基于深度学习神经网络技术的脊柱椎弓根螺钉自动规划研究  

Automatic layout planning of spinal pedicle screws based on deep learning neural network technology

作　　者：赵经纬[1] 张蕴显 施崭 张琦 杨智[2] 刘波[1] 何达[1] ZHAO Jingwei;ZHANG Yunxian;SHI Zhan;ZHANG Qi;YANG Zhi;LIU Bo;HE Da(Department of Spinal Surgery,Beijing Jishuitan Hospital,Intelligent Orthopedic Innovation Unit of China Academy of Medical Sciences,Beijing 100035,China;School of Biomedical Engineering,Capital Medical University)

机构地区：[1]北京积水潭医院脊柱外科,中国医学科学院智能骨科创新单元,北京100035 [2]首都医科大学生物医学工程学院

出　　处：《中国数字医学》2024年第4期84-91,共8页China Digital Medicine

基　　金：首都卫生发展科研专项(2024-2-2076);北京市自然科学基金-海淀原始创新联合基金资助项目(L202005);北京市医院管理中心青年职工创新工作室项目(2023);北京积水潭医院学科新星(XKXX202210)。

摘　　要：目的:针对骨科手术机器人螺钉手工规划效率低下的问题,实现基于CT的脊柱椎弓根螺钉自动、高效、高质量规划。方法:采用深度学习神经网络对标注分割和螺钉的CT图像进行有监督的机器学习,实现脊柱椎弓根螺钉的自动规划;本实验使用44例腰椎CT共440枚螺钉作为训练集,使用11例CT生成110枚螺钉作为测试集,以手工规划作为对照组,通过盲法专家评价评估螺钉规划效果,并通过记录规划时间评估规划效率。结果:该自动规划方法生成的螺钉规划临床可用率为95.4%,自动规划时间与平均手工规划时间分别为68.8 s和177.6 s。结论:该自动规划方法可初步实现高效、高质量的脊柱椎弓根螺钉自动规划,但仍需临床医生监督复核。Objective To address the low efficiency of manual screw layout planning during robot-assisted orthopedic surgery,and to achieve automatic,efficient,and high-quality layout planning of vertebral pedicle screws based on CT images.Methods A supervised machine learning method was used with a deep learning neural network to segment and layout screws for vertebral pedicles using CT scans.440 screws from 44 cases of lumbar spine CT scans were used as training sets,and 110 screws generated from 11 cases were used as testing sets.The results of manual layout planning were control group,and blinding was used in the experts'evaluation for the layout planning,while the layout duration was recorded for the planning efficiency evaluation.Results The clinical availability rate of the automatic layout planning method was 95.4%,and the average automatic layout planning duration and manual planning duration were 68.8 seconds and 177.6 seconds,respectively.Conclusion This method can initially achieve efficient and high-quality automatic layout planning of vertebral pedicle screws.However,clinician supervision and review are still necessary.

关 键 词：智能骨科 深度学习神经网络 AI辅助诊疗 手术自动规划 

分 类 号：R197.3[医药卫生—卫生事业管理] R319[医药卫生—公共卫生与预防医学]