Fully automated determination of robotic pedicle screw accuracy and precision utilizing computer vision algorithms

Area of Science:

• Spine Surgery
• Medical Imaging
• Computer Vision
• Robotics

Background:

• Traditional pedicle screw accuracy assessment relies on CT scans and visual inspection.
• Accurate pedicle screw placement is crucial to prevent complications and revision surgeries.
• Existing methods lack standardized, detailed 3D accuracy and precision metrics.

Purpose of the Study:

• To evaluate the accuracy and precision of pedicle screw insertion using a novel computer vision algorithm.
• To establish a standardized protocol for assessing pedicle screw placement accuracy and precision.

Main Methods:

• Utilized three cadaveric specimens with bilateral T2-L4 instrumentation via robotic-assisted navigation.
• Employed automated segmentation and computer vision to align preoperative and postoperative CT scans.
• Assessed registration accuracy using embedded tantalum beads for landmark alignment.

Main Results:

• Automated CT registration achieved sub-voxel accuracy.
• Mean 3D errors for screw tip and tail were 1.67 mm and 1.78 mm, respectively.
• Mean angular deviation was 1.58°, with mean absolute errors of 0.75 mm (medial-lateral) and 0.60 mm (superior-inferior) for mid-pedicular accuracy.

Conclusions:

• Introduced automated algorithms for precise pedicle screw placement assessment.
• Achieved accuracy and precision comparable or superior to existing robotic-assisted studies.
• The computerized workflow provides standardized, detailed 3D translational and angular accuracy metrics.