A comprehensive framework for planning pedicle screw trajectory using ANTs template-based registration approach.

PURPOSE: This study proposes a comprehensive registration-based framework for automated planning of pedicle screw trajectories in the lumbar region by using the average population model generated from the Advanced Normalization Tools (ANTs). METHODS: The ANTs templates are generated from training datasets. The entry points and target points of the pedicle screw trajectory on the both left and right sides at each lumbar vertebra on the ANTs templates are annotated by an orthopedic expert, and then are transferred to the patient's coordinates by deformable registration to predict the corresponding entry points and target points. To validate the accuracy of the predicted positions, two kinds of metrics are calculated: (1) the Root Mean Square Error (RMSE), and (2) the angular deviation between the expert-annotated references and the predicted trajectories. RESULTS: The proposed framework achieved encouraging geometric agreement across the entire lumbar vertebrae (from L1 to L5). The minimum RMSEs for entry points: (1) on the left side is 1.92 ± 0.86 (mm), and (2) on the right side is 1.64 ± 0.86 (mm); for target points: (1) on the left side is 5.36 ± 3.00 (mm), and (2) on the right side is 6.16 ± 1.95 (mm). For the angular deviation, the minimum is 5.88 ± 2.05 (degree) on the left side, and 3.89 ± 2.26 (degree) on the right side at the L1 vertebra. CONCLUSION: The proposed framework of using the ANTs template-based registration approach demonstrates the feasibility of automated patient-specific preoperative planning of lumbar pedicle screw trajectories with encouraging geometric agreement.