基于人工智能深度学习的CT-MRI多模态影像自动融合分割技术在前交叉韧带重建术前规划中的应用  

Application of artificial intelligence deep learning-based multimodal CT-MRI images automatic fusion and segmentation in preoperative planning for anterior cruciate ligament reconstruction

作　　者：于浩淼 董继祥[1] 李海鹏 王明新 王龙 王耀霆 葛毅 刘星宇 张逸凌 李春宝 YU Haomiao;DONG Jixiang;LI Haipeng;WANG Mingxin;WANG Long;WANG Yaoting;GE Yi;LIU Xingyu;ZHANG Yiling;LI Chunbao(Department of Sports Medicine,Senior Department of Orthopaedics,the Fourth Medical Center of Chinese PLA General Hospital,Beijing 100089,China;Orthopaedic Center,Beijing Friendship Hospital,Capital Medical University,Beijing 100050,China;School of Biomedical Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]中国人民解放军总医院第四医学中心骨科医学部运动医学科,北京100089 [2]首都医科大学附属北京友谊医院骨科中心,北京100050 [3]清华大学生物医学工程学院,北京100084

出　　处：《中华骨与关节外科杂志》2025年第1期27-35,共9页Chinese Journal of Bone and Joint Surgery

摘　　要：目的:探讨基于人工智能(AI)深度学习算法的CT-MRI多模态影像自动融合分割技术,开发并构建前交叉韧带重建术(ACLR)术前自动规划系统的可行性,以实现ACLR高效准确的术前规划。方法:通过中国人民解放军总医院第四医学中心影像中心得到2023年4月至2024年1月就诊的200例前交叉韧带(ACL)、后交叉韧带(PCL)及半月板正常的膝关节疼痛患者的膝关节CT和MRI影像,由运动医学专业医师对骨皮质、ACL、PCL、半月板等结构进行手工标注,并使用AI深度学习算法对标注图像进行学习,构建CT-MRI多模态影像自动融合分割系统。基于CT-MRI配准融合图像,再次使用AI深度学习技术,强化ACLR股骨、胫骨骨道内外口关键点位的识别,构建ACLR术前自动规划系统。招募12例ACL损伤患者并使用其CT影像,3D打印技术打印其假骨模型并使用ACLR术前自动规划系统对胫骨及股骨骨道位置进行规划,以此为依据在假骨模型上钻取骨道,并对股骨和胫骨骨道长度、关节腔内间距的差异进行统计学分析。结果:CT及MRI多模态影像融合分割后可形成包含骨骼及软组织结构的个体化3D膝关节模型,多模态影像融合精度Dice指数为0.864。ACLR术前自动规划系统进行术前规划的平均时间为(3.0±0.5)min。假骨模拟手术中股骨骨道、胫骨骨道长度及关节腔内间距与术前规划的差异均无统计学意义(P均>0.05)。结论:基于AI深度学习的CT-MRI多模态影像自动融合分割技术的ACLR术前自动规划系统更为智能、快速、精准,可显著提高ACLR术前规划能力,有望降低ACLR术后并发症,提升手术效果。Objective:To explore the feasibility of developing and constructing an automatic preoperative planning system for anterior cruciate ligament reconstruction(ACLR)based on artificial intelligence(AI)deep learning algorithm and multimodal CT-MRI images automatic fusion and segmentation, in order to achieve efficient and accurate preoperative planning for ACLR. Methods:CT and MRI images of the knee joints from 200 patients with normal anterior cruciate ligament (ACL), posterior cruciate ligaments (PCL), and meniscus admitted from April 2023 to January 2024 were obtained from the imaging center of the Fourth Medical Center of Chinese PLA General Hospital. Sports medicine specialists manually annotated structures such as the cortical bone, ACL, PCL, and meniscus, and used AI deep learning algorithm to learn the annotated images to build a multimodal CT-MRI image automatic fusion and segmentation system. Based on the CT-MRI registration and fusion images, AI deep learning technology was further applied to enhance the identification of key points at the entry and exit of the femoral and tibial tunnels in ACLR, constructing an automatic preoperative planning system for ACLR. Twelve patients with ACL injuries were recruited, and their CT images were used to guide the creation of pseudo bone models using 3D printing technology. The ACLR automatic preoperative planning system was used to plan the positions of tibial and femoral tunnel. which were then drilled on the pseudo bone models. Statistical analysis was performed to assess the differences in the lengths of the femoral and tibial tunnel and intra-articular spacing. Results: The CT-MRI multimodal image fusion and segmentation generated an individualized 3D knee joint model containing bone and soft tissue structures, with the Dice index of multimodal image fusion accuracy was 0.864. The mean time for preoperative planning using the ACLR automatic preoperative planning system was (3.0±0.5) min. In the pseudo-bone experiments, the differences between the planned and actua

关 键 词：前交叉韧带重建术 人工智能 多模态影像融合 术前规划 

分 类 号：R687[医药卫生—骨科学]