Abstract
Background: Surgical navigation becomes inaccurate when spinal alignment changes, requiring repeat scan such as O-arm CT to restore accuracy, but this introduces additional radiation exposure, workflow disruption, and delay. We developed and confirmed ex vivo validation of a novel hand-held system (HHS) that can update a preoperative CT (pCT) using intraoperative stereovision (iSV), leveraging a level-wise registration framework to account for intervertebral motion.
Methods: iSV guidance was compared against Medtronic Stealth Station guidance for lumbar pedicle screw placement in 6 live Yucatan mini-pigs. The animals were positioned supine with hyperextension of the lumbar spine for pCT. They were then positioned prone with flexion of the lumbar spine during surgery. iSV data of exposed spine was acquired to generate an updated CT (uCT) to reflect the current spinal alignment. Half of the pedicle screws were randomly selected to be placed using pCT and the other half with uCT. The surgeon was blinded and was instructed to place the screws solely based on guidance provided by the navigation system.
Results: A total of 66 pedicle screws were implanted in 6 live animals. Only 1 of the uCT navigated screws showed a grade 3 (>4 mm) breach of 5.5 mm, whereas the pCT navigated screws had 3 instances with values of 18 mm, 9.5 mm, and 10.9 mm. The registration error of the iSV uCT was lower than that of the pCT (3.9±1.9 mm vs 6.1±5.4 mm; p<.05).
Conclusions: Intraoperative Stereovision uCT resulted in lower registration error and improved pedicle screw accuracy in live animals.
Keywords: Stereovision, Image guidance, Registration, Pedicle screw, In vivo, Deformation, Radiation-free, Motion
Introduction
Image guided surgery has improved efficiency and patient outcomes in spine surgery by providing real time location information of critical anatomy within the surgical field, such as pedicle cortical bone margins, nerve roots, and spinal cord [[1], [2], [3]]. Lumbar pedicle screw perforation may be reduced from 6% to 1% with surgical navigation [3]. For the continued increasing rate of lumbar fusion surgery, navigation is becoming the standard of care at many hospitals with widespread availability of intraoperative 3D fluoroscopic imaging (such as O-arm from Medtronic, Minneapolis, MN, and OEC 3D from GE Healthcare, Chicago IL) [[4], [5]].
However, spinal surgery is dynamic, and vertebral alignment can change during the procedure, degrading accuracy of surgical navigation if significant intervertebral motion occurs after registration of preoperative computed tomography (pCT) scans for image guidance.
To restore accuracy, a new registration CT is required. Although fluoroscopic CT systems provide automatic surgical field registration that can be repeated during a case, these systems are expensive, cumbersome, and increase radiation exposure, a significant risk for both the patient [[6], [7]] and staff [6].
We developed a handheld intraoperative stereovision system to address spinal alignment change from pCT by using images of the exposed spine surface leveraging a marker-less multilevel registration framework [[8], [9]]. Other visible light systems exist for spine registration [[10], [11], [12], [13], [14]]. HHS requires no manual localization of levels to be deformed, deforms multiple levels, and generates an updated image set for open, navigated spine surgery.
7D Surgical FLASH(Orthofix, Lewisville, TX) [[10], [11], [15]] also performs marker-less registration of spine during open procedures. Unlike HHS, FLASH currently updates up to 2 levels at a time, and a surgeon must identify the levels to be deformed intraoperatively with stylus probe and on pCT. Other visible light systems use depth projections of skin to register CT, [[12], [13]] or endoscope images [14] for minimally invasive spine surgery (MISS).
The HHS was validated in cadaver swine pedicle screw placement studies. Results showed navigation using an iSV updated CT after alignment change improved registration error and pedicle screw accuracy compared to navigation with pCT [16]. Results are presented from in vivo validation experiments for iSV-updated navigation for pedicle screw placement after spinal alignment change in live pigs to assess performance degradation from movement associated with respiration, bleeding and other physiological factors compared to cadaver animals.
Material and methods
Intraoperative stereovision
Methods follow the same protocol as previous cadaver studies [[17], [18]]. An intraoperative, hand-held stereovision system was constructed. Two webcams (Logitech C920, Newark, CA) were rigidly affixed to a custom designed 3D printed handle. Both cameras were attached with cyanoacrylate glue on opposing wedges allowing for both cameras field of view (FOV) to intersect at 15 cm. Two laser diodes which converged beams at 15 cm were inserted into holes on either side of the cameras. LED strip lights were affixed to front of HHS with adhesive. An active optical tracker (Medtronic, Dublin, Ireland) was affixed to a wedge on the back side of the HHS.
HHS was calibrated with a rigid checkerboard made of aluminum (Calib.io, Svendborg, Denmark). Checkerboard was affixed to custom, tracked frame [[19], [20]]. HHS was rigidly clamped in a test tube holder on a ring stand. Tracked snapshots of the checkerboard were collected at 15 cm with illumination from halogen bulb desk lamp. Reference frame was rotated and translated across all 6 degrees of freedom.
100 Calibration images were collected, a randomized subset of 25 images were selected for spatial calibration validation. Camera intrinsic parameters and extrinsic parameters were calibrated with the Matlab Stereo Calibrator app (Mathworks). Rigid transformation between manually localized checkerboard corners in the space defined by active tracker and HHS localized points serve as spatial calibration [[19], [20]].
HHS was dropped and broken, after live animal 1. A new system with same hardware was assembled. The frame was narrowed, cameras were removed from their stock enclosures and placed into a cup-like enclosure with the front exposed.
Image data of the surgical field were collected with HHS [15]. HHS was controlled with a custom script in LabView [21]. The data collector had a real time view of both HHS cameras, and visibility of all optical trackers. A snapshot of spine was collected when prompted by the HHS operator
Deformable registration
Level-wise registration was employed to align each vertebra in the reconstructed iSV data automatically with its counterpart in pCT. Stereo pairs were rectified, and correspondence was computed with a GPU optimized optical flow algorithm [22]. Depth were extracted from disparity map.
Surfaces were transformed to tracking space with spatial calibration. Bony surface of spine was segmented manually with polygons on left camera image. pCT was segmented as one body (L1-L6) manually using 3D slicer [[8], [23]]. pCT and iSV point clouds were rectified using nonlinear principal component analysis (NLPCA) [24]. Rectified iSV and pCT dorsal surfaces created 2D depth projection images. SIFT correspondence of depth maps was used to rigidly register depth projections and correspond overlapping pixels. 3D points in pCT were rigidly registered to iSV with SIFT correspondence.
Anterior surface of rectified pCT was segmented by level with Hough transformation to find horizontal lines corresponding to the intervertebral disk. Segmentations were refined with manually selected depth projection of spinous process and facet joints. Segmented pCT was used as template to segment iSV by level.
Corresponding levels were registered with a modified iterative implementation of Horn’s quaternion method as follows [25]. Three-dimensional coordinates of pCT were registered to SIFT corresponded iSV coordinates with Horn’s method, [25] and points with L2 norms greater than 50 mm are removed. Registration is iterated without update to correspondence as L2 norm exclusion was reduced by 2.5 mm until 1.5 mm. multilevel segmented pCT voxels are rigidly registered to corresponding iSV point cloud locations generating uCT with intensity values from pCT. Surgical field registration was identity as defined in spatial calibration of HHS. uCT images and registration were uploaded to StealthStation for navigation.
A worst-case scenario positional change between pCT imaging and instrumentation was accentuated with a roll of surgical towels placed under lumbar spine.
Cobb Angles for both pCT and iCT were calculated and their difference was quantified to measure posture change between preoperative imaging (pCT) and intraoperative surgical positions (iCT) [26].
Surgical procedure
The study was approved by the Institutional Animal Care and Use Committee (IACUC) at Dartmouth. Six (6) live Yucatan mini-pigs (Sinclair Bioresources, Auxvasse, MO), weighing 40–60 kg, underwent lumbar pedicle screw placement after acclimation. No exclusion criteria, other than weight, existed. Fig. 1 shows a flowchart of the experimental procedure Fig. 2.
Fig. 1.
Experimental flowchart. Operations move from left to right and separate after A), B), and C) are performed into Stealth registration (top, solid line; d1 and e1) and uCT registration (bottom, dashed line; d2 and e2), prior to recombining at F) for blinded screw placement with navigation by uCT and Stealth. Fig. generated with Bioreneder.com.
Fig. 2.
Posterior vertebral surfaces of uCT (green circles) and pCT (red x’s) compared to the co-registered iSV surface (blue dots). A: iSV and pCT surfaces. B: iSV and uCT surfaces.
Animals were anesthetized with ketamine and anesthesia was maintained with isoflurane and fentanyl. Fentanyl maintained analgesia control which concurrently stabilized anesthesia. Once the animals were confirmed to be under anesthesia by jaw muscle tension, and with no reaction to application of pressure between toes, the operation commenced. Vitals were monitored throughout procedure.
Posterior surface of the lumbar region was exposed. A midline incision exposed the entire bony surface inclusive of the transverse process from L1 to L6. A paralytic was injected as bolus when involuntary muscular contractions were apparent during exposure.
Fiducial markers consisting of three mini-screws (1.5 mm diameter and 2 mm length; Stryker, Kalamazoo, MI) were implanted in each vertebra (in the spinous process, and left and right transverse processes), the incision was closed with sutures, and the animal was flipped supine. pCT was acquired with lordotic bump.
The animal was then positioned prone in surgical position, and the lumbar region was re-exposed. Retractors were placed on the caudal and cranial end of the cavity. A ∼2 cm incision was made to expose the left iliac crest. A passively tracked reference frame was rigidly clamped to the ilium facing caudally and tracked by a commercial navigation system (StealthStation S7, Medtronic).
An intraoperative CT (iCT) scan was acquired to measure alignment change, and for registration analysis. The mini-screws were localized in the surgical field via a tracked stylus probe which provided position data that later served as ground-truth locations for registration analysis. An evenly distributed subset of these mini-screws (N=10) were used for fiducial based registration (FBR) of pCT.
Pedicle screws of the same size (Vertex Select; Medtronic) were inserted on each vertebra by a surgeon blinded to the navigation method. The sequence of levels was randomized, as was selection of right and left sides to be guided by Stealth (FBR pCT) or uCT, respectively.
Surgeons were asked to suspend clinical judgement, and insert screws based solely on navigational information. If significant medial breach was expected, the screw was placed rigidly in the vertebral arch without entering the spinal canal intraoperatively, and later inserted fully after euthanasia for breach assessment.
Pedicle screw breach analysis
Our primary outcome, pedicle screw perforation grade was assessed. Pedicle screws placed rigidly in the vertebral arch were advanced after euthanasia and trajectory was confirmed on CT. Instrumented vertebrae were dissected by level. Medial breach of pedicle screws were measured with calipers.
Breach severity was graded according to thresholds defined by Gertzbein and Robbins [27]: 0 mm (no breach) =Grade 0, <2 mm = Grade 1, 2–4 mm = Grade 2, and >4 mm = Grade 3. Grade 0 was assigned, if the screw threads were visible in the spinal canal but not measurable. Pedicle width of implanted levels were measured manually on pCT with 3D Slicer [17].
Registration error
Computed registration error served as our secondary outcome measure, because screws navigated with stealth were not given opportunity to correct for spinal alignment. Fiducial mini-screws were localized manually on pCT, uCT, and iCT with StealthStation. Mini-screw heads were localized on HHS surface from left camera images used for deformation.
Accuracy of image guidance was evaluated in terms of point to point registration errors (ppRE) using implanted mini-screw fiducials. ppRE was calculated as the distance L2 norm between the fiducial localized on the CT transformed to physical space via corresponding surgical field registration, and its counterpart position intraoperatively located via a stylus probe.
Since ten (10) implanted fiducials were used for pCT registration, pCT ppRE was decomposed into a fiducial registration error (FRE) component and target registration error (TRE) contribution from the remaining fiducials (not used in pCT registration).
Additionally, we evaluated HHS accuracy using the same ppRE measure, defined as distance between the point manually localized on the tracked, reconstructed iSV surface and its counterpart localized intraoperatively via the tracked stylus probe.
Statistical analysis
A t-test was performed to compare pCT and uCT ppREs [26]. Two Spearman's rank correlation coefficients were calculated: one to evaluate associations between ppRE and posture change for uCT (N=99) and pCT (N=99), respectively.
The other to determine relationships between screw insertion order and breach, again for both uCT and pCT. All data analysis was completed in MATLAB (Mathworks, Natick, MA).
HHS hardware investigation
After all in vivo experiments, the HHS was subjected to torture tests. The system was rapidly shaken, and the system was wiped down with disinfectant wipes.
HHS was calibrated with snapshots, and after the test, new snapshots were used to verify change in reprojection error.
Results
Intraoperative stereovision
Stereo-parameters were calibrated to <1 pixel of reprojection error, and HHS was calibrated spatially to <0.5 mm prior to surgery. Mean iSV accuracy across the 6 cases was 2.4±1.4 mm.
HHS images with hand motion less than 0.1 mm/s were kept. HHS images with incomplete tracking data were rejected.
Deformable registration
uCT images only included L1-L6 where stealth navigation included all bone. uCT segmentation was imperfect, overhanging bone, or poorly rectified regions were segmented through bone which caused visible artifacts.
Overall computational efficiency of uCT registration was ∼10–15 minutes, comparable to pCT registration with FBR (∼10 minutes).
Lordosis, as measured by cobb angle from L1-L6 ranged from 12.2 o to 41.3 o with a mean value of 24.8 o.
Guidance analysis
Sixty-six (66) pedicle screws were placed (33 using each navigation method) and evaluated in terms of guidance method accuracy and pedicle screw placement breach.
Case characteristics and accuracy assessment results are shown in Table 1 for each animal experiment.
Table 1.
Summary of cases, levels involved, iSV accuracy, registration errors (ppRE, TRE, and FRE), Cobb angle changes between pCT and iCT, pedicle width, and breach values (uCT and stealth).
| Case ID | 1 | 2 | 3 | 4 | 5 | 6 | Mean Values |
|---|---|---|---|---|---|---|---|
| Levels Instrumented | L1-L5 | L1-L5 | L1-L6 | L1-L6 | L1-L5 | L1-L6 | |
| iSV (mm) | 1.2±0.6 | 1.6±0.5 | 3.5±0.6 | 4.3±0.6 | 1.1±0.5 | 2.1±0.5 | 2.4±1.4 |
| uCT PPRE (mm) | 2.9±1.0 | 1.8±1.2 | 4.2±1.6 | 5.6±1.7 | 3.8±2.4 | 3.8±1.0 | 3.9±1.9 |
| Stealth PPRE (mm) | 5.5±5.1 | 3.8±3.0 | 8.3±7.3 | 4.4±4.1 | 7.6±5.8 | 6.8±4.9 | 6.1±5.4 |
| Stealth TRE (mm) | 7.0±5.0 | 3.7±2.3 | 9.6±5.6 | 5.0±3.7 | 10.4±1.4 | 7.6±4.3 | 7.2±4.7 |
| Stealth FRE (mm) | 4.7±5.0 | 3.8±3.6 | 7.4±8.4 | 3.9±4.4 | 6.1±6.7 | 6.1±5.4 | 5.4±5.7 |
| Lordosis | 21.3° | 12.2° | 32.1° | 19.0° | 41.3° | 23.0° | 24.8° |
| Pedicle Width (mm) | 5.2±0.4 | 7.6±0.4 | 7.5±1.5 | 7.6±0.8 | 6.6±0.6 | 6.3±0.4 | 7.0±1.3 |
| uCT Breach (mm) | None | None | None | 2.5 | 5.5 | 4 & 4 | 4 |
| Stealth Breach (mm) | 18 | None | None | None | 9.5 | 10.9 | 12.8 |
Breach values separated with “&” designate multiple incidences of breach. Mean values across experimental cases are also compiled.
Although both uCT and Stealth guided screws caused some medial breaches, Stealth guided placement outcomes were more severe relative to the uCT navigated results (18 mm, 9.5 mm, and 10.9 mm for Stealth navigation relative to 2.5 mm, 4 mm, 4 mm and 5.5 mm for uCT guidance). uCT guidance outperformed Stealth navigation in 5 out of 6 cases. Average ppREs of uCT registration were significantly lower than those of Stealth registration (3.9±1.9 mm vs 6.1±5.4 mm; p<.05). Values of Stealth ppREs were higher towards the cranial end of exposed spines (especially L1; 14.5±6.0 mm, and L2; 6.6±3.6 mm). Fig. 3 shows level-wise ppRRE values for both uCT and Stealth registrations.
Fig. 3.
Boxplot of ppREs between uCT (green) and Stealth (red) across levels. Vertebral levels (L1-L6) are separated by vertical lines. Boxes comprised of data within the 25% and 75% intervals. Median lines are indicated.
Statistical analysis
Spearman rank correlation analyses indicate strong association between lumbar lordotic angle change during surgery and Stealth ppRE (ρ=0.94, p<.05), whereas uCT ppRE was not correlated with the alignment change (ρ=0.09, p=.92).
The same correlation analysis was inconclusive for order of insertion versus breach for both uCT (ρ=0.20; p=.28) and Stealth (ρ=−0.05; p=.77).
HHS hardware investigation
HHS was calibrated before each hardware test. After rapidly shaking the HHS, reprojection error went from 0.26±0.1 mm to 0.28±0.1 mm.
After a wiping the HHS down with a high-grade disinfecting wipe, the reprojection error went from 0.37±0.2 mm to 1.28±0.2 mm.
Discussion
In this study, we conducted in vivo validation of iSV, comparing performance of stereovision-updated uCT navigation vs Stealth image guidance without correction of posture change for lumbar pedicle screw placement. The overall registration errors of Stealth were high and led to severe perforations (all >9 mm) when an average of 25 degree change occurred in lumbar lordosis between preoperative CT imaging and intraoperative surgical position.
In contrast, stereovision-updated uCT registration resulted in less severe breach, with only one Grade 3 perforation (5.5 mm). iSV level-wise uCT registration was able to account for changes in spinal alignment and improved navigation accuracy. The near-zero Spearman rank correlation coefficient for uCT between ppRE and change in cobb angle(ρ=0.09) also showed that uCT accuracy was similar across levels, whereas Stealth accuracy was more accurate L2-L6.
The errors in fiducial-based registration for Stealth were likely lower than in clinical settings as mini-screws are not applicable, and natural landmarks are typically used as fiducials, resulting in higher localization errors, and subsequently, higher registration errors.
The Grade 3 perforation with uCT navigation in Case 4 may have occurred because the iSV errors were highest among all cases (4.3±0.6 mm) and beyond the clinically accepted range for an intraoperative imaging system. Unfortunately, iSV errors were propagated by the uCT registration which matched the iSV surfaces; thus, uCT was biased by the inaccurate iSV data, leading to the highest ppRE among cases (5.6±1.7 mm).
We have since investigated our HHS construction and found that its hardware mounts potentially changed after being wiped down with disinfecting wipes between calibration and OR use in some of cases, which caused a rigid shift of the iSV surfaces from their ground truths, leading to higher reconstruction errors relative to other cases as well as previous studies [[17], [18]].
We subsequently corrected for the rigid shift, and improved ppRE errors from 4.2±1.6 mm to 2.1±1.3 mm for Case 3 and from 5.6±1.7 mm to 2.3±1.4 mm for Case 4, and into the clinically acceptable range. The impact of corrected iSV data on the pedicle breach results cannot be determined retrospectively at this time. Work is underway to perform simulated pedicle screw placements to evaluate this system under ideal conditions.
Vertebral positioning is known to change intraoperatively [[28], [29]]. Guha et al. showed in a single level simulation study that 38.5% of instrumented vertebrae deviated by ≥2 mm during lumbar instrumentation [29]. Additionally, both Guha and Glossip recorded average breathing motions of 1.96 mm [29] and 1.3 mm, [28] respectively. Our cadaver study shows correlation between breach and sequence of insertion for uCT navigated screws, [18] which suggests that a dynamic surgical field correlates with higher grade of breach for uCT navigated pedicle screw placements.
The low Spearman rank correlation coefficient and high p-value in this study did not substantiate the claim (ρ=0.20, p=.28), and further experiments may be required to resolve the discrepancy. Preliminary simulation studies show that our registration model can be executed iteratively to account for a dynamic surgical field [30]. This iterative implementation could further account for intraoperative change without adding the risk of radiation exposure that repeated O-arm or iCT acquisitions would introduce. Future studies are planned to incorporate our image updating model iteratively throughout instrumentation.
Compared to previous cadaver studies, [18] the live animal results produced larger errors in terms of screw position and registration fidelity for several reasons. Surgical exposures were larger in cadaver studies. Here, they were more commensurate with standard practice. As a result, our HHS, which had a fixed working distance of ∼15 cm, acquired incomplete profiles of the vertebra at the ends of the surgical openings in some cases (e.g., L1 or L5/L6).
As a result, ppREs of uCT were higher at the ends (5.08±2.43 mm; L1 and 4.6±0.8 mm; L6) compared to other levels (3.5±1.7 mm; L2-L5). Cranial and caudal ends of the surgical cavity were poorly illuminated, and were filled with blood and interstitial fluid. Effort is underway to improve registration accuracy with poor illumination and incomplete profiles of the spine.
Breathing introduced motion between iSV acquisitions, increasing iSV surface reconstruction errors and leading to discrepancies between the times of iSV acquisition and pedicle screw placement. Although the lumbar region largely moved as a whole due to breathing, small alignment changes did occur.
Significant motion was observed in soft tissues and organs on the ventral side of the spine during screw placement in live experiments, whereas cadaver organs had been removed and less motion was observed. Muscle contractions caused alignment changes during the operation. A paralytic agent was used during the initial exposure, but was discontinued during instrumentation.
The surgeon was instructed to suspend surgical judgement, and use the navigation images as they were portrayed. Having a clear view of the surgical field, and apparent uCT artifacts may have influenced the placement of pedicle screws. Despite the limitations and associated procedural improvements that can be introduced, this study is an essential step towards clinical applications as the live animals provided a physiological environment similar to clinical conditions.
Our level-wise registration algorithm uses a custom iterative solver. This solver does not update correspondence between iterations. The SIFT based correspondence is quite robust, but methods exist which update correspondence and can be fine-tuned further to converge to a global solution [[31], [32]]. Additionally, the pCT bony surface was altered when down sampled by NLPCA. Work is underway to evaluate the efficacy of other iterative registration methods with an accurate bone boundary.
HHS has not been assessed for small exposures to date, although simulations suggest performance will not degrade substantially [33]. The level-wise registration framework can be adapted to other intraoperative modalities that are amenable to minimally-invasive exposures, such as ultrasound. Algorithms are being developed to detect bony surfaces and reconstruct 3D surfaces based on 2D tracked ultrasound images.
Additional error sources from live procedures have been identified in this study, which lay a solid foundation for further refinements of iSV image updating of pCT for image guidance during spine procedures, and adaptation to clinical settings with improved accuracy and efficiency.
Study limitations
uCT was compared to FBR with accentuated lordosis (mean value 24.8°) which may be a greater change than is commonly observed under clinical conditions in the operating room. Data acquired with Stealthstation (Medtronic, Dublin, Ireland) was based on the initial FBR with no intraoperative correction as might occur in practice. Stealth FBR mini-screws were selected evenly from L1-L6. Although the StealthStation removed mismatched fiducial automatically, especially around L1 leading to bias in accuracy towards the cranial end of the spine in this case.
Surgeons were instructed to suspend clinical judgement and navigate screw placement solely based on image guidance in order to assess navigational accuracy based on the two technical approaches as closely as possible while minimizing surgeon influence. Although conceptually appealing, elimination of all behavioral variables was not achieved and some loss of experimental control did occur which is difficult to quantify. uCT PPRE was still greater than the targeted clinical threshold of 2 mm (3.9±1.9 mm) PPRE was corrected post procedure, and its impact on the initial pedicle screw placement is unknown at this time.
Conclusion
This study is the first in vivo study utilizing a marker-less, multilevel registration method driven by a nonionizing HHS for navigation. Severity of breach was improved as compared to StealthStation. (18 mm, 9.5 mm, and 10.9 mm for Stealth navigation relative to 2.5 mm, 4 mm, 4 mm and 5.5 mm for uCT guidance) Registration error is improved compared to pCT but still above the clinically acceptable threshold of ≤2.0 mm. (3.9±1.9 mm for uCT vs 6.1±5.4 mm for Stealth; p<.05) Work is underway to improve our model’s accuracy for narrow exposures as seen at L1 and L6 (PPRE; 5.08±2.43 mm; L1 and 4.6±0.8 mm; L6), and simulations are planned to test our model under ideal circumstances with enhanced registration methods.
Author Contributions
Ryan B. Duke, provided conceptualization, data curation, formal analysis, and writing of the original draft. Xiaoyao Fan, provided conceptualization, data curation, software, supervision, investigation, and reviewed and edited the manuscript. Steven Baltic, provided investigation. William ``Ross'' Warner, provided conceptualization, and investigation. Songbai Ji, provided software, supervision, and validation. Linton Evans, provided supervision, investigation, and methodology. Sohail Mirza, provided conceptualization, formal analysis, supervision, funding acquisition, validation, methodology, and writing – review and editing. Keith Paulsen, provided conceptualization, resources, supervision, validation, methodology, project administration, and writing – review and editing.
Declarations of competing interests
One or more of the authors declare financial or professional relationships on ICMJE-NASSJ disclosure forms.
Acknowledgments
This work was funded by the National Institutes of Health (USA) (grant R01EB025747 and R01CA CA159324). Medtronic provided the StealthStation. The funding sources did not play a role in the investigation. The authors are inventors on patents and patents-pending related to stereovision assigned to the Trustees of Dartmouth College. Drs. Fan, Mirza, and Paulsen are involved with early-stage commercialization of some of the technologies described in the paper through startup companies: InSight Surgical Technologies and PEER Technologies.
Footnotes
FDA device/drug status: Not applicable.
Author disclosures: RBD: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); Other Support: Medtronic (None). XF: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); Patent: US Patent Office, Dartmouth College (US12239494B1: Amount not disclosed), Patent: US Patent Office, Dartmouth College (US20260017912A1: Amount not disclosed); Scientific Advisory Board: InSight Surgical (Amount not disclosed), PEER Technologies (Amount not disclosed); Other Support; Medtronic (None). SPB: Nothing to disclose. WRW: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); US Patent Office, Dartmouth College (US20260017912A1: Amount not disclosed); Other Support: Medtronic (None). SJ: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); US Patent Office, Dartmouth College (US12239494B1: Amount not disclosed); Other Support: Medtronic (None). LTE: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); Other Support: Medtronic (None). SKM: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); US Patent Office, Dartmouth College (US12239494B1: Amount not disclosed); Scientific Advisory Board: PEER Technologies (Amount not disclosed); Other Support: Medtronic (None). KDP: Research Support: National Institutes of Health (F), National Institutes of Health (I); Medtronic (None); US Patent Office, Dartmouth College (US12239494B1: Amount not disclosed), Patent: US Patent Office, Dartmouth College (US20260017912A1: Amount not disclosed); Scientific Advisory Board: InSight Surgical (Amount not disclosed), PEER Technologies (Amount not disclosed); Other Support: Medtronic (None).
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.xnsj.2026.100939.
Appendix. Supplementary materials
References
- 1.Tian N.F., Xu H.Z. Image-guided pedicle screw insertion accuracy: a meta-analysis. Int Orthop (SICOT) 2009;33(4):895–903. doi: 10.1007/s00264-009-0792-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Mirza S.K., Wiggins G.C., Kuntz C.I., et al. Accuracy of thoracic vertebral body screw placement using standard fluoroscopy, fluoroscopic image guidance, and computed tomographic image guidance: a cadaver study. Spine. 2003;28(4):402. doi: 10.1097/01.BRS.0000048461.51308.CD. [DOI] [PubMed] [Google Scholar]
- 3.Silbermann J., Riese F., Allam Y., Reichert T., Koeppert H., Gutberlet M. Computer tomography assessment of pedicle screw placement in lumbar and sacral spine: comparison between free-hand and O-arm based navigation techniques. Eur Spine J. 2011;20(6):875–881. doi: 10.1007/s00586-010-1683-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Kobayashi K., Ando K., Nishida Y., Ishiguro N., Imagama S. Epidemiological trends in spine surgery over 10 years in a multicenter database. Eur Spine J. 2018;27(8):1698–1703. doi: 10.1007/s00586-018-5513-4. [DOI] [PubMed] [Google Scholar]
- 5.Buser Z., Ortega B., D’Oro A., et al. Spine degenerative conditions and their treatments: national trends in the United States of America. Glob Spine J. 2018;8(1):57–67. doi: 10.1177/2192568217696688. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Jenkins N.W., Parrish J.M., Sheha E.D., Singh K. Intraoperative risks of radiation exposure for the surgeon and patient. Ann Transl Med. 2021;9(1):84. doi: 10.21037/atm-20-1052. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Crawford A.M., Striano B.M., Giberson-Chen C.C., et al. Projected lifetime cancer risk associated with intraoperative computed tomography for lumbar spine surgery. Spine. 2023;48(13):893. doi: 10.1097/BRS.0000000000004685. [DOI] [PubMed] [Google Scholar]
- 8.Ji S., Fan X., Olson J.D., et al. In: Computational methods and clinical applications for spine imaging. Lecture notes in computer science. Vrtovec T., Yao J., Glocker B., Klinder T., Frangi A., Zheng G., Li S., editors. Springer International Publishing; Switzerland: 2016. Patient registration via topologically encoded depth projection images in spine surgery; pp. 27–37. eds. [Google Scholar]
- 9.Cai Y., Wu S., Fan X., et al. A level-wise spine registration framework to account for large pose changes. Int J Comput Assist Radiol Surg. 2021;16(6):943–953. doi: 10.1007/s11548-021-02395-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Kalfas I.H. Machine vision navigation in spine surgery. Front Surg. 2021;8 doi: 10.3389/fsurg.2021.640554. https://pubmed.ncbi.nlm.nih.gov/33738298/ [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Jakubovic R., Guha D., Gupta S., et al. High speed, High density intraoperative 3D optical topographical imaging with efficient registration to MRI and CT for craniospinal surgical navigation. Sci Rep. 2018;8(1) doi: 10.1038/s41598-018-32424-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Kumar A., Dwarakanath T.A., Bhutani G., Srinivas D., Moiyadi A. Autonomous frameless spine registration and robot-assisted surgery. Orthop Surg. 2026;18(7):1471–1490. doi: 10.1111/os.70342. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Manni F., Mamprin M., Holthuizen R., et al. Multi-view 3D skin feature recognition and localization for patient tracking in spinal surgery applications. BioMed Eng OnLine. 2021;20(1) doi: 10.1186/s12938-020-00843-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Zhang P., Shen Y., Giraud F., et al. Markerless inside-out tool tracking for endoscopic spine surgery: a benchmarking study and clinical dataset. Int J CARS. 2026 doi: 10.1007/s11548-026-03625-z. [DOI] [PubMed] [Google Scholar]
- 15.Stewart G. Visible light navigation in spine surgery: my experience with my first 150 cases. Int J Spine Surg. 2022;16(S2):S28–S36. doi: 10.14444/8274. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Abbasi H., Abbasi A., Abbasi H., Abbasi A. Using porcine cadavers as an alternative to Human cadavers for teaching minimally invasive spinal fusion: proof of concept and anatomical comparison. Cureus. 2019;11(11):11. doi: 10.7759/cureus.6158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Evans L., Olson J.D., Cai Y., et al. Stereovision Co-Registration in image-guided spinal surgery: accuracy assessment using explanted porcine spines. Oper Neurosurg. 2018;15(6):686–691. doi: 10.1093/ons/opy023. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Fan X., Mirza S.K., Li C., Evans L.T., Ji S., Paulsen K.D. Accuracy of stereovision-updated versus preoperative CT-based image guidance in multilevel lumbar pedicle screw placement. JB JS Open Access. 2022;7(1) doi: 10.2106/JBJS.OA.21.00129. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Duke R., Fan X., Ji S., Mirza S.K., Paulsen K.D. Vol. 11598. Robotic Interventions, and Modeling; Bellingham, Washington: 2021. Semi-automated calibration of a hand-held stereovision system using a tracked calibration frame for image guided surgery; pp. 517–523. (Medical imaging 2021: image-guided procedures). [Google Scholar]
- 20.Fan X., Durtschi M.S., Li C., Ji S., Mirza S.K., Paulsen K.D. Calibration of a hand-held stereovision system for image-guided spinal surgery. Medical imaging 2019: image-guided procedures. 2019;10951:497–503. [Google Scholar]
- 21.Fan X., Durtschi M.S., Li C., et al. Hand-held stereovision system for image updating in open spine surgery. Oper Neurosurg. 2020;19(4):461. doi: 10.1093/ons/opaa057. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Ji S., Fan X., Roberts D.W., et al. Springer International Publishing; Bellingham, Washington: 2013. Machine learning in medical imaging. Lecture notes in computer science; pp. 106–113. [Google Scholar]
- 23.Fedorov A., Beichel R., Kalpathy-Cramer J., et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging. 2012;30(9):1323–1341. doi: 10.1016/j.mri.2012.05.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Hsieh W.W. Nonlinear principal component analysis by neural networks. Tellus A. 2001;53(5):599–615. [Google Scholar]
- 25.Horn B.K.P. Closed-form solution of absolute orientation using unit quaternions. J Opt Soc Am A. 1987;4(4):629–642. [Google Scholar]
- 26.Warner W.R., Fan X., Duke R.B., et al. Vol. 12034. Robotic Interventions, and ModelingSPIE; Bellingham, Washington: 2022. Preoperative-to-interoperative shift in spine pose measured as change in lordosis Cobb angle and its effect on navigational accuracy; pp. 542–547. (Medical imaging 2022: image-guided procedures). [Google Scholar]
- 27.Gertzbein S.D., Robbins S.E. Accuracy of pedicular screw placement in vivo. Spine (Phila Pa 1976) 1990;15(1):11–14. doi: 10.1097/00007632-199001000-00004. [DOI] [PubMed] [Google Scholar]
- 28.Glossop N., Hu R. Assessment of vertebral body motion during spine surgery. Spine. 1997;22(8):903. doi: 10.1097/00007632-199704150-00014. [DOI] [PubMed] [Google Scholar]
- 29.Guha D., Jakubovic R., Gupta S., et al. Intraoperative error propagation in 3-dimensional spinal navigation from nonsegmental registration: a prospective cadaveric and clinical study. Glob Spine J. 2019;9(5):512–520. doi: 10.1177/2192568218804556. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Warner W.R., Fan X., Duke R.B., et al. Vol. 12466. Robotic Interventions, and Modeling; Bellingham, Washington: 2023. Towards accounting for intraoperative spine motion: a simulation study of registration between stereovision surfaces; pp. 350–357. (Medical imaging 2023: image-guided procedures). [Google Scholar]
- 31.Segal AV, Haehnel D, Thrun S. Generalized-ICP. 2010. doi:10.7551/mitpress/8727.003.0022.
- 32.Yang J., Li H., Campbell D., Jia Y. Go-ICP: a globally optimal solution to 3D ICP point-set registration. IEEE Trans Pattern Anal Mach Intell. 2016;38(11):2241–2254. doi: 10.1109/TPAMI.2015.2513405. [DOI] [PubMed] [Google Scholar]
- 33.Duke R., Fan X., Warner W.R., et al. Vol. 12928. Robotic Interventions, and Modeling; Bellingham, Washington: 2024. Simulation study of minimally invasive surgical scenes and their effect on hand-held stereovision driven level-wise registration techniques; p. 408. (Medical imaging 2024: image-guided procedures). [Google Scholar]
Further readings
- 34.Sklar M., Fatemi P., Wadhwa H., Leung C., Zygourakis C. First case report using optical topographic-guided navigation in revision spinal fusion for calcified thoracic disk. J Clin Neurosci. 2021;91:80–83. doi: 10.1016/j.jocn.2021.06.031. [DOI] [PubMed] [Google Scholar]
- 35.Kwak S.G., Kim J.H. Central limit theorem: the cornerstone of modern statistics. Korean J Anesth. 2017;70(2):144–156. doi: 10.4097/kjae.2017.70.2.144. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.