Although robotic spine systems can offer many advantages, including reduced radiation exposure and decreased invasiveness, the increment in precision remains essential in favour of their use.
In the spine, there are several methods to measure implant accuracy. Some use the criteria of "in" or "out" 6, others use pedicle breach analysis 7–9, and yet others attempt to quantify the amount of facet joint violation 10. Although these approaches can correlate clinically with patient outcomes, they do not clearly define accuracy. It is reasonable to use the "breach" concept (Gertzbein-Robbins scale) to check accuracy when the implant is inserted in a broad pedicle (> 7mm). In this context, for a 5mm diameter PS to still be considered acceptable (< 2mm breach), it is necessary to have no more than a 3mm shift from the original trajectory. However, in a narrow pedicle (< 4mm), the same 3mm shift is not an appropriate measure. This qualitative method is vague and cannot be applied to all patients and spinal levels. In discordance with this method, we used a spatial orientation or quantitative approach to compare PS accuracy in the current study.
Our experiment measured preoperative and postoperative screw distances and angles in two anatomic planes. A variation of this method has been used before 11,12, but we added new features which offer several advantages. To begin with, ours is the first to assess the performance of robotic instrumentation. Second, comparing planned preoperative trajectories with postoperative measurements allowed us to extrapolate Postoperative Screw Error Patterns. This involved using anatomically derived measures of screw position, which described 3D screw trajectories, and analyzing how final screw placements deviated from the planned trajectories. By classifying trajectory errors, it allows robot calibration, avoiding future deviances. Another benefit of our measurement methodology is that we used the contralateral pedicle as one of the measures in the axial plane, as opposed to the more commonly used ipsilateral pedicle 7–9. Several authors described how the implants' metallic artifact sometimes obscures the screw boundaries, making pedicle breach assessment almost impossible 13. None of our screw's metallic artifacts tampered with the measurement using the contralateral pedicle as a reference.
The discussion of how to define pedicle screw accuracy and its importance is sparse in the literature. Rampersaud et al. are among the few groups addressing this question 14. In their paper, accuracy requirements differ depending on the spine level. These requirements often exceed the accuracy of current image-guided surgical systems based on clinical utility errors reported in the literature. In their conclusion, the authors state that maximum permissible translational/rotational error tolerances ranged from 0.0 mm/0.0° at T5 to 3.8 mm/12.7° at L5. In our cadavers, we divided the non-optimal PS into two classes: Unacceptable (> 2.0mm for distance and > 5º for angulation) and Inaccurate (between > 0.5mm and ≤ 2.0mmn and > 2.5º and ≤ 5º, respectively). Ideally, a robotic system should produce no noticeable screw trajectory errors, but technology has yet to mature. In addition, it is challenging to measure screw trajectories from CT images without a repeatability error of less than 0.5 mm or 2.5º. To minimize this error, we measured preoperative and postoperative images using the same planning software. A third-party application was considered less consistent.
A few others have used the quantitative concept to measure their implants utilizing direct distances and angles to compare planned and postoperative images 11,12,15. Ortel et al. quantified only the axial angle using a "midsagittal line" passing in the middle of the vertebra, taking the spinous process as a reference 15. This line is not always straightforward, especially in cases where the vertebrae are asymmetrical. Instead, we used the vertebral canal and body as a reference; this method was less affected by anatomical irregularities. Kleck and al. also measured the axial angle but included a direct distance from the entry point to the tip of the screw 11. Although valid, this does not give enough information to fully represent the screw position in both preoperative and postoperative images. Guha et al. measured both axial and sagittal angles 12. They included the distance between the screw entry point and the mid-sagittal line (bisecting the vertebral body, spinal canal, and spinous process). Although their technique provides some information about the 3D orientation of the screw, it does not specify the screw's tip coordinates inside the vertebra. We demarcated the precise spatial orientation inside the vertebra by considering the screw’s proximal and distal directions. The 3D anatomical definition was crucial to characterize the screw accuracy.
In our study, the incidence of unacceptable screws in our sample (10.8%) was consistent with other studies 3,4,16–19, which used the Gertzbein-Robbins scale for measuring accuracy 8. However, our data added new insights when labelling implants as inaccurate. Based on our results, only 43.2% of the PS were found to precisely replicate the preoperative plan, with the remaining placed screws being either inaccurate (37.8%) or wholly unacceptable (18.9%). This is the first time a study has exposed the limits of the current robotic technology, as suggested by some authors 14. In addition, our study protocol allowed us to analyze the accuracy of different registration techniques (O-arm and preopCT/C-arm).
In numerous articles, o-arm accuracy has been compared against fluoroscopy; overall, it performs better 16,20–23. However, its efficacy in robotic surgery is still to be defined. Two studies tested accuracy using the Gertzbein-Robbins scale and found no difference between the O-arm and preopCT/C-arm registration techniques for unacceptable screws (> grade I) 5,24. Our results had a different outcome. In our series, the O-arm was superior to the preopCT/C-arm. It is essential to point out that the upper cervical screw angle is particularly challenging for any technique, navigated or not. They are usually very cranial (high sagittal) and medial (high axial) in angle. Moreover, because of the high sagittal degree, the navigated dynamic reference array (sphere trackers) can be easily blocked from the navigation camera. So, theoretically, the most challenged screws were placed using the O-arm method and still yielded more accurate results. The reasons may be linked to the acquisition process. The preoopCT/C-arm method requires imaging fusion and matching, and the algorithm behind it depends on bone density. So the merging process between the preopCT and the C-arm images may be affected by osteoporosis. In fact, some authors have described that the robotic system cannot recognize the vertebral anatomy from the poor-quality intraoperative fluoroscopic 2,19. In our sample, the S1 screw of cadaver W was not inserted for this reason. Based on these data, viewing the O-arm as the superior method is not difficult, but a confirmatory investigation is required.
In terms of surgical access, apart from the advantages or disadvantages of the two techniques, some scientific studies claim that open exposure can tamper with screw accuracy when soft tissue retraction is not optimal or if soft tissues encumber the surgical robotic arm 2,3,25. Our results comparing percutaneous and open approaches may have been influenced by inadequate exposure to robotic standards. When considering all screws, irrespective of the registration method, the less invasive procedure (percutaneous) had better performance. For example, we noticed inappropriate muscular exposure for the L3 and L4 screws during the cadaver W surgical procedure. This may have interfered with the postoperative result of these two implants. In part, our results were similar to Kantelhardt et al. 25. Although both open and percutaneously placed screws have not generally differed in their study, some significance was found towards less invasiveness, but only for implants entirely inside the bone. In our case, the overall O-arm superiority may have influenced the better percutaneous performance. Unfortunately, we did not use open access with the O-arm to confirm these results. Nevertheless, the results were similar when both open and percutaneous access were compared for preopCT/C-arm registration.
Our study has several limitations. Despite what some consider a positive factor 8 lacks inter-investigator variability. Second, the small sample size limits the broad generalizability of the results, but by adding several measurements per screw, our results reached significance. It is essential to mention that although our group has approximately 150 hours of robotic cadaver training, our learning curve has not yet plateaued.