|Home | About | Journals | Submit | Contact Us | Français|
Deep brain stimulation (DBS) electrode placement using interventional MRI has been previously reported using a commercially available skull mounted aiming device (Medtronic Nexframe MR) and native MRI scanner software. This first-generation method has technical limitations that are inherent to the hardware and software used. A novel system (SurgiVision ClearPoint) consisting of an aiming device (SMARTFrame) and software has been developed specifically for iMRI interventions including DBS.
A series of phantom and cadaver tests were performed to determine the system’s capability, preliminary accuracy and workflow.
18 experiments using a water phantom were used to determine predictive accuracy of the software. 16 experiments using a gelatin-filled skull phantom were used to determine targeting accuracy of the aiming device. 6 procedures in three cadaver heads were performed to compare workflow and accuracy of ClearPoint with Nexframe MR.
Software prediction experiments showed an average error of 0.9±0.5 mm in magnitude in pitch and roll (mean pitch error −0.2±0.7 mm, mean roll error +0.2±0.7 mm) and an average error of 0.7±0.3 mm in X-Y translation with a slight anterior (0.5±0.3 mm) and lateral (0.4±0.3mm) bias. Targeting accuracy experiments showed average radial error of 0.5±0.3 mm. Cadaver experiments showed a radial error of 0.2±0.1 mm with the ClearPoint system (average procedure time 88±14 minutes) vs 0.6±0.2 mm with the Nexframe MR (average procedure time 92±12 minutes).
This novel system provides the submillimetric accuracy required for stereotactic interventions including DBS placement. It also overcomes technical limitations inherent in the first-generation iMRI system.
Frame-based stereotaxis has been a well established technique for a variety of neurosurgical procedures since the development of stereotactic frames in the first half of the last century1, 2. These procedures include brain biopsy, cyst aspiration, catheter placement, lesioning procedures and deep brain stimulation (DBS). Image-guided surgical workstations such as the StealthStation® (Medtronic, Minneapolis, MN) and VectorVision® (BrainLAB, Heimstetten, Germany) systems now allow many of these procedures to be done using frameless stereotaxis with skin or bone fiducials and optical or electromagnetic instrument tracking technology3–5.
Regardless of the platform, frame-based and frameless stereotaxis rely on imaging obtained prior to surgery. In the operating room, the surgeon must perform a registration process to make “imaging space” match physical space as closely as possible. In frame-based stereotaxy, this involves setting up the frame coordinates; in frameless stereotaxy, this involves fiducial registration. Once the operation begins and the dura is opened, loss of cerebrospinal fluid and development of pneumocephalus combine to cause unpredictable and varying degrees of brain shift, even in deep brain structures6–8. These stereotactic procedures are therefore inherently susceptible to inaccuracies as a result of errors in registration, errors due to brain shift and/or human errors in technique. Moreover, none of these traditional techniques allow true intraoperative visualization of the instrument in the brain or the intended target.
Interventional MRI (iMRI) was developed in the mid 1990s as a means of providing real or near real time imaging during neurosurgical procedures including stereotactic applications9–12. In 2002 our group began to develop a first-generation approach, using a commercially available skull mounted aiming device (Nexframe MR, Medtronic, Minneapolis, MN), to implant DBS electrodes into the subthalamic nucleus (STN) of patients with Parkinson’s disease using 1.5T iMRI13, 14. The motivations for this included the desire to improve the safety and accuracy of placement by eliminating the need for microelectrode or physiologic mapping, reducing brain penetrations, allowing intraoperative visualization of both the STN and the DBS electrode, providing immediate detection of complications and shortening procedure time. In addition, the technique increased patient comfort by eliminating the need for stereotactic frame or fiducial placement and allowing the procedure to be done under general anesthesia. Having patients asleep does eliminate the potential difficulties they may have tolerating an awake, physiologically-guided DBS procedure; however it also introduces the possibility of complications related to general anesthesia itself. We recently reported on our experience with this technique in 53 STN DBS implantations and found the accuracy superior to frame-based and frameless stereotaxy for DBS with comparable clinical outcomes and a low complication rate15.
Despite this initial success, the Nexframe MR has a number of limitations. The device is well designed for frameless DBS using optical tracking in a regular operating room, but its physical design and geometry limit its functionality in iMRI applications. Moreover, there is no dedicated software designed to perform iMRI guided interventions; as a result, many steps of the procedure have to be done manually and require significant expertise, while other steps in the procedure require image interpretation that is limited by human ability14. Here we summarize the development of a second-generation platform, the ClearPoint system (SurgiVision, Inc, Irvine, CA), and evaluate its performance on a series of phantom and cadaver tests to determine its capability, accuracy and workflow efficiency.
The ClearPoint implantation system is based on a specially designed trajectory guide (SMARTFrame) with control software (SW). A head fixation device integrated with radiofrequency receiving coils has been developed for the ClearPoint system, but here we evaluated only the core software and SMARTFrame trajectory guide. The methodologies developed for ClearPoint are based on those previously described for use with the Nexframe MR, but have some significant differences13. Notably, the new implantation system comes with control software dedicated to this specific application and features a trajectory guide with substantial functional enhancements and a narrower lateral profile compared with the Nexframe MR. The latter property makes it substantially easier to perform simultaneous bilateral procedures and imparts greater freedom in trajectory selection.
The trajectory guide, or SMARTFrame, is a single-use burr hole mounted device that permits motion in four independent directions (Figure 1). The device contains an MR visible targeting cannula, which has a spherically shaped tip. Two degrees of motion (pitch, roll) pivot the frame around this fixed point in orthogonal directions. The motion is driven by gears and each rotation of the control knob corresponds to a four degree change in angulation of the frame. This corresponds to approximately a 6.3 mm change in position at typical target depths (90 mm) for each full rotation of the pitch/roll controls. Thus, sub-millimetric adjustments can be difficult to achieve with only these controls. The other two sets of controls (X-Y stage) permit the user to dial-in offsets, which create parallel trajectories for given settings of the pitch and roll controls. These controls offer much finer sensitivity, with each rotation of the X-Y stage offsetting the trajectory by just 1 mm. The X-Y stage, however, offers limited range, with a maximum offset in either orthogonal direction of ±2.5 mm. Thus, the pitch and roll is initially adjusted to closely approximate the desired trajectory, and fine adjustments are subsequently achieved with the X-Y stage. The X-Y stage can also be used to correct minor inaccuracies on an initial pass into the brain. The SW provides instructions on how to adjust the SMARTFrame; therefore, the angulation of the frame with respect to the patient and the target must be known. This is achieved by imaging of three MR-visible fiducial markers built into the base of the frame, in a plane that is perpendicular to the default null pitch and roll angulations. All adjustments of the SMARTFrame can be made by turning knobs directly on the guide itself, or an extender device can be attached to the frame that permits remote manipulation of the guide (Figure 2). The targeting cannula has an open central channel through which a mandrel (stylet) may be inserted. The guide comes with a rigid ceramic mandrel laminated in plastic (length=305 mm, diameter=1.4mm) that is non-conductive, produces minimal MR artifact, and fits within a peel-away sheath.
The SW exists on a stand-alone workstation that is external to the magnet room but has a second monitor within the magnet room to provide the surgeon with the necessary feedback. The software communicates with the host computer of the MR system via a network link over which DICOM data are sent. The general workflow is divided into three different sections: burr hole planning (ENTRY), target selection and trajectory visualization (TARGET) and alignment of trajectory guide and insertion monitoring (NAVIGATE). The SW is designed for either unilateral or bilateral procedures, with a single target selected on each side.
In the ENTRY stage, the patient is placed in a head fixation device and a sterile field is established within the magnet bore using a custom drape. One or two sterile MR visible surface grids (Figure 3) are then placed on the patient’s head prior to the acquisition of a gadolinium enhanced T1-weighted volumetric MR scan (“3D Grid”, Table 1). Each grid contains 36 cells that are identified spatially by letters and numbers along each axis, as well as an extra cell in one corner that allows the SW to identify the orientation of the grid on the scalp. The acquired images are sent via a DICOM link to the control computer, where the SW automatically identifies the anterior and posterior commissures (AC, PC) as well as the mid-sagittal plane, and segments the cells of the surface grid(s). Default target coordinates, with respect to AC-PC, can be prescribed as a preliminary target, and the surgeon then determines the most appropriate entry point and trajectory to the target region. The SW permits orthogonal views of proposed trajectories and permits the surgeon to “fly-through” a specified trajectory to identify structures that will be traversed during implantation. It additionally provides guidance as to whether the proposed trajectory is likely to lie within the angular reach of the trajectory guide (±330 in pitch and ±260 in roll) and whether the bore of the MR scanner is likely to physically interfere with device insertion. Once a desired trajectory has been established, the SW provides the grid coordinates at which the burr hole(s) should be created (Figure 3). A specially designed bone-marking tool is then used to mark the skull at the desired burr hole location(s) and the patient is moved to the rear end of the magnet. Skin incision, burr hole preparation, dural opening and mounting of the SMARTFrames follows.
The patient is moved into the bore such that the superior aspect of the forehead is positioned at the center of the magnet. This positioning minimizes the distance from the bore center to both the external trajectory guide and deep brain structures. A new T1-weighted volumetric acquisition is obtained with sufficient coverage to reveal the brain structure of interest as well as the orientation of the SMARTFrame(s) (“3D SF”, Table 1). AC-PC and the mid-sagittal plane are again automatically detected, but can be manually refined by the clinician. Additional sequences, such as slab acquisitions focused on the region of interest in the brain, can additionally be obtained and fused with the volumetric data (“T2 Target”, Table 1). A target can then be selected either based on AC-PC coordinates, direct visualization of the local anatomy, or with a combination of both and supporting information from an atlas overlay if desired. The final trajectory, based on the actual mounted position of the trajectory guides and the selected target, can then be visualized. The SW again checks to assure that the selected target is within the angular range of the device and whether a bore collision during mandrel insertion is anticipated.
Once a trajectory has been selected the process of aiming the guide towards the target can begin. The SW provides scan plane parameters, including slice offsets and angulations, for acquiring imaging data through the base of the trajectory guide. This volumetric data should be high resolution (isotropic 1 mm) and is used to refine the coordinates of the fiducial markers on the SMARTFrame and of the spherical tip of the targeting cannula (“Pivot”, Table 1). The SW then prescribes the angulations and offsets of a scan plane through the upper portion of the targeting cannula (“Align”, Table 1). This rapid 2-dimensional scan plane is perpendicular to and centered on a ray that originates at the deep brain target, extends through the pivot point, and continues external to the patient. The scan provides a cross sectional view of the upper portion of the targeting cannula, which is automatically detected by the SW. Based on the detected position of the targeting cannula, instructions are then provided on what adjustments to the pitch and roll controls are necessary to make the targeting cannula collinear with the desired trajectory (Figure 4). After an adjustment is made, the same scan can then be acquired and the process repeated until the SW predicts that the trajectory guide is aimed at the target (typically within ~1 mm). The time required for scan acquisition in this step is approximately 5 seconds. At this point a locking screw is inserted to prevent any further changes in pitch and roll and the process of fine adjustments with the X-Y stage is initiated. The SW provides the angulations and offsets for oblique coronal and sagittal planes along the specified trajectory. MR scans are acquired along these perpendicular views and the resulting data are sent to the workstation (“Cor/Sag T1”, Table 1). The SW segments the targeting cannula and fits a linear function to its 3-dimensional orientation. This line is extended towards the target and a prediction of the magnitude and direction of the error at target depth is provided. Instructions are provided on how to adjust the X-Y translation stage to correct for this predicted error and the process can be repeated until an acceptable predicted error is achieved. The SW then provides the length on the insertion mandrel at which to set the depth stop such that the mandrel terminates at the selected target when inserted. Mandrel insertion can be monitored in a stepwise fashion and the SW provides the necessary scan plane offsets and angulations to view the insertion in oblique coronal and sagittal planes (“Cor/Sag T2”, Table 1). A final evaluation step is used to determine the targeting error, which is usually evaluated by re-acquiring the MR dataset on which the target was originally defined (“T2 target”, Table 1). If targeting accuracy is unacceptable, then the user can define a desired correction that is based on the magnitude and direction of the targeting error. The SW calculates the necessary adjustment to the X-Y stage to achieve this correction and presents the values. The surgeon can then remove the insertion mandrel, execute the necessary adjustment on the X-Y stage, and then re-insert to target depth.
The methodology summarized above was used in an assortment of phantom and cadaver studies to validate the approach and to determine the targeting accuracy. All imaging was performed on a 1.5T MR system (Philips Achieva, Best, The Netherlands) and two flexible 20 cm diameter surface coils were placed laterally on the phantom/cadaver, such that they covered both the phantom/cadaver target regions and the external trajectory guide. The imaging parameters for the various sequences that were utilized in these studies are summarized in Table 1.
The SW makes predictions about targeting error both while making pitch and roll adjustments as well as during adjustment of the X-Y stage. The predictions of error made during pitch and roll adjustments utilize geometric information about the position of the pivot point and upper part of the targeting cannula and their angular relationship to the desired trajectory. The predictions of error made during adjustment of the X-Y stage utilize an automatic fitting and projection tool that extends a ray from the center of the targeting cannula into the brain and compares this trajectory with the desired trajectory. The SW predictions are made prior to the insertion of any device and provide guidance as to whether the trajectory guide is properly aligned.
We evaluated the accuracy of these predictions on a cylindrical phantom that was filled with water. Two SMARTFrames were mounted on the phantom in a bilateral fashion and ceramic mandrels were inserted into the phantom to a depth of >90 mm. Once in position, the trajectory guides and mandrels were left stationary for the remainder of the experiment. Standard ClearPoint workflow, as described above, was then followed and a target point was selected directly on the mandrel itself at a depth of 90 mm. Thus, the SW should subsequently report no error since we defined the target at a spatial location where the inserted mandrel already exists. The software was advanced to the NAVIGATE portion of the procedure, where the SW reported predicted error and recommended changes in the pitch/roll adjustments. The error was noted but no changes were made in the pitch or roll. The SW was then advanced to the X-Y stage adjustment step, where again the predicted error was noted but no changes were made. At total of 9 bilateral procedures (18 targets) were simulated with varying trajectory angulations to determine the predictive accuracy of the software.
For this assessment, a SMARTFrame trajectory guide was mounted on a gelatin-filled, skull shaped phantom (Figure 1) containing eight discrete targets at depths consistent with deep brain structures. This phantom was identical to that previously used to assess targeting accuracy with the Nexframe MR system13. The skull phantom was secured to the MR table top with a head fixation frame (Malcolm-Rand, CMI Composites, San Clemente, CA). The ClearPoint system was then used to localize targets ipsilateral to the trajectory guide, and first pass targeting accuracy was determined. A total of 16 unilateral procedures were performed over four separate imaging sessions, with two to eight unilateral procedures done in each sitting.
A series of three cadaver studies were performed by three different neurosurgeons of varying experience levels with the device to assess the workflow efficiency and accuracy. Two unilateral mandrel insertions were sequentially performed in each cadaver head; one utilizing the Nexframe MR trajectory guide and following previously described methodologies and the other utilizing the SMARTFrame trajectory guide and ClearPoint15, 16. In each case a target approximating the STN was selected for the first procedure and a symmetric target with respect to the mid-sagittal plane was selected for the subsequent procedure. Each cadaver study was performed by a different neurosurgeon and the order in which each unilateral procedure was performed was alternated (Nexframe MR first twice, ClearPoint first once). Targeting accuracy and workflow efficiency, as measured by the time of the procedure from initial imaging to localization of the deep brain structure, were assessed. To determine targeting accuracy, the position of the mandrel was taken at the geometric center of its MR artifact at the depth of the selected target. The radial error in this plane, measured from intended target to mandrel position, was evaluated for each insertion.
All values are reported as Mean±Standard Deviation. A total of 18 different targets were evaluated with coronal angulations ranging from 9.5–25.60 and sagittal angulations ranging from −4.3 to 11.60. At the pitch/roll stage the software reported a mean error of 0.9±0.5 mm in magnitude. The mean pitch error was −0.2±0.7 mm, while the mean roll error was +0.2±0.7 mm, indicating there was minimal bias in the direction of the reported error. At the X-Y translation stage the average reported error was 0.7±0.3 mm. At this stage there was a tendency for the software to systematically predict that the mandrel would land slightly anterior (0.5±0.3 mm) and lateral (0.4±0.3mm) to its true position (Figure 5).
Sixteen simulated unilateral procedures were performed over four separate days (Figure 6). One base was found to be loosely secured to the skull phantom at the conclusion of the procedure and this data point was discarded. The average radial error over the remaining 15 insertions was 0.5±0.3 mm.
Insertion of a titanium (Nexframe MR) or ceramic (ClearPoint) mandrel to a selected deep brain target was performed in six individual implantations (Figure 7). The average target depth was 90.4 mm. Average procedure time was 92±12 minutes for the Nexframe MR approach and 88±14 minutes for the ClearPoint approach. The titanium mandrel produced substantially greater artifact on MR images, making the determination of its precise location challenging. Radial error for the three mandrels inserted with the ClearPoint system averaged 0.2±0.1 mm, while the radial error for the three mandrels inserted with the Nexframe MR system averaged 0.6±0.2 mm.
ClearPoint is a second-generation iMRI implantation system that provides a method for iMRI-guided access to deep brain structures. The system is more intuitive, appears more accurate in the small number of experiments performed here and is easier to implement than the earlier technique based on the Nexframe MR14, 17. These improvements were achieved by designing a skull mounted aiming system specifically with iMRI applications in mind, as well as designing a software environment that automates many steps of the procedure and reduces the likelihood of inaccuracies due to human influences.
The Nexframe MR is widely used in standard operating rooms for implantation of DBS electrodes using frameless stereotaxy. There are elements of its design, however, that limit its functionality in the MRI environment. Its base is funnel-shaped and has a widest diameter of 88 mm, which can present a challenge when trying to mount two devices side-by-side for bilateral simultaneous implantations. The fluid filled MRI visible alignment stem that is placed in the Nexframe MR to align the device to the target requires replacement with a multi-lumen insert for subsequent mandrel and DBS lead placement. Placing the multi-lumen insert correctly is difficult because the patient’s head is a considerable distance from the opening of the bore. Finally, the Nexframe MR is aligned by reaching into the magnet bore while simultaneously watching live MR images, which is physically awkward. The software component of the Nexframe MR technique also has significant shortcomings. It relies on the MR scanner console software to perform all of the imaging, targeting, trajectory planning, alignment and confirmation steps of the procedure. This software was not designed with neurosurgical interventions in mind; therefore, many steps of the procedure require manual entry of scan parameters and imaging planes as well as operator-controlled graphical reconstruction of the three dimension coordinates of implantation hardware.
The ClearPoint system addresses many of these shortcomings. The SMARTFrame is narrow in profile such that there is no interference between two of them during a bilateral simultaneous implantation, yet it exceeds the angular range of potential targets that can be reached by the Nexframe MR18. In real-world applications, the achievable angle is now likely limited by the diameter of the burr hole and the thickness of the skull. The targeting cannula acts as both an MR-visible aiming device and the guide for mandrel and lead insertion, eliminating mechanical errors that may occur during exchange of device inserts. Finally, the SMARTFrame can be steered remotely (with feedback from the SW), making the process of alignment much easier, faster and more comfortable.
The ClearPoint workflow is quite similar to that of existing stereotactic workstations. It accepts DICOM imaging data from any MR scanner and presents the scans in a consistent format independent of the scanning platform used. Many steps are now automated but can always be edited by the user, saving time while maintaining flexibility and control over the procedure. The procedure times with the Nexframe MR system were strongly dependent on the MR operator, while the ClearPoint system is significantly easier to follow with minimal training. Finally, the software’s ability to localize and segment objects in the surgical field and predict error based on their position reduces the number of steps subject to human error.
The phantom and cadaver studies in this report were designed to determine the capability, accuracy and workflow efficiency of the new system. The SW prediction feature in particular is new to the second-generation device and required validation. The phantom target conditions used to generate the SW predictions were designed to replicate angles and depths that would be encountered clinically. This series of experiments were unique in that the mandrel was placed in the phantom and the SW was then used to target the mandrel itself. The predicted error should therefore be zero, so any predicted error reported by the SW must arise from either non-linearities in MR space or the SW prediction paradigm, and not the aiming accuracy of the SMARTFrame. With pitch and roll adjustment, the mean error was 0.9±0.5 mm with no bias in either the pitch or roll directions (both having magnitude of 0.2±0.7 mm). With X-Y stage translation, the mean error was lower (0.7±0.3 mm) but with a tendency of the software to systematically predict errors in the anterior and lateral directions by 0.5±0.3 mm and 0.4±0.3mm, respectively. This directional bias may be specific to field inhomogeneities and image distortion in our particular MR scanner and not the ClearPoint system itself, or could represent a bias in the SW prediction paradigm specific to the X-Y stage translation step. We used the same scanner for these phantom studies as we do for clinical iMRI DBS implantations using the first-generation system, and have noted similar direction and magnitude specific biases during those procedures. Thus, slight non-linearities in MR space may be the dominant contributor to the errors that were detected in this analysis.
The phantom accuracy testing was a more traditional phantom experiment where a MR visible target was localized using ClearPoint and a mandrel was passed to the target as directed by the SW. In addition to determining the accuracy of ClearPoint itself, a preliminary comparison could be made with the Nexframe MR, as the same phantom, same MRI scanner and same investigators were used for both the current experiments and experiments previously reported using the Nexframe MR. The ClearPoint system was found to have a radial error of 0.5±0.3 mm. In addition to providing submillimetric accuracy, ClearPoint appears to improve on a radial error of 0.8±0.5 mm reported for the Nexframe MR13. The apparent improvement in accuracy may be attributable to improvements in the software environment as noted above, although improvements in the aiming device due to the integrated targeting cannula may also play a part.
The cadaver comparisons studies were designed to be a head-to-head comparison of the Nexframe MR and ClearPoint with regards to both workflow efficiency and to some degree accuracy. These differed from the phantom testing in that the actual entry point planning and burr hole placement portion of the workflow was done for each system in these tests. This was not done in the phantom accuracy testing, where the same pre-drilled burr holes were used repeatedly. It was also the best opportunity to preliminarily compare the accuracy of the two systems, as one of each aiming system was mounted on the same cadaver head and tested in the same session with the same surgeon. The average procedure time with three surgeons of varying levels of experience was only 4 minutes shorter using ClearPoint; however, given the large amount of clinical experience we have with the Nexframe MR and the virtual lack of clinical experience with ClearPoint, the fact that the new generation system is already comparable or slightly faster is likely significant. Unfortunately with a small sample size (N=3 for each system) it is not possible to draw firm conclusions on relative accuracy, but these initial findings with ClearPoint are encouraging.
Finally, while we have been clinically focused on the use of iMRI for implantation of DBS electrodes, this technology has other potential applications including depth electrode placement for epilepsy monitoring, brain biopsy and catheter or cannula placement. Our group has already used a slightly modified version of ClearPoint to perform iMRI guided infusions in non-human primates, with the goal of using this technology to guide convection enhanced delivery of therapeutics to the brain and monitor them using real time MRI. Such a technique would be a natural extension of gene therapy trials that have already been done for Parkinson’s disease, and would have further applications in other neurodegenerative disorders or in the chemotherapeutic treatment of neoplasms19–23. Although ClearPoint can be used for a variety of different iMRI-guided interventions, we tested it specifically with DBS implantation in mind. Given the fact that the safety of imaging patients with implanted DBS electrodes at 3T is not established, we only tested the system in a 1.5T environment. For procedures such as brain biopsy or catheter placement, higher field strengths could be considered, although the improvements in tissue discrimination with stronger magnets must be weighted against the greater potential for field inhomogeneity.
This novel iMRI stereotactic system provides improved functionality and workflow when compared to the first-generation iMRI system. Various phantom and cadaver experiments demonstrate software predictions of targeting error and accuracy of the aiming device in the submillimetric range necessary for stereotactic interventions including DBS implantation. Further validation in the clinical setting is warranted.
The authors would like to thank Christine Arnold for her assistance with the cadaver specimens and Dr. Kiarash Shahlaie for his critical comments on the ClearPoint system.
This work was funded in part by UC Discovery Grant #156321 and an NIH grant (R21 EB008888).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Dr.s Larson, Starr and Martin have received grant funding from SurgiVision, Inc, but do not have any financial interest in the company. Geoffrey Bates and Lisa Tansey are employees of SurgiVision, Inc.