Image-guided surgery is poised to become an increasing part of standard operative care because it improves surgical precision and reduces trauma from surgical exposure1,2. In spinal surgery, it facilitates complex procedures, enables minimally invasive interventions, reduces radiation exposure, and increases the accuracy of pedicle screw placement3,4. Most spinal surgery still involves open exposure, and <15% of surgeons use navigation routinely3,4, in part because it increases procedure duration and cost, provides unreliable accuracy in some cases, and involves complex equipment and additional training3,4. Image guidance must transform the surgical space to image space accurately; the process is compromised by any change in vertebral position during surgery5. Motion can be compensated for by registering each vertebra individually. Here, segmental registration and attaching the reference frame to the level of interest are common in cases in which accuracy becomes critical (e.g., small pedicles in the thoracic or cervical spine or L1). However, these processes take time and can be cumbersome for some surgeons. Intraoperative fluoroscopic computed tomography (CT), such as use of the O-arm (Medtronic), provides accurate mapping and automated registration1, but it adds radiation exposure (to the patient and, potentially, the surgical team) and operative time6,7. The 7D Surgical image-guided technology registers the surgical field with preoperative CT (pCT) using anatomical landmarks, but it involves structured light equipment and manual selection in both pCT and the surgical field, and limits navigation to 1 vertebra at a time. The efficacy of intraoperative magnetic resonance imaging (MRI) is limited by availability, the lack of MR-compatible instruments and equipment8, and imaging artifacts caused by instrumentation in the surgical field9.
We developed a novel method of compensating for spinal motion between preoperative scans and intraoperative positioning, and registering the surgical field automatically on the basis of intraoperative photographs acquired with a handheld camera system rather than more expensive structured light equipment or repeated CT scans with ionizing radiation10,11. This optically tracked, radiation-free, handheld intraoperative stereovision (iSV) device acquires depth-of-field data of exposed vertebral surfaces12. It uses a validated deformable registration to align each vertebra with pCT automatically and without the manual identification of landmarks. The process modifies pCT data to reflect each vertebra’s current position10,11.
In this study, we measured the magnitudes of motion-induced errors using a state-of-the-art commercial spinal navigation system during in vitro pedicle screw placement in whole-pig cadavers and compared the results to those of iSV registration, which compensated for intraoperative vertebral motion.
Materials and Methods
We designed the study to compare registration performance with conventional image guidance (StealthStation i7; Medtronic) and our iSV system under conditions of accentuated vertebral motion to determine how large registration errors can become with standard methods and to assess the clinical need for updated images and whether our iSV approach can compensate for the induced misalignment. We recognize that the study was not comparing expected clinical performance of standard image guidance with our new technology per se because strategies to mitigate registration errors from vertebral motion with standard methods, for example vertebra-by-vertebra re-registration, were not evaluated. We used 6 cadaveric whole-pig specimens because of their anatomical similarities with human lumbar spines13. The sample size was based on variations observed in previous experiments. No inclusion/exclusion criteria for specimens were established. A surgical procedure flowchart is shown in Figure 1. Both StealthStation and iSV registrations were performed once at the beginning of the procedure. The guidance system used for each screw was randomly assigned, and the surgeon was blinded to the guidance system used for each screw. The same experienced spine surgeon placed pedicle screws in lumbar vertebrae and was instructed to suspend clinical judgment and place screws based solely on image guidance. We compared registration errors, the accuracy of screw position, and efficiency (the time required for registration) of the 2 approaches.
Standard midline posterior lumbar exposure was performed from L1 to L6. Eighteen mini-screws (1.5-mm diameter; Stryker) were implanted as fiducial markers in the spinous and transverse processes at each level. The incision was closed, the animal was positioned supine with accentuated lumbar lordosis at L3, and pCT was acquired (pixel spacing: 0.21 × 0.21 to 0.35 × 0.35 mm; slice thickness: 0.6 mm). The animal was repositioned prone and secured to the operating table to minimize intraoperative alignment change. The surgical site was reopened. A dynamic reference frame (Medtronic) was attached rigidly to the ilium. Ground-truth locations of 18 fiducials were digitized with a tracked stylus (Medtronic).
Fiducial-based registration (FBR) was performed on the StealthStation with pCT using the same 10 (the maximum number of registration points allowed by StealthStation) of 18 fiducials to align the entire lumbar exposure. The 10 fiducial locations were distributed evenly across the lumbar spine following literature suggestions14. We used fiducials instead of anatomical landmarks typical of clinical cases to minimize localization errors.
The handheld iSV12,15 acquired 3 to 4 image pairs with tracking data to reconstruct a complete 3D profile of the exposed spine. Figure 2 shows a photograph of iSV acquisition in a typical experiment. The accuracy of iSV reconstruction was assessed as the distances between tracked fiducial locations and their counterparts localized on iSV surfaces. Preoperative CT was registered with iSV surfaces using a deformable registration (illustrated in Fig. 3, technical details published elsewhere10,11). An updated CT (uCT) image volume was generated by deforming the pCT, and uploaded to the StealthStation.
Pedicle Screw Insertion
Pedicle screws of the same size (Vertex Select; Medtronic) were inserted on each vertebra based on StealthStation image guidance using pCT or iSV-updated image guidance using uCT, respectively, although screw diameters varied between cases because of the limited availability of instruments for animal cadaveric studies. The sequence of screw insertions was randomized. A second randomization was employed to choose the registration method. For each insertion, a corresponding registration and scan were loaded into the StealthStation for image guidance: pCT for StealthStation registration (Fig. 4-A) and uCT for iSV registration (Fig. 4-B).
Assessment of Accuracy
We assessed the point-to-point registration error (ppRE) of 18 fiducials, calculated as the distances between tracked fiducial locations in physical space and their counterparts as transformed from image to physical space. With respect to StealthStation measurements, the fiducial registration error (FRE)16 was calculated from the 10 fiducials involved in StealthStation registration, and the target registration error (TRE)16 was calculated using the other 8 fiducials. In addition, the ppRE was calculated using the 3 fiducials on each level. The ppRE of iSV registration is equivalent to its TRE since the 18 fiducials were not involved in iSV registration, whereas the ppRE of StealthStation registration is a combination of its FRE and TRE.
Postoperative CT images (pixel spacing: 0.21 × 0.21 to 0.31 × 0.31 mm; slice thickness: 0.6 mm) were acquired to visualize screw positions. Subsequently, specimens were dissected at the instrumented levels, and perforation of each implanted pedicle screw was measured in medial-lateral and superior-inferior directions6. Anterior perforation was not measured because of the limited selection in pedicle-screw lengths. Perforation severity grades were assigned according to thresholds defined by Gertzbein and Robbins17: 0 mm (no breach) = Grade 0, a perforation distance of <2 mm = Grade 1, 2-4 mm = Grade 2, and >4 mm = Grade 3. If the perforation distance was not measurable with calipers but screw threads were visible upon dissection, Grade 0 was assigned. Figure 5 shows representative images of medial (Fig. 5-A) and lateral (Fig. 5-B) perforation, and Grade 0 perforation with no visible breach (Fig. 5-C) and with a visible but unmeasurable breach (Fig. 5-D). We also investigated relationships between perforation magnitude and (1) screw location (vertebral level), (2) sequence of insertion, and (3) the ppRE at the vertebral level. The Spearman rank correlation coefficient, ρ, was calculated to assess associations between these variables, and a p value was found in testing a no-correlation hypothesis. All data analyses were performed in MATLAB (MathWorks).
Source of Funding
This study was funded by the National Institutes of Health (grant R01EB025747-01). Medtronic Navigation (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 start-up companies, InSight Surgical Technologies and PEER Technologies.
Case characteristics are summarized in Table I. Pedicle widths were measured on pCT (column 3: range, 2.7 to 8.5 mm) and were narrower than those of humans as reported in the literature (range, 6.4 to 17.5 mm18). Pedicle screw diameters are reported in column 4 of Table I.
TABLE I -
Spinal Levels Instrumented, Range of Pedicle Width, and Size of Inserted Screws
||Pedicle Width (mm)
||Screw Diameter (mm)
*The number of levels instrumented for each case is reported in parentheses.
The accuracy of StealthStation and iSV registrations is reported in Table II. The FREs and TREs of StealthStation registration appear in columns 2 and 3, respectively. Points with large registration errors were excluded by the StealthStation in its FRE calculations (the number of excluded points is reported in parentheses in column 2). The TRE (average [and standard deviation], 8.37 ± 1.76 mm) was larger than the FRE (average, 4.40 ± 1.35 mm) in StealthStation registration. The TRE of iSV registration (column 4 in Table II) (average, 2.81 ± 0.91 mm) was smaller than both the FRE (column 2) and TRE (column 3) of StealthStation registration. The overall iSV reconstruction accuracy across the 6 cases (column 5) was 1.22 ± 0.19 mm, indicating that iSV data acquisition was accurate. The change in lumbar lordosis between preoperative imaging and the intraoperative surgical position was measured using the Cobb method19 and is reported in column 6 of Table II.
TABLE II -
Accuracy of StealthStation (FRE and TRE) and iSV (TRE) Registrations and Reconstructed iSV Surfaces, and Change in Lumbar Lordotic Angle Between Preoperative Supine Imaging and Intraoperative Prone Position*
||Reconstruction Accuracy of iSV†(mm)
||Lumbar Lordotic Angle Change‡(°)
||7.38 ± 3.50
||2.34 ± 0.81
||1.55 ± 0.54
||7.86 ± 4.22
||3.93 ± 1.47
||1.16 ± 0.31
||7.38 ± 2.98
||2.87 ± 1.16
||0.96 ± 0.34
||8.80 ± 4.49
||2.51 ± 2.34
||1.16 ± 0.48
||7.06 ± 3.75
||3.71 ± 0.77
||1.22 ± 0.44
||11.74 ± 6.02
||1.48 ± 0.95
||1.27 ± 0.43
||4.40 ± 1.35
||8.37 ± 1.76
||2.81 ± 0.91
||1.22 ± 0.19
||48 ± 13
*iSV = intraoperative stereovision, FRE = fiducial registration error, and TRE = target registration error.
†The values are given as the mean and standard deviation. The number of points excluded from StealthStation registration is reported in parentheses in column 2.
Change between preoperative imaging and intraoperative surgical position, as measured using the Cobb method19
We also compared the ppREs of StealthStation and iSV registrations at all 18 fiducial locations and at each vertebral level (Fig. 6-A and 6-B, respectively). The overall ppREs of iSV registration were smaller than those of StealthStation registration at all 18 fiducial locations as well as at all 6 levels. Furthermore, the ppREs of StealthStation registration were larger toward the ends of the exposed spine (L1 average, 16.45 ± 3.32 mm; L6 average, 12.19 ± 5.37 mm), whereas the ppREs from iSV registration were distributed evenly across all screws and all levels. The ppRE for L6 (average, 3.68 ± 0.85 mm) was larger than at other levels. The standard deviations show that the ppRE of StealthStation registration (range, 4.09 to 7.52 mm) was more variable than that of iSV registration (range, 0.81 to 2.50 mm).
Pedicle Screw Position
Sixty-eight pedicle screws were inserted (34 using each registration method). The distributions of pedicle screws across perforation severity grades are shown in Figure 7 for StealthStation and iSV registrations. The results show that iSV registration had a higher rate of Grade 0, similar rates of Grades 1 and 2, and a lower rate of Grade 3 perforations relative to StealthStation registration.
All screws from both registrations had Grade 0 perforation in superior-inferior directions. In the medial direction (Fig. 7-A), iSV registration had a higher rate of Grade 0 screws than StealthStation (97% versus 68%) and lower rates for all other grades. StealthStation registration had a high rate of Grade 3 perforation (1 screw in each case; range: 4.5 to 14.0 mm). In the lateral direction (Fig. 7-B), iSV registration had higher rates of Grades 1 and 2 perforation (6% and 2 screws in each grade). All Grade 1 and 2 screws resulted from Case 3, in which the pedicle width was smaller (2.9 to 4.2 mm). Neither registration had lateral Grade 3 perforation. Overall performance is shown in Figure 7-C.
Spearman correlation results show that perforation from iSV registration was associated with the sequence of insertion (ρ = 0.37; p = 0.03), whereas StealthStation registration was not correlated with the sequence of insertion (ρ = −0.04; p = 0.79). Figure 8 shows the relationships between perforation magnitude and screw location (Fig. 8-A), sequence of insertion (Fig. 8-B), and ppRE at the vertebral level (Fig. 8-C) for StealthStation and iSV registrations. Figure 8-A shows that StealthStation registration had larger perforation toward the ends (L1 and L6), whereas perforation from iSV registration was more evenly distributed across all levels. Figure 8-B shows perforation ordered by sequence of insertion (horizontal axis). Figure 8-C shows that perforation was severe with a large ppRE, and Spearman correlation shows a strong association (ρ = 0.38; p = 0.001) when all data points were analyzed.
The overall time required for iSV registration (computational efficiency) was ∼10 to 15 minutes. The overall cost in time of FBR on the StealthStation was similar (∼10 minutes), and involved manual selection of homologous points on CT scans and in the surgical field.
Registration accuracy was assessed using fiducials, but errors at fiducials may not represent the accuracy at pedicles. We implanted pedicle screws on the basis of image guidance only and measured the perforation of pedicle screws to assess registration performance. The natural anatomy of the pedicle canal may have compensated for some errors, as the cortical margin of the pedicle may redirect the tap and screw during placement. If so, the correction would affect both iSV and StealthStation systems equally, and our results show that iSV registration outperformed StealthStation registration in terms of medial/lateral perforations under the experimental conditions in the study. Severe medial perforation (Grade 3) was observed in 6 of 8 pedicle screws with a large ppRE (>10 mm; all from StealthStation registration).
We also found that perforation was associated with the sequence of insertion in iSV registration (Fig. 8-B). One possible explanation is that additional alignment change was introduced during pedicle screw insertion. Compared with other techniques, iSV registration is low-cost and efficient, and can be repeated during surgery to account for recurring alignment changes without radiation exposure.
FRE and TRE depend on the severity of alignment change in FBR. For reference, we acquired intraoperative CT (iCT) in the operative prone position and performed FBR with iCT, which served as the “gold standard” because the image stack matched the intraoperative position exactly20. The accuracy of iCT-based registration was superb (0.49 ± 0.07 mm), and any remaining errors were mainly due to localization effects (e.g., tracking system errors and human errors in localizing fiducials in image and physical space). StealthStation registration using pCT was subject to similar localization effects, but TREs were caused largely by posture changes between supine imaging and prone positioning (average change in lumbar lordosis, 48° ± 13°). Here, we only investigated multilevel procedures, and observed that alignment change was larger and ppREs were larger toward the ends of spine exposures. Since fiducials were evenly distributed, registration points toward exposure ends were often excluded by the StealthStation because of their large errors. As a result, the TRE was larger than the FRE.
StealthStation accuracy was lower relative to published case series6,21,22 and systematic reviews/meta-analyses23,24 because we deliberately evaluated errors that resulted when vertebral motion between imaging and surgical position was not addressed with conventional registration systems. Discrepancies between the StealthStation and iSV would likely be reduced if individual vertebrae were registered on the StealthStation as suggested by others5, or with shorter segments25. For example, ppRE can be improved to 1.92 ± 0.50 mm by registering only 3 levels and limiting navigation to 1 level at a time. However, for some surgeons, and perhaps those new to navigation, repeated segmental registration can be cumbersome and increase operative time, and the resulting accuracy is surgeon-dependent26. In this study, intervertebral motion caused a range of pedicle breaches with the StealthStation when the whole lumbar spine was registered. Intraoperative stereovision registration, on the other hand, was able to compensate for this motion to a large extent and navigate the entire surgical field. It avoided the repositioning of instruments, requiring multiple level-wise registrations, and the repeating of CT scans. We plan to compare the performance of iSV registration with segmental registration and iCT registration in a future study.
Several limitations must be acknowledged. First, iSV registration is only possible on exposed levels. Levels at exposure ends are more difficult to acquire due to line-of-sight restrictions, and errors can be higher. The issue is mitigated by positioning instruments away from the vertebrae. Second, soft-tissue and ligamentous structures remaining on the dorsal surface of the vertebrae, which were not present in CT-segmented spines, likely contributed to errors in iSV registration. Although injury could not be assessed here, a previous report18 suggested that distances between the pedicle cortex and neural structures were 1.7 to 2.0 mm medially, and 2.4 (L5) to 9.6 (L1) mm laterally, in the lumbar section. We have since improved our algorithms and reduced the overall ppRE to <2.0 mm (average, 1.94 ± 0.37 mm; range, 1.26 to 2.24 mm) in the 6 cadavers to meet clinical acceptance criteria. Third, a long and wide exposure was made in this study to reveal the entire lumbar section, including partial transverse processes. Simulations show that registration accuracy is similar for various lengths of exposure27. Preliminary data indicate that a narrower exposure (up to the facet joints) with reduced muscle stripping and tissue damage may still be sufficient for iSV registration, but accuracy can be affected. Effort is underway to evaluate iSV image updating performance as a function of exposure size. Finally, the exposed spine was selected manually to remove irrelevant background from iSV images, but that required minimal expertise, and the reconstruction and registration processes were automatic. Effort is underway to develop machine learning-based algorithms to segment the spine automatically from iSV images to eliminate user input.
The iSV technique may be applicable to open exposure, the cervical spine, and uninstrumented revision surgery, where landmarks are not clearly available or posterior structures have been removed in the primary surgery. For instrumented revisions, further investigation would be needed to demonstrate the adequacy of using pCT imaging with hardware artifacts. The technique also has possible clinical applications for navigation in nonspinal surgery.
Note: The authors thank John Peiffer, Michaela Whitty, Michael Pearl, Robert Ferranti Jr., and Theresa Haron from the Center for Surgical Innovation, and Dr. Vyacheslav Makler from Neurosurgery at Dartmouth-Hitchcock Medical Center for their assistance with data collection, and William R. Warner at Dartmouth College for assistance with data analysis.
1. Holly LT, Foley KT. Image guidance in spine surgery. Orthop Clin North Am. 2007 Jul;38(3):451-61, abstract viii.
2. Peters TM. Image-guidance for surgical procedures. Phys Med Biol. 2006 Jul 21;51(14):R505-40.
3. Härtl R, Lam KS, Wang J, Korge A, Kandziora F, Audigé L. Worldwide survey on the use of navigation in spine surgery. World Neurosurg. 2013 Jan;79(1):162-72.
4. Choo AD, Regev G, Garfin SR, Kim CW. Surgeons’ perceptions of spinal navigation: analysis of key factors affecting the lack of adoption of spinal navigation technology. SAS J. 2008 Dec 1;2(4):189-94.
5. Holly LT, Foley KT. Intraoperative spinal navigation. Spine (Phila Pa 1976). 2003 Aug 1;28(15)(Suppl):S54-61.
6. Mirza SK, Wiggins GC, Kuntz C 4th, York JE, Bellabarba C, Knonodi MA, Chapman JR, Shaffrey CI. Accuracy of thoracic vertebral body screw placement using standard fluoroscopy, fluoroscopic image guidance, and computed tomographic image guidance: a cadaver study. Spine (Phila Pa 1976). 2003 Feb 15;28(4):402-13.
7. 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 Jun;20(6):875-81.
8. Woodard EJ, Leon SP, Moriarty TM, Quinones A, Zamani AA, Jolesz FA. Initial experience with intraoperative magnetic resonance imaging in spine surgery. Spine (Phila Pa 1976). 2001 Feb 15;26(4):410-7.
9. Rutherford EE, Tarplett LJ, Davies EM, Harley JM, King LJ. Lumbar spine fusion and stabilization: hardware, techniques, and imaging appearances. Radiographics. 2007 Nov-Dec;27(6):1737-49.
10. Cai Y, Olson JD, Fan X, Evans LT, Paulsen KD, Roberts DW, Mirza SK, Lollis SS, Ji S. Automatic geometric rectification for patient registration in image-guided spinal surgery. Med Imag Image-Guid Proc Robot Interv Mod. 2016:9786.
11. Cai Y, Wu S, Fan X, Olson J, Evans L, Lollis S, Mirza SK, Paulsen KD, Ji S. A level-wise spine registration framework to account for large pose changes. Int J Comput Assist Radiol Surg. 2021 Jun;16(6):943-53.
12. Fan X, Durtschi MS, Li C, Evans LT, Ji S, Mirza SK, Paulsen KD. Hand-Held Stereovision System for Image Updating in Open Spine Surgery. Oper Neurosurg (Hagerstown). 2020 Sep 15;19(4):461-70.
13. 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 Nov 14;11(11):e6158.
14. West JB, Fitzpatrick JM, Toms SA, Maurer CR Jr, Maciunas RJ. Fiducial point placement and the accuracy of point-based, rigid body registration. Neurosurgery. 2001 Apr;48(4):810-6, discussion 816-7.
15. Fan X, Durtschi MS, Li C, Ji S, Mirza SK, Paulsen KD. Calibration of a hand-held stereovision system for image-guided spinal surgery. Med Imag Visual Image-Guid Proc Mod. 2019:10951.
16. Fitzpatrick JM. Fiducial registration error and target registration error are uncorrelated. Med Imag Visual Image-Guid Proc Mod. 2009;7261.
17. Gertzbein SD, Robbins SE. Accuracy of pedicular screw placement in vivo. Spine (Phila Pa 1976). 1990 Jan;15(1):11-4.
18. Lien SB, Liou NH, Wu SS. Analysis of anatomic morphometry of the pedicles and the safe zone for through-pedicle procedures in the thoracic and lumbar spine. Eur Spine J. 2007 Aug;16(8):1215-22.
19. Okpala FO. Comparison of Four Radiographic Angular Measures of Lumbar Lordosis. J Neurosci Rural Pract. 2018 Jul-Sep;9(3):298-304.
20. Tormenti MJ, Kostov DB, Gardner PA, Kanter AS, Spiro RM, Okonkwo DO. Intraoperative computed tomography image-guided navigation for posterior thoracolumbar spinal instrumentation in spinal deformity surgery. Neurosurg Focus. 2010 Mar;28(3):E11.
21. Fu TS, Wong CB, Tsai TT, Liang YC, Chen LH, Chen WJ. Pedicle screw insertion: computed tomography versus fluoroscopic image guidance. Int Orthop. 2008 Aug;32(4):517-21.
22. Choi WW, Green BA, Levi ADO. Computer-assisted fluoroscopic targeting system for pedicle screw insertion. Neurosurgery. 2000 Oct;47(4):872-8.
23. Tian NF, Xu HZ. Image-guided pedicle screw insertion accuracy: a meta-analysis. Int Orthop. 2009 Aug;33(4):895-903.
24. Bourgeois AC, Faulkner AR, Pasciak AS, Bradley YC. The evolution of image-guided lumbosacral spine surgery. Ann Transl Med. 2015 Apr;3(5):69.
25. Papadopoulos EC, Girardi FP, Sama A, Sandhu HS, Cammisa FP Jr. Accuracy of single-time, multilevel registration in image-guided spinal surgery. Spine J. 2005 May-Jun;5(3):263-7, discussion :268.
26. Ughwanogho E, Flynn JM. Current Navigation Modalities in Spine Surgery. University of Pennsylvania Orthopaedic Journal. 2010;20:65-69.
27. Fan X, Durtschi MS, Li C, Evans LT, Ji S, Cai Y, Mirza SK, Paulsen KD. Stereovision-updated image guidance in multi-level open spine surgery: short vs. long exposure. Med Imag Imag-Guid Proc Robot Intervent Mod. 2020:11315.