Comment on: “Development and Validation of a Computer-Assisted Screw Trajectory Planning Model Based on Iterative Closest Point Registration and Weighted K-Nearest Neighbors Algorithms for Cortical Bone and Pedicle Screw Techniques”

Abstract

validated a computer-assisted screw trajectory planning model based on iterative closest point (ICP) registration and weighted k-nearest neighbors (kNN) for pedicle screw (PS), cortical bone trajectory (CBT), and modified cortical bone trajectory (MCBT) techniques. The unified framework and its performance across bone-quality subgroups are useful additions to preoperative spinal planning, but we would like to raise four methodological points. First, the reference standard used to validate the algorithm consisted of trajectories manually planned by two surgeons, and no interobserver or intraobserver reliability metric was reported for these plans. Prior work comparing manual pedicle screw planning between observers found mean sagittal-inclination differences of about 18 degrees between raters, 2 a magnitude similar to the model-versus-expert deviations reported here. Without knowing how much the two expert plans diverged from each other, it is hard to judge whether the reported 2-to 5-mm and 2-to 5-degree deviations reflect genuine algorithmic error or fall within ordinary surgeon disagreement. Second, the template library was built entirely from an internal cohort aged 20 to 40 years with normal bone density, then transferred by weighted kNN to external cohorts that included older and osteopenic patients. Lumbar vertebral morphology changes with age, including increasing osteophyte volume and endplate irregularity not represented in a young, normal-density template set. 3 The acceptance rates reported in older subgroups were reassuring, but this mismatch between template and target population leaves open whether performance would hold up in patients with more advanced degenerative change than those sampled here