Preoperative evaluation of C2 pedicle screw placement using a deep learning model: Development and validation study

Area of Science:

• Spinal Surgery
• Medical Imaging
• Artificial Intelligence

Background:

• Preoperative assessment for C2 pedicle screw placement is challenging due to low consistency and high skill requirements.
• Existing methods for C2 pedicle screw placement assessment are operationally complex and demand significant surgical expertise.

Purpose of the Study:

• To develop and validate a deep learning model for rapid and accurate assessment of C2 pedicle screw placement feasibility.
• To create an automated pipeline for C2 pedicle delineation and screw placement probability evaluation.

Main Methods:

• Developed C2-Net, a deep learning pipeline with image segmentation and screw placement probability modules.
• Utilized CT images for C2 pedicle delineation and evaluated model performance against 3D-printed screws and surgeons of varying experience.

Main Results:

• C2-Net achieved 89.4% accuracy, 90.0% sensitivity, and 89.0% specificity on the test set.
• The model's performance was comparable to senior surgeons and superior to junior surgeons, demonstrating higher consistency.
• Attention maps provided visual interpretation, and predicted probabilities differentiated C2 pedicle structural variations.

Conclusions:

• C2-Net demonstrates high accuracy and efficiency in assessing C2 pedicle screw placement, outperforming junior surgeons.
• The model offers rapid, consistent evaluations and visual interpretations, showing potential as an assistive tool in spinal surgery.
• C2-Net can aid clinical decision-making for C2 pedicle screw placement.