Automated intraoperative segmentation and path planning framework for femoral neck fractures under fluoroscopic guidance

In closed reduction and internal fixation (CRIF) for femoral neck fractures, intraoperative fluoroscopy often suffers from severe noise and blurriness due to factors such as low-dose radiation protocols and patient positioning constraints. This significantly obscures critical anatomical boundaries, making Kirschner-wire (K-wire) insertion highly dependent on subjective surgeon experience and consequently increasing operative risks, including malpositioning and revision surgeries. To address these challenges, we propose an automated framework that maps image segmentation to surgical planning, alongside a novel dataset termed FFSP, which provides pixel-level anatomical and clinical geometric annotations. We introduce U-Comer, a dual-branch deep learning architecture specifically designed to fuse global structural context with local boundary details through cross-path interactions, effectively mitigating the limitations imposed by degraded fluoroscopic image quality. Leveraging U-Comer’s precise anatomical segmentations, we formulate an interpretable, geometry-based algorithm for automated K-wire trajectory design, evaluated by newly established quantitative metrics. Extensive experimental evaluations confirm that U-Comer achieves state-of-the-art segmentation performance on the FFSP dataset. Furthermore, the automatically generated puncture paths align closely with expert clinical guidelines, underscoring the framework’s potential as a robust and reliable intraoperative decision-support tool.