Toward an Intelligent Robotic Pedicle Screw Placement—Evaluation of a Force-Based Controlled Drilling System

Aims: Malposition of pedicle screws—the central element of lumbothoracic spine fusion procedures—still pose a relevant risk in spine surgery. Existing robotic systems retain precisely the trajectory for screw placement, not more. So far, they comprise the same potential risk of malpositioning associated with navigated image guidance. However, robots can potentially be smart tools if equipped with sensors that provide information far beyond the haptic feedback of the surgeons. The aim of this work was to evaluate a concept of intelligent pedicle drilling with force-based measurement to predict a screw misplacement based on information from high-resolution computed tomography (CT) data.

Methods: We hypothesized that a sensory robotic drill bit could be located with a maximum inaccuracy of 1 mm and less within the cortical borders of the lumbar spine. A robotic drilling arm equipped with a force-torque sensor was utilized. The system was able to detect the resistance while traversing the cortical and spongious bone with a drill. Subsequently, n = 20 trajectories were drilled various segments of a human lumbar spine specimen. Pre- and postexperimental high-resolution CT scans of the drilled trajectories were acquired. Per each drilled trajectory, the axial force applied to the drill bit was extracted and correlated with n = 1000 candidate trajectory in a 3 × 3 mm search space (resolution: 0.3 mm, angle variation: ± 2.5°, Δ0.5°).

Results: Seventeen datasets were available for accuracy analysis. Intentionally, in 4 of 17 trajectories a medial breach was created and 7/17 trajectories a lateral breach. The remaining 6/17 trajectories were fully passed through the pedicle. The pose of the drill trajectory could be identified as accurate as 0.25 ± 0.11 mm (n = 17), with available accuracies during lateral and medial breach to be 0.27 ± 0.06 mm (n = 7) and 0.34 ± 0.11 mm (n = 4), respectively. Available accuracy in the pedicle was 0.15 ± 0.08 mm (n = 6).

Conclusion: The maximum inaccuracy was < 0.5 mm. This technique indicates a potentially very powerful approach that operates independently of an image-guidance model (including inherent problems of image co-registration, optical tracking, and instrument calibration). The highly accurate detection of a cortical bone breach represents a novel and clinically relevant safety feature. By adding sensor technologies, robots can become “smart” and lead the way to the next level of robot-assisted spine surgery.