Intraoperative cone beam CT in hybrid operating room set-up as an alternative to postoperative CT for pedicle screw breach detection

CT is considered the gold standard for detecting pedicle breach. However, CBCT may be a viable and low radiation dose alternative, to provide intraoperative feedback to surgeons to permit in-room revisions of misplaced screws To assess the ability and reliability of intraoperative cone-beam CT (CBCT) from a robotic C-arm in a hybrid operating room (OR) two hundred forty-one pedicle screws were inserted in cervical, thoracic and lumbar spine of 7 cadavers, followed by CBCT and CT imaging. The CT images served as the standard of reference. Agreement on screw placement between both imaging systems was assessed using Cohen’s Kappa coecient (κ). Sensitivity, Specicity, Receiver operating characteristic (ROC), area under the empirical and tted ROC curves (AUC) were computed to assess CBCT as a diagnostic tool compared to CT. The patient effective radiation dose (ED) was calculated for comparison. A systematic literature review was performed to provide perspective to the obtained results. sensitivity and respectively. empirical and 0.96, respectively.


Introduction
Accurately positioned pedicle screws offer optimal biomechanical xation, making them a well-accepted treatment choice for a variety of spine pathologies 1,2 . However, up to 42% of pedicle screws were reported to be malpositioned 3,4 . While initially silent clinically, these may eventually present with an unstable construct, reduced fusion, pseudoarthrosis and accelerated adjacent-level degeneration 5 .
In comparison to CT, radiography detects only 52% of misplaced screws 6,7 , making intraoperative CT imaging necessary, especially for complex deformity, to avoid revision surgeries. Availability of intraoperative 3D imaging resulted in revision of 9% of screws intraoperatively, corresponding to 35% of Page 3/13 the treated patients8. Lower threshold for intraoperative revisions based on objective intraoperatively available imaging data, leads to fewer secondary revision surgeries 8 . Avoidance of postoperative revisions makes the initial investment in intraoperative 3D imaging technologies economically more attractive 9 .
Hybrid operating rooms (OR) equipped with a motorized C-arm coupled with a radiolucent surgical table as well as with integrated navigation capabilities, have been recently used for spine surgery 10 . The C-arm provides intraoperative cone-beam CT (CBCT) imaging.
The purpose of this cadaver study was to determine the diagnostic performance of intraoperative CBCT using a C-arm in a hybrid OR for identifying screw misplacements in comparison to gold-standard postoperative CT. To put our results in perspective, a systematic literature review of studies comparing intraoperative 3D imaging for screw misplacements with postoperative CT was conducted.

Materials And Methods
All methods were carried out in accordance with relevant guidelines and regulations. Ethics approval for a cadaver study was obtained from the ethic committee of the Christian-Albrecht-University Kiel, Germany.
All Cadavers were anonymous and located at the department of forensic pathology, University of Hamburg.

Surgical technique
Surgeries were performed by one surgeon in a hybrid OR (AlluraClarity Flexmove, Philips, the Netherlands) ( Fig. 1). Cadavers were placed prone on the operating table, adhesive skin markers applied and a planning CBCT scan obtained. Pedicle screw trajectories were planned using Augmented Reality Surgical Navigation system (ARSN, Philips, the Netherlands) 11 . Pedicles were instrumented bilaterally from C7 to L5 using a tracked Jamshidi needles, aligned along the planned trajectory. Cannulated pedicle screw sizes were 5.5 x 40 mm from C7 to T9; and 6.5 x 45 mm from T10 to L5.
CBCT imaging: the study group The C-arm was positioned on the side, orthogonal to the operating table along the axial plane of the patient and rolled in this plane to perform CBCT optimized for spine imaging (XperCT Spine, Philips, the Netherlands). The CBCT reconstruction is the result of 302 x-ray projections acquired over a 180-degree clockwise rotation, starting with the x-ray tube on the right side of the table. The acquisition settings were 120 nominal kV tube voltage and modulated tube current-exposure time (mAs) generated by the automatic exposure control which is estimated based on patient size thickness at each angular projection during the rotation. Radiation dose-area products were collected from the radiation dose structured report (RDSR) and converted to patient effective dose (ED).
The CBCT reconstruction corresponds to a 512x512x396 matrix with isotropic voxels of 0.49 mm.
Consequently, the dimension of the imaged region of interest (ROI) corresponds to a 25x25 cm in the axial plane by 19.5 cm in the cranial-caudal direction 12 (Fig. 2).

CT: the reference group
Default settings for spine imaging on a 16-slice CT scanner was used to image the cadavers (Brillance 16, Philips, the Netherlands). CT image reconstructions were made based on a helical acquisition at 120 kV tube voltage acquisition and a modulated tube current-exposure time (mAs), derived from the anterioposterior and lateral scout scans. Radiation dose-length products were collected from the RDSRs and converted to patient ED.
The CT image reconstructions were performed at 0.75 mm slice thickness and dimension of the reconstructed ROI image was 37x37 cm in the axial plane, in crania-caudal length direction covering the whole instrumented spine (Fig. 2).

Data analysis
The CBCT and CT images were reviewed on a PACS system with 3D volumes displayed in a multi-planar format. The window width and level were set to optimize visualization. Screw position assessments were performed using the Gertzbein grading 13 : -grade 0 (screw within the pedicle without cortical breach) -grade 1 (0-2 mm breach, minor perforation including cortical encroachment) -grade 2 (2-4 mm breach, moderate breach) -grade 3 (more than 4 mm breach, severe misplacement).
Each screw was assigned a grade from 0 to 3, for each imaging modality. Consequently, a 4x4 contingency table to compare the similarity in rating was created. Figure 3 illustrates an example of each of the 4 grades on both CBCT and CT from the diagonal of the contingency table (i.e. agreement between both imaging modality for each grade). All images in Fig. 3 were displayed for comparison at equal display settings optimized for bone and hardware imaging.
Grades 0 and 1 were considered clinically accurate, whereas grades 2 and 3 were considered misplaced.

Statistical analysis
Cohen's Kappa coe cient (κ) was calculated to assess the agreement between CBCT and CT based on the contingency tables and was classi ed according to Landis et al. 14 as detailed in Table 1. Sensitivity was de ned as the proportion of misplaced screws detected on CBCT which were con rmed as such on CT; whereas speci city was de ned as the proportion of screws accurately placed according to CBCT which was con rmed as such on CT. Receiver operating characteristic (ROC) analysis was performed and area under the empirical and tted ROC curves (AUC) were computed to assess performance of CBCT as a diagnostic tool whose performance was classi ed according to Hanley et al. 15 (Table 1). Two-sided Fisher's exact and Wilcoxon signed rank tests were used for comparison where appropriate. Two-sided AUC comparison test was performed. Con dence intervals (CI) were computed at 95% and expressed between brackets. A p-value (p) less than 0.05 was considered as signi cant.

Literature review
A systematic review was conducted in order to perform a proper comparison with all the existing literature on assessing the diagnostic performance of intraoperative 3D imaging versus CT.
The review was performed using PubMed Central, PubMed, Embase, Web of Science, and Cochrane with the following search strategy: ("pedicle screw placement" AND "sensitivity" AND "speci city" AND "imaging"). After ltering for duplicates from the various search engines, 5 exclusion criteria were applied during abstract screening: (a) articles not written in English, (b) conference abstracts, (c) articles that do not report about pedicle screw (e.g. sacroiliac screws), (d) articles that are not comparative studies, and (e) studies that do not consider an intraoperative 3D system as study group (e.g. 2D radiography, electromyography, etc.). After text screening, papers that did not report clinical accuracy for screw placement by using a Gertzbein-like grading i.e. (2-mm increment scale relative to breach) were excluded as also data with inconsistency or missing information. All preclinical and clinical studies were included.
Literature data with available contingency tables, or with detailed numbers of screws per grade for each group from which contingency tables could be generated, were used to calculate sensitivity, speci city, and AUC and statistical tests performed to ensure a fair diagnostic performance comparison between all studies.

Accuracy comparison between CBCT and CT
All cadavers had their full instrumented spine imaged with 3 CBCT as seen in comparison with CT (Fig. 2). Table 2 corresponds to the contingency table comparing the grading assessment between both imaging modalities for each screw. Out of the 241 placed screws, 134, 22, 6, and 17 screws were rated by both imaging modalities as grade 0, 1, 2, and 3, respectively; yielding a total of 179 screws with equal ratings. With respect to screws considered as accurate placement on CT, twenty-four screws and one screw were rated as grade 1 and 2 respectively on CBCT while they were rated grade 0 on CT, whereas thirteen screws and three screws were rated as grade 0 and 2 respectively on CBCT while they were rated as grade 1 on CT. With respect to screws considered as misplaced on CT, two screws were rated grade 0 and three screws were graded 1 on CBCT while they were rated as grade 2 on CT. Finally, two and twelve screws were rated as grade 1 and 2 on CBCT while they were rated as grade 3 on CT.

Radiation dose of the subjects
The patient ED in CBCT imaging was signi cantly lower than CT for all 7 cadavers (median 8.2 mSv vs 12.9 mSv, p < 0.05). The average ED was 7.2 ± 5.4 mSv in CBCT vs 14.3 ± 3.4 mSv in CT with a 53% ED reduction on CBCT.

Literature review and comparison
The initial database search identi ed 117 studies without any duplicates. Of these, 104 articles were excluded after scrutiny of the abstracts. The two most common reasons for exclusion were articles not being comparative studies (n = 35), and articles where the study group did not correspond to an intraoperative 3D imaging such as 2D radiography, EMG, or other techniques (n = 35). Fifteen studies were not related to pedicle screws (e.g. iliac screws), and an equal amount corresponded to conference abstracts. Four abstracts corresponded to papers not written in English. From the remaining 13 articles, 5 other exclusions were applied at text screening level. The most common reason of exclusion was not using a Gertzbein-like grading with 2 mm increment for breach (n = 5) [16][17][18][19][20] . One study was excluded because of inconsistency between the data in the abstract and in the text 21 . One study did not use a grading scale but revision surgery as surrogate for screw misplacement22, another study did not con rm whether the obtained accuracy was indeed breach < 2mm or all breaches (i.e. grades 1, 2 and 3) 23 , and one study did not provide enough information to compute accuracy for breach < 2mm and considered all breaches as misplaced screws 24 .
There were nally 4 studies that were discussed in this review [25][26][27][28] (Fig. 4). Two studies were cadaveric work comparing intraoperative CBCT vs macroscopic dissection 25,26 from which one compared to CT as well 26 ; and 2 clinical studies compared CBCT vs CT 27,28 . Details of these studies are in Table 3 with sensitivity, speci city, and AUC where applicable. Sensitivity and speci city from literature ranged between 0.69-0.86 and between 0.71-1.00, respectively. The AUC could be only derived from one study 28 with an available contingency table yielding AUC from the empirical and tted ROC curves of 0.91 [0.83-0.98] and 0.93 [0.87-1.00], respectively. Figure 5 compares the ROC curves and corresponding AUC.

Discussion
Hybrid ORs support multidisciplinary use of 2D and 3D imaging and navigation for open and minimal invasive procedures 10 . This cadaver study sought to assess the diagnostic performance of CBCT from a C-arm within a hybrid OR compared to diagnostic CT. Unlike mobile C-arm with CBCT capability and mobile CT, the C-arm system in the hybrid OR has an image acquisition and processing chain comparable to diagnostic CT scanners, enabling better management of image quality and radiation dose exposure 29 .
While macroscopic dissection is the gold reference for evaluating pedicle screw placement, CT is the clinically viable alternative 26 . Choosing a cadaver study over a clinical study allowed for immediate comparison between CBCT and CT. Furthermore, cadaver studies allow uniformity of CT and CBCT protocols on the same scanner to ensure fair comparisons. In clinical studies, follow-up scans are often performed several months after surgery. Bone remodeling and regrowth or screw loosening occurs during the follow-up period which can in uence the Gertzbein grading introducing thus possibilities for comparison errors.
The almost perfect inter-modality agreement of this study was suggestive of CBCT as a viable replacement for postoperative CT. The sensitivity obtained for the CBCT images was 0.84, indicating that 84% of misplaced screws detected on CT, were detected on CBCT. Of the 7 screws that were missed as a misplaced screw in CBCT, 6 were in the thoracic region. A speci city of 0.98, indicates in 2% of the cases, CBCT identi ed screws as misplaced when they are assessed to be well placed on CT. The screws that had the degree of cortical wall penetration overestimated in the CBCT, were mainly in the thoracic region. Interestingly, 5/7 screws were missed as misplaced on CBCT and 3/4 instances where the screw was labelled as misplaced on CBCT but a well-placed screw on CT, had a difference of 1 grade. In Gertzbein, there is an increment of 2mm in the amount of displacement between the consecutive grades. It is likely that the metal artefacts and the small pedicle sizes in the thoracic region, could potentially be the cause for the 2mm error in the assessment.
A higher sensitivity compared to speci city suggests that CBCT based assessment tends to underestimate pedicle breaches compared to CT. Clinically, it is important to detect a truly misplaced screw. Therefore, sensitivity is the most important parameter. However, sensitivity and speci city should be both compared together to get the overall picture. AUC analysis from ROC curves is the radiologically accepted technique to assess a system's diagnostic performance described by a single number, as it considers sensitivity-speci city for all grades. The CBCT used in this study yielded AUCs suggestive of its excellence as a diagnostic tool for assessing screw placement during surgery.
The conducted systematic literature review yielded 4 studies meeting the search criteria. The sensitivity and speci city of this study was in the upper range from the reported values in the literature ( Table 3).
The Cohen's Kappa and the AUC are indicators of the strength of agreement and diagnosis comparisons respectively. The strength of agreement was superior to the study performed with a mobile CBCT by Fujimori et al. 27 and comparable to the study performed with CBCT in a hybrid OR28. The study by Cordemans et al. is the only study from which the AUC could be derived for comparison of diagnostic strength28. These values were lower than in this current study but not statistically signi cant (p = 0.20 and p = 0.27 for the empirical and tted AUC, respectively). The tted AUC accounts for the continuous nature of measuring breach while the empirical AUC only takes the 2mm increment as de ned by Gertzbein.
Inclusion of a high number of misplaced screws (18% with a grade 2 or higher breach as per Gertzbein) and screws in the cervical or upper thoracic regions where the pedicle sizes are small (36% screws were in these regions) provide a better evaluation of the technology. This is re ected by the higher con dence interval compared to literature, thereby providing a higher reliability of sensitivity.
The CBCT offers signi cant patient ED reduction (53% reduction) compared to CT. In institutions where a threshold for patient radiation dose exposure exists for perioperative procedures, one can justify the use of intraoperative CBCT as a diagnostic tool at a lower radiation dose exposure compared to CT, without altering diagnostic judgement of screw placement.
Cervical screws were limited in this study and detection of pedicle breach in the cervical region may be more challenging, given the smaller pedicle diameter and the shoulder superimposition in the image. Additional studies addressing other factors impacting image quality such as contrast-to-noise ratio, severity of metal artefacts, image reconstruction algorithm, and radiation dose protocol should be performed to complete the diagnostic evaluation of CBCT versus CT.

Conclusion
Intraoperative CBCT by C-arm in a hybrid OR is highly reliable in identi cation of screw placement at signi cant dose reduction. A systematic literature search demonstrated superior performance compared to the existing CBCT of mobile C-arm systems con rming reported performance by another study using a hybrid OR system.

Declarations
Ethics Approval: Ethics approval was obtained for a cadaver study from the ethic committee of the Christian-Albrecht-University Kiel, Germany Consent for Publication: N/A Availability of Data and Materials: The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. Grade 3 0 0 2 17 Table 3 Literature data comparison. C, T, and L correspond to cervical, thoracic, and lumbar spinal levels, respectively. Κ corresponds to Cohen's Kappa coefficient. AUCe and AUCf correspond to the area under the empirical and fitted receiver operating characteristic curve, respectively. NA indicates not available data.