OBJECTIVE: The clinical application of the Nine-grid Area Division Method for pedicle puncture in L-OVCF is limited by high technical thresholds and low efficiency. This study aimed to develop an AI-integrated automated system for L-OVCF diagnosis and pedicle puncture planning by combining the nine-grid method with deep learning, and to validate its diagnostic accuracy, planning consistency, and clinical application efficiency. METHODS: A multicenter CT dataset of L-OVCF patients was collected from three hospitals affiliated with Capital Medical University (January 2020-December 2022). A two-stage improved U-Net architecture was constructed for automated lumbar vertebral segmentation, and a 3D ResNet50 network was used for L-OVCF identification. A geometric algorithm was developed to realize automated puncture path planning based on the Nine-grid Area Division Method. The performance of the segmentation and diagnosis modules was evaluated with DSC, AUC, and Hausdorff Distance. 20 cases were randomly selected to compare the consistency of puncture planning between the AI system and senior surgeons' manual planning, and the planning efficiency and resource consumption of the two methods were analyzed. RESULTS: The proposed two-stage U-Net achieved an overall DSC of 0.934 for vertebral segmentation, significantly outperforming the single-stage nnU-Net model. The L-OVCF identification model yielded a high degree of accuracy with AUC of 0.918 (95% CI: 0.885-0.925). The automated planning results showed extremely high consistency with manual planning (DSC = 0.958, IoU = 0.921, HD = 486.7 μm), and the planning efficiency was significantly improved with memory consumption within the capacity of standard clinical workstations. CONCLUSIONS: The developed AI-integrated system accurately reproduces the preoperative planning logic of senior surgeons, with high diagnostic accuracy and puncture planning consistency, while markedly improving planning efficiency and reducing the technical threshold of the nine-grid method. It has good clinical translational potential and can provide a reliable auxiliary tool for the precise and minimally invasive treatment of L-OVCF.
Zuo et al. (Wed,) studied this question.