to the robot penetrating the burr hole. The probe insertion is performed after the NBTU applicator is attached to the robot. Once the probe is inserted to the desired depth, a confirmation MR image-set is acquired. After which, a scan plane perpendicular to the probe path and passing through the ablation elements is set using the MatMR-TD monitor application. The ablation device is powered ON and the MR-thermometry module starts temperature and thermal dosage monitoring and displays it in real-time at an update rate of 2.78 seconds. Once a predetermined treatment time has elapsed, the ablation device is manually powered OFF, while the neurosurgeon monitored the delivered thermal dosage throughout the ablation delivery. At the end, NBTU is extracted and detached from the robot. Fig 7 shows the complete surgical workflow.

Sterilization is important for in-vivo survival studies. Desiccation sterilization is chosen for the animal studies which could be substituted by other procedures for human trials. As shown in Fig. 3, the ablation probe assembly, probe rotation gear, cannula, cannula depth stop, and thumb screws for probe and cannula attachment are sterilized before every surgical procedure. These parts are put in a sterile bag and left in a desiccation chamber for 48 hours before the surgery. The steps for attaching these sterile components to maintain the sterile field are described in Section III-B.

III. Experimental Setup

Demonstration of the accuracy and usability of the integrated system was first assessed in an MRI with gelatin phantom experiments. After which, the system was further assessed with 10 porcine studies. The porcine studies explored the overall accuracy of the system within a full surgical environment. The effects of different power and time settings were also explored for the NBTU applicator for in-vivo brain tissue.

A. In-Phantom Accuracy Assessment in MRI

Phantom studies were performed under live MRI guidance using a 3T Achieva scanner (Philips, The Netherlands) to evaluate the targeting accuracy of the system. A 3D printed square phantom 90mm × 90mm × 60mm with a 45 mm homogeneous region, a 15 mm tall grid composed of 12.5 mm squares and an laser cut acrylic plate with five 5 mm diameter circular holes (white circles in Fig. 9) in the lower portion was filled with a mixture of 70 g KNOX Original Unflavored Gelatin (Kraft Foods, USA) and 340 mL of water. The phantom was placed at a 45° angle inside the MRI. Fig. 8 shows the experimental setup. An NBTU applicator was robotically aligned and then inserted into the phantom. A confirmation image-set was collected using the 3D-FFE imaging protocol shown in Table I. The aforementioned experimental sequence was repeated for 5 orientations in each of 5 target locations predefined on a planning image set. A total of 25 insertions were collected to assess the system accuracy. The experimental setup was designed as a mock-up of a typical clinical procedure in which the robot is registered only once and then used to ablate multiple target regions.

Fig. 9:

Left: gelatin (green) filled 3D printed phantom (white) and acrylic plate with 5 holes (black), the targets and segmented five needle trajectories for target T5, Right: 2D view showing targets T1-T5 (black cross) within the while circles (holes in the acrylic plate) and five measured tip positions (dots) for each target.

Fig. 8:

Experiment setup showing the robot attached on the scanner bed and a gelatin phantom for accuracy assessment.

The desired target positions defined in 3D Slicer were compared with actual probe tip positions (manually segmented from MRI confirmation images) to assess translational targeting accuracy. Orientation targeting accuracy was accessed by manually selecting approximately 40 points along the probe trajectory in the confirmation images and then fitting a line to compare the achieved trajectory with the planned one. This process was repeated for all 25 insertions.

B. Animal Studies

Upon approval of UMass Medical School’s Institutional Animal Care and Use Committee (IACUC), animal studies were performed to evaluate the integrated system performance in swine model. Animal was prepared in a sterile room adjacent to the MRI suite. A neurosurgeon created the burr hole, filled it with the sterile gelatin with gadolinium to fill the void for better localization during imaging, and then stitched it. Before the animal was brought in the MR room, the robot base platform was attached to the scanner bed and registered to the scanner coordinate system. Then the animal was laid on the scanner bed in head first prone position with appropriate life support systems. Robot was attached to the robot base and a planning image-set (see Table I) was acquired. Using 3D Slicer, a surgical plan consisting of the target and entry locations was determined by a neurosurgeon based on the planning images-set. The robot then aligns to the planned trajectory when the foot pedal was engaged by the neurosurgeon. Once the robot was at the desired orientation and position, it was covered in a sterile drape and a small cutout was made to attach the cannula depth-stop shown in Fig. 3. The burr hole was exposed and a sterile cannula was inserted to guide the probe. Once the cannula was locked in place using the thumbscrew, sterile ablation probe was attached to the needle driver module shown in Fig. 3. The clinician engaged the foot pedal to robotically insert the ablation probe to the previously defined insertion depth. A confirmation image-set (see Table I) was acquired to confirm that the probe was inserted to the desired target location. The scan plane’s position and orientation for MR thermometry was set to a plane perpendicular to the probe axis intersecting the ablation element which is at a known distance form the tip. The scan plane geometry is determined using the robot pose calculated by the robot control application. The ablation element was powered ON at a power setting determined by evaluating the ablation results from the previous studies. The real-time temperature change was observed using MR thermometry as shown in Fig. 10. The ablation element was powered OFF when a predetermined amount of time had elapsed, while the neurosurgeon monitored delivered MR thermal dosage and size of the ablation region in real-time. At the end, the ablation probe was extracted and the incision was stitched.

Fig. 10:

Representative thermal dosage monitoring for study 5 at 0, 61.08, 122.16 and 180.46 Sec., Top: temperature change, middle: delivered thermal dosage and bottom: overlay of the anatomical image and the delivered thermal dosage thresholded at 10 CEM43, and the red rectangle represents the region of interest for thermal dosage monitoring, only that portion of the image is processed and displayed for temperature and thermal dosage monitoring (top and middle images). All the image dimensions are in mm

IV. Results and Discussion

Fig. 9 shows all the target locations and five insertion trajectories obtained from the confirmation images for the in-phantom accuracy assessment. The average RMS error is summarized for each of the five targets in Table II, with maximum tip RMS error of 2.10 mm in the R direction and maximum rotation RMS errors of 1.98 deg about the R_R axis. The errors were within the expected range given the MRI resolution of 0.44 mm per voxel and the manual segmentation method used being susceptible to artifacts introduced by the insertion of the probe in the gelatin.

TABLE II:

Robot accuracy assessment results for the MRI phantom study

Target No	Planned Target			RMS Error
	R(mm)	A(mm)	S(mm)	R(mm)	A(mm)	S(mm)	RR(deg)	RS (deg)
1	9.03	54.75	−7.62	0.23	0.61	0.64	1.36	1.41
2	−25.01	55.95	−10.02	1.41	0.36	0.80	1.98	1.35
3	−23.48	75.68	−30.43	2.10	0.42	0.28	1.28	1.12
4	10.68	75.68	−30.43	1.84	1.04	0.73	1.48	0.82
5	−6.21	65.32	−20.66	1.36	0.44	0.72	0.28	0.31
Average STD				1.39	0.57	0.63	1.27	1.00
				0.64	0.25	0.18	0.56	0.41

Table III shows error analysis and procedure duration for eight of the ten animal studies. No targeting and/or ablation results were acquired from the remaining two studies primarily due to pig motion during the experiment and deflection of the probe when entering the thick swine skull. Note that the procedure duration reported here is the time between the first (robot registration) and the last (MR thermometry) MRI scans during the study; also as the studies progressed and team got more experience with the system, the procedure time decreased. Moreover, the robot was primarily designed for surgery in humans which leads to workspace and targeting challenges when dealing with a swine model. The large errors in study No. 3 were due to incorrect robot configuration resultant from updating the registration frame location to its new position under the robot base where it remained for the latter studies.

TABLE III:

Results for the animal studies: ablation probe placement accuracy and study duration

Study No	Planned Trajectory					|| Error ||					Duration (hh:mm)
	R(mm)	A(mm)	S(mm)	RR (deg)	RS (deg)	R(mm)	A(mm)	S(mm)	RR (deg)	RS (deg)
2	−6.77	69.44	7.61	−42.46	−16.27	1.71	3.66	4.73	4.35	6.44	03:35
3	−4.07	72.28	54.00	−30.37	5.22	1.12	16.26	6.10	9.09	4.93	04:09
5	6.32	83.96	212.00	−72.79	3.51	3.09	1.37	0.16	6.27	6.02	03:25
6	−5.41	77.85	32.90	−69.03	−16.43	1.18	0.50	0.23	2.46	12.61	04:06
7	−9.94	59.03	34.15	−78.10	15.76	0.59	2.92	0.50	6.45	0.76	03:43
8	−15.64	88.35	48.21	−56.10	1.23	1.29	1.99	1.11	5.10	2.20	02:44
9	−1.46	85.40	63.10	−61.41	13.91	0.05	0.86	2.14	1.83	0.12	02:26
10	1.86	87.75	75.95	−47.89	−8.50	0.64	2.78	1.16	12.16	8.50	02:31
Average STD						1.21	3.79	2.01	5.96	5.20	03:19
						0.85	4.82	2.08	3.19	3.92	00:38

Real-time monitoring of the delivered thermal dosage for study 5 at 4 different timestamps is shown in Fig. 10, while Fig. 11 shows the final dose delivered as measured by MR thermometry for seven of ten animal studies. Note that different power and time settings were used since we were fine tuning the NBTU ablator for in-vivo brain tissue. The irregular shape of some of the ablation regions can be attributed to one or more of: Difference between desired and actual probe location leads to MR temperature mapping at incorrect location (see Table III); Proximity to boundaries such as the bone in the skull can lead to ultrasound reflection. Also, during these studies, though the threshold for the desired thermal dosage was set to 10 CEM43 [43], the ablation system was always powered OFF after the predetermined treatment time had elapsed. During study No. 9, the ablation system had a technical issue and the treatment was powered OFF withing a short period and hence we did not observe any temperature changes in real-time MR-TI. The ablation results were used to compare the gross histology and the measured thermal dosage to establish the relationship between them in [44].

Fig. 11:

Thermal dosage for seven animal studies. Left image shows the MR-TI Slice, while the right image shows close-up view of the thermal dosage delivered, AP is acoustic power delivered and T is the treatment duration in seconds. Both parameters were varied between trials in order to evaluate their effect on lesion size.

V. Conclusion

The development of a fully integrated robotic system for MRI-guided closed-loop brain tumor ablation is presented. The system is designed with considerations that it could be used for multiple stereotactic neurosurgery procedures such as DBS or tumor ablation. The surgical planning and navigation along with the ablation system are integrated to produce an ablation lesion at a desired target location. Accuracy of 1.39±0.7 mm was measured in MR phantom studies while accuracy in a swine model was of 3.79mm ± 4.82 mm. The measured errors may be related to backlash, robot registration, patient motion, or other unmodeled errors during the probe artifact segmentation.

Future survival animal trials are in progress to better understand the effects of NBTU on live brain tissue which may lead to a clinically suitable system. Improvements to the MR thermometry can lead to more accurate ablation regions.