This cross-sectional study developed and evaluated a method for automated MR image segmentation to construct 3D lumbar models at the L4/L5 level, incorporating bone, nerve roots, dural sac, intervertebral disc, and skin. Using these models, the authors quantitatively analyzed virtual working channel trajectories for percutaneous endoscopic lumbar discectomy (PELD). The automated segmentation model achieved high accuracy with mean Dice scores above 0.89 for most structures. By varying the coronal plane angle (CPA) and cross-sectional angle (CSA), the study quantified how the trajectories intersected with anatomical structures such as nerves, dura, and facet joints. Results indicated that an optimal surgical path at L4/L5 occurs at approximately CSA 15° and CPA 15°–20°, balancing access to the disc while minimizing intrusion into critical structures. The overlap trends between different anatomical features were non-linear, demonstrating the complex trade-offs in minimally invasive lumbar access planning.