Automated Image Analysis and Surgical Planning for High-Precision Oncologic Orthopaedic Surgery
Abstract
Purpose
Accurate planning of pelvic fixation is critical for complex orthopaedic and oncologic surgeries, yet current automated tools focus primarily on normal anatomy and do not account for patient-specific variations caused by tumors. We aim to develop an automated pipeline to personalize surgical planning in the presence of pelvic tumors by segmenting the tumor and surrounding pelvic anatomy and subsequently defining patient-specific screw trajectories.
Methods
This study included 100 patients with orthopaedic malignancies who underwent spinopelvic surgery. Tumor, sacrum, left and right pelvis were manually segmented by a radiologist on preoperative CT scans. A 3-D U-Net architecture was trained with 90 patients and evaluated on 10 patients using Dice similarity coefficient (DSC) and 95th percentile of the Hausdorff distance (H95). A statistical shape model was constructed using an atlas of 32 patients with existing left and right pelvic surface meshes and 4 predefined screw trajectories per side. To map screw trajectories onto patient specific anatomy, the shape model was registered, using rigid and deformable registration, onto 10 patients. Trajectory usability was assessed in terms of conformance to bone corridors and a 5-point Likert scale by an orthopaedic surgeon.
Results
The proposed pipeline generated accurate segmentation of tumor and pelvic anatomy, achieving a DSC of 0.64(0.21) and H95 of 26.57mm (16.91mm) for tumor, 0.93 (0.04) and 1.96mm (1.59mm) for sacrum, and 0.89(0.13) and 6.62mm (11.26mm) for left and right pelvis, respectively. The four auto-planned screw trajectories received mean Likert scores of (1) 3.9, (2) 3.7, (3) 3.2, and (4) 3.9, averaged across left and right pelvis.
Conclusion
These preliminary findings demonstrate that the proposed method can accurately delineate normal and tumor-involved pelvic anatomy from preoperative CT imaging. The automated surgical planning framework generated clinically reasonable screw corridors, supporting integration into complex surgical workflows and future patient-specific implant development.