基于3D-Densenet神经网络技术助力低剂量CT测量骨密度:与定量CT一致性研究  

Assisting low dose CT measurement of bone mineral density with 3D-Densenet neural network technology:a study on consistency with quantitative CT

作　　者：马铎珊 苏丹阳 王岩[1] 高剑波[1] 吴艳[2] MA Duoshan;SU Danyang;WANG Yan;GAO Jianbo;WU Yan(Department of Radiology,the First Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China;Medical 3D Printing Center,the First Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China)

机构地区：[1]郑州大学第一附属医院放射科,河南郑州450052 [2]郑州大学第一附属医院医学3D打印中心,河南郑州450052

出　　处：《实用放射学杂志》2024年第9期1518-1522,共5页Journal of Practical Radiology

基　　金：河南省引进高端外国专家计划项目(HNGD2022032);河南省科技攻关项目(222102310238)。

摘　　要：目的 评估基于3D-Densenet神经网络技术的人工智能(AI)骨密度(BMD)测量系统与定量CT(QCT)测定BMD的相关性和一致性,并评估其诊断骨质疏松症(OP)的有效性.方法 回顾性选取1 201例接受低剂量CT(LDCT)检查的健康体检者,分别使用AI BMD测量系统和QCT测量T12、L1、L2椎体BMD及平均BMD.采用Bland-Altman、Pearson及Kappa分析2种方法测量BMD的一致性与相关性.以QCT结果为参考标准,绘制受试者工作特征(ROC)曲线评估AI BMD测量系统诊断OP的准确性.结果 2种方法测量的平均BMD的相关系数(γ)=0.997,决定系数(γ^(2))=0.993.AI BMD测量系统对BMD正常、低骨量、OP诊断的Kappa值为0.905.AI BMD测量系统用于诊断OP的曲线下面积(AUC)为0.998,敏感度为0.888,特异度为0.997.结论 基于3D-Densenet神经网络技术的AI BMD测量系统与QCT测量结果具有高度相关性及一致性,可准确地诊断BMD正常、低骨量与OP.Objective To evaluate the correlation and consistency between an artificial intelligence(Al)bone mineral density(BMD)measurement system based on 3D-Densenet neural network technology and quantitative computed tomography(QCT)in measuring BMD,as well as to assess its effectiveness in diagnosing osteoporosis(OP).Methods A total of 1201 participants who underwent low dose computed tomography(LDCT)were retrospectively included.The AI BMD measurement system and QCT were utilized to measure the BMD of T12,L1,L2 vertebrae,and the average BMD.Consistency and correlation of BMD measurements between the two methods were assessed using Bland-Altman,Pearson,and Kappa analyses.With QCT results as the reference standard,the receiver operating characteristic(ROC)curve was drawn to evaluate the accuracy of AI BMD measurement system in diagnosing OP.Results The γ and γ^(2) for the average BMD measured by the two methods were O.997 and 0.993,respectively.The Kappa value for the diagnosis of normal BMD,low bone mass,and OP using the AI BMD measurement system was 0.905.The area under the curve(AUC)for diagnosing OP using the AI BMD measurement system was 0.998,with a sensitivity of 0.888 and specificity of 0.997.Conclusion The AI BMD measurement system based on 3D-Densenet neural network technology has a high correlation and consistency with the QCT measurement result,which can accurately diagnose normal BMD,low bone mass,and OP.

关 键 词：骨质疏松症 骨密度 定量CT 人工智能 

分 类 号：R681[医药卫生—骨科学] R814.42[医药卫生—外科学] TP18[医药卫生—临床医学]