This study is the first description of a fully automated algorithm for determination of the technical accuracy of pedicle screw placement that does not require human intervention in the rigid-body alignment or measurement process. The technical accuracy results from this study of the Mazor X Stealth robotic system are equivalent to or better than reported accuracy from any prior in vivo or cadaver study [5, 19, 20, 26, 27, 32, 33, 38,39,40,41,42]. At the mid-pedicle position, our MAE was 0.75 mm in the ML direction, 0.60 mm in the SI direction, and 1.04 mm in total. Our total angular error of 1.58° was also superior to any prior cadaver or in vivo study [5, 19, 26, 27, 29, 32, 33, 38,39,40,41,42].

Prior to the development of technical accuracy measurement systems, the G–R classification system served a useful role providing categorical accuracy and safety data. If we accept the premise that Grade A and Grade B are clinically acceptable, then only computer-assisted techniques (freehand navigation and robotically assisted navigation) consistently achieve the goal of 100% Grade A and B results (Table 3) [5, 19, 20, 26, 27, 32, 33, 38,39,40,41,42]. The current cadaveric study supports this conclusion, with all screws identified on CT as Grade A (88.6%) or B (11.4%), although only 3 screws were confirmed as breached after open dissection (8.4% false-positive rate).

The difficulties assessing pedicle breach and differentiating between Grade A and Grade B screws secondary to scatter cannot be underestimated, as well as observer bias and lack of a standardized approach to assessing the CT scans. It is likely that prior studies of breach based upon CT evaluation may also suffer from high false-positive rates; however, it is most likely to impact review of pedicle screws with less than 2 mm of breach, which are generally deemed clinically acceptable.

It is difficult to assess the technical accuracy of freehand techniques since freehand surgeons generally do not obtain preoperative 3D imaging or plan their screws on 3D planning software. However, a recent non-consecutive retrospective study of freehand pedicle screw categorical accuracy in 318 pediatric spinal deformity patients with 6,358 screws reported 2.63% of the pedicle screws were Grade C or worse, and 0.26% of screws necessitated UPROR [6]. Another non-consecutive multicenter retrospective review reported 0.26% incidence of both neurologic injury and misplaced instrumentation [7]. A single-center retrospective review of all pediatric patients who underwent spinal fusion over a 30-year period revealed a 1.1% incidence of UPROR related to malpositioned pedicle screws, neurological changes, or pneumothorax (presumably related to implants) [43]. Meta-analyses report overall accuracy rates between 90.6% and 94.9% for freehand and freehand navigation techniques [44, 45]. Collectively, these studies provide useful baseline data for accuracy and revision rates for pedicle screws using non-robotic techniques. The retrospective and non-consecutive nature of the pediatric multicenter database studies limit their utility for comparison with consecutive series of computer-navigated surgical accuracy in vivo and single-center consecutive case reviews [5, 19, 20, 26, 27, 32, 33, 38,39,40,41,42,43].

The advent of computer-assisted surgical navigation has fostered the development of technical accuracy, namely geometric comparisons between the 3D preoperative CT plan and the postoperative CT scan. Freehand navigation and robotically assisted navigation studies have variably described the technical accuracy of pedicle screw placement with increasing sophistication, but without any consensus on terminology or analysis methodology [5, 13,14,15,16,17, 19, 20, 22, 26,27,28,29, 32, 33, 38, 40,41,42]. The protocol described in this work is greatly influenced by a small number of studies utilizing either cadavers, bone models, or in vivo human studies which have iteratively improved the granular reporting of 3D technical accuracy (Table 3) [5, 19, 20, 26, 27, 32, 33, 38,39,40,41,42]. The largest studies of computer-assisted freehand and robotic navigation report total angular errors between 2.0 and 6.3 degrees compared with 1.58 degrees in the current study. Only one study by Volk et al. evaluated mid-pedicle error, which was 1.75 mm in the ML direction and 1.52 mm in the SI direction, compared to 0.75 mm and 0.60 mm in the current study [26].

A combination of standardized clinical safety data (e.g., G–R classification) and technical accuracy including precision is necessary to properly compare different navigation systems and workflows. We advocate a screw-aligned coordinate system as the natural reference frame to present useful feedback for surgeons, and we present “target analysis” visualizations (Fig. 4) to illustrate accuracy (mean error) and precision (2 SD) [19, 26, 33, 38, 40]. Although most prior authors reference screw tip and tail accuracy, we agree with Volk et al. that accuracy data at the mid-pedicle is most critical, as this describes proximity of the pedicle screw to the spinal canal [26]. Furthermore, in addition to MAE, we would strongly endorse reporting SME, as this provides valuable information about directionality of systematic errors and allows direct comparison between screws and across studies. Similarly, the variance (or SD) of the SME indicates the precision/consistency of the surgical technique.

Coronal error at screw tail (a), mid-pedicle (b) and tip (c). The origin (center of the target) is the planned screw trajectory and the blue (circle) marks represent actual individual screw positions; right-side screws have been mirrored across the sagittal plane to standardize laterality. Purple stars indicate the signed mean error, while the shaded ellipse represents two standard deviations in screw position. Grey target circles show 2 and 4 mm coronal error relative to planned position. ML = medial(+)/lateral(−), SI = superior( +)/inferior(−)

A limitation of prior accuracy studies is the requirement for human experts to perform the overlay of the preoperative plan with the postoperative screw position. This study reports the development of a fully automated protocol for determination of pedicle screw accuracy utilizing standard preoperative and postoperative CT scans. Our reported registration accuracy (0.48 mm) was limited by the imaging resolution (0.625 mm), while perturbation analysis showed the algorithm to be extremely consistent. In future studies, we intend to reduce the uncertainty of our registration accuracy, and we believe such assessments should be a prerequisite for technical accuracy data reporting. Some limitations of the current study include the small number of pedicle screws assessed, as well as the use of cadavers instead of in vivo screw assessment. The cadavers did not have any spinal deformity, which likely would impact accuracy results. We recognize that only a subset of accuracy errors in pedicle screw placement result in patient harm. However, improvements to accuracy, precision, and reliability of pedicle screw placement using robotically assisted surgical navigation have the potential to reduce the incidence of patient harm by decreasing the 0.26% incidence of UPROR secondary to malpositioned pedicle screws [6, 7, 43]. A standardized, systematic approach to the reporting of pedicle screw accuracy with computer-assisted pedicle screw insertion techniques using standardized nomenclature as well as a screw-centric 3D coordinate system is vital. Standardized cadaver and human models will greatly facilitate testing, comparison, and improvement of robotic systems (ASTM F2554-18).