Safety and accuracy of cervical pedicle screw navigation using artificial intelligence-generated, MRI-based synthetic CT versus conventional CT.

Pedicle screw placement in the cervical spine is a very demanding technique that may lead to critical complications due to the surrounding neurovascular structures. The aim of this investigator-initiated study was to test whether radiation-free, MRI-based, synthetic CT (sCT)-guided spinal navigation is noninferior to CT-guided spinal navigation in terms of safety and accuracy of cervical pedicle screw placement in a cadaveric model. The cervical spines of 5 cadavers were scanned with both thin-slice CT and the sCT-MRI sequence. From MRI, sCT scans were artificial intelligence-generated with a previously validated model. Preoperatively, screw trajectories were planned on both CT and sCT. Four spine surgeons performed surface matching and navigated Kirschner wire placement from levels C2 to T2 bilaterally. Randomization (1:1 ratio) was performed for modality, surgeon, and side. Postoperative CT scans were acquired and virtual screws with predefined sizes were projected on the wires. Distance and angulation between intra- and postoperative virtual screw positions were analyzed. Medial and lateral breaches were assessed by an independent researcher using the Gertzbein-Robbins classification, with grades A and B considered satisfactory ( 0.05). The mean angulation was 4.6° (SD 2.5°) for CT and 5.4° (SD 2.9°) for sCT (p > 0.05). Of the CT-guided virtual screws, 84% were grade A, 13% grade B, and 3% grade C. Of the sCT-navigated virtual screws, 86% were grade A and 14% grade B. A complete radiation-free cervical pedicle screw navigation using MRI-based synthetic CT scans in a cadaveric experiment is feasible and as safe and accurate as conventional CT-guided navigation.