Automatic Pelvic Structure Restoration: A Sim-To-Real Approach via Recursive Pose Estimation Network

Pelvic fractures rank among the most complex injuries in orthopedics, where preoperative reduction planning is essential for obtaining positive functional outcomes. The conventional manual planning process depends heavily on the surgeon’s expertise and is time-consuming and error-prone. We propose a pose estimation module tailored for the joint reduction step that predicts the target transformation of dislocated bone. Given the scarcity of annotated data for model training, we employed a sim-to-real approach, simulating traumas on healthy pelvic scans, and applied a recursive strategy for application in real clinical scenarios. The proposed method out-performed mirror and mean template techniques as well as an optimization-based joint matching method, achieving an error of (2.58 ± 0.97) mm in simulated data and (3.07 ± 0.76) mm in clinical data, with an average execution time of 4.35 s. Demonstrating high accuracy and robustness, the method represents a promising stride toward a fully automated pipeline.