AI-derived versus surgeon-performed instrumentation in adolescent idiopathic scoliosis: a biomechanical simulation analysis.

Abstract

Posterior spinal fusion (PSF) with pedicle screws is the standard treatment for adolescent idiopathic scoliosis (AIS) with curves > 45°, yet significant variability in instrumentation strategies persists. Current planning guidelines lack patient-specific detail and 3D optimization. This study aimed to biomechanically evaluate the simulated 3D spinal correction and implant forces of artificial intelligence (AI)-derived instrumentation strategies compared with surgeon-performed instrumentation in AIS patients undergoing PSF. Thirty-five AIS patients (Lenke 1-2, aged 12-19) treated with PSF were included, with data obtained from the MIMO Clinical Trial and affiliated hospitals. Patient-specific 3D biomechanical models were reconstructed from preoperative radiographs using validated geometric algorithms and multibody dynamics. For each patient, nine instrumentation strategies (AI-generated or surgeon-performed) were simulated with standardized surgical steps. Screw positions were assigned using predefined rule-based patterns derived from neural network-based multi-task learning model (NNML)-predicted parameters. Outcome measures included main thoracic Cobb angle, thoracic kyphosis, apical vertebral rotation, screw pullout forces, number of screws, and fused levels. Statistical comparisons were performed using repeated-measures ANOVA and Friedman tests. All reported outcomes represent simulated biomechanical results rather than clinical follow-up data. Surgeon-performed instrumentations achieved slightly greater simulated Cobb angle correction (≈4°, p < 0.001) but required more screws and fused levels. AI-generated strategies produced higher thoracic kyphosis (≈4°, p = 0.008) and, when optimized, matched or exceeded overall correction in 77% of patients. Best-AI configurations reduced screw count modestly (≈1 screw, p = 0.048) without compromising simulated 3D correction. Biomechanical modeling suggests that AI-derived plans can achieve comparable simulated 3D correction with fewer implants. These findings reflect simulated performance and require clinical validation, but indicate potential value of AI-assisted 3D modeling for future AIS preoperative planning.