Balancing Iliac Crest Avoidance and Neural Safety in L5 / S1 TELD : Quantitative Identification of a Candidate Balanced Trajectory Window Using AI ‐Assisted 3D MRI Reconstruction

OBJECTIVE: Transforaminal endoscopic lumbar discectomy (TELD) at L5/S1 is challenging due to the iliac crest barrier. Conventional planning relies on CT or 2D imaging, inherently lacking soft-tissue contrast to visualize neural structures. This study aimed to define a quantitative candidate balanced trajectory zone using AI-assisted 3D MRI reconstruction to simultaneously evaluate iliac crest obstruction and neural safety constraints. METHODS: A deep learning model was trained on 50 thin-slice lumbar MRIs to achieve precise segmentation and 3D reconstruction of neural, dural, and osseous structures. Thirty lumbar models were selected after image- and reconstruction-quality screening for surgical puncture simulation. Using a 5° increment as the angular increment, 49 virtual working trajectories (coronal plane angles, CPA 0°-30° and cross-sectional angles, CSA 0°-30°) were established for each bilateral intervertebral foramen, yielding a total of 2940 trajectories. Intersection volumes between each trajectory and the iliac crest, L5 nerve root, dural sac, and articular processes were quantitatively analyzed. RESULTS: The model achieved an 86.4% mean Dice coefficient (73.2%-92.4%). Among 2940 trajectories, 1207 (41.1%) completely avoided the iliac crest. Spatial analysis revealed a trade-off: increasing CSA reduced iliac and S1 superior articular process obstruction but increased L5 osseous removal, whereas steeper CPAs spared the L5 nerve root but risked the L5 inferior articular process. Quantitative analysis identified a candidate balanced trajectory window (CPA 20°-25°, CSA 25°-30°) that achieved a balance between iliac crest avoidance, neuro-osseous preservation, and facet integrity. CONCLUSIONS: This candidate trajectory window (CPA 20°-25°/CSA 25°-30°) may provide a useful reference for balancing iliac crest avoidance with neuro-osseous safety. These findings provide quantitative constraints for manual planning and for the future development of computer-assisted trajectory-planning systems.