AI-guided Spinal Screw Planning for Cervical Spine Procedures

Study Design: Proof-of-concept study. Background: Optimal cervical screw placement is a complex, user-dependent procedure. While advanced techniques, including image-guided surgery and 3D-printed guides, have shown to improve accuracy of this procedure, their success depends on the quality of preoperative planning. However, traditional preoperative planning is labor-intensive, user-dependent, and may overlook patient-specific anatomic nuances. Objective: To develop and evaluate an AI-guided method for planning atlantoaxial and subaxial lateral mass screw trajectories in the cervical spine. Methods: The training of the AI model involved defining the orientation of the atlantoaxial and subaxial screws in all cervical levels on preoperative CT imaging of 41 patients who underwent cervical spine surgery. The model’s performance was evaluated using ten thin-sliced head and neck CT images, effectively simulating clinical applications for various anatomic scenarios. Results: Qualitative evaluation of the AI model demonstrated a 99.3% success rate for defining an identifiable screw trajectory and a 98.6% success rate for identification of the correct vertebral level. Quantitative evaluation of the AI model showed an overall mean screw length of 19.2±3.9 mm, bicortical purchase achievement in 84% of the subaxial screws, an adequate head-to-rod distance in 95.8% of all screws, and a mean runtime of 13.4±4.7 minutes. Conclusions: The developed model represents a valuable step toward the clinical implementation of AI-guided, human-monitored surgical planning for cervical screw placement. By enabling clinicians to primarily verify and fine-tune automated plans, the model reduces workload, saves time, improves accuracy, and minimizes user dependency, which supports safer and more efficient cervical procedures. Further expanding the training dataset to encompass broader spinal pathologies and screw insertion techniques is expected to enhance its clinical value even more.