基于3D打印矫形器三点力学数据与机器学习的特发性脊柱侧弯Cobb角预测及临床评价  

Prediction and Clinical Evaluation of Cobb Angle in Idiopathic Scoliosis Using Machine Learning and Three-Point Mechanical Data of 3D-Printed Orthotics

作　　者：马寻君 李娅 蔚俊 刘海涛 吴云成 王金武[5] MA Xunjun;LI Ya;YU Jun;LIU Haitao;WU Yuncheng;WANG Jinwu(School of Medical Information and Engineering,Xuzhou Medical University,Xuzhou 221004,Jiangsu,China;School of Rehabilitation Medicine,Shandong Second Medical University,Weifang 261053,Shandong,China;Institute of Translational Medicine,Shanghai Jiao Tong University,Shanghai,200240,China;School of Health Sciences and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Department of Orthopedics,Shanghai Ninth People’s Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai 200011,China)

机构地区：[1]徐州医科大学医学信息与工程学院,江苏徐州221004 [2]山东第二医科大学康复医学院,山东潍坊261053 [3]上海交通大学转化医学研究院,上海200240 [4]上海理工大学健康科学与工程学院,上海200093 [5]上海交通大学医学院附属第九人民医院骨科,上海200011

出　　处：《医用生物力学》2025年第2期364-370,共7页Journal of Medical Biomechanics

基　　金：科技部重点研发计划项目(2022YFF1202600);上海市科学技术委员会“科技创新行动计划”国内科技合作领域项目(22015820100);上海交通大学医学院附属第九人民医院临床研究型MDT项目(201914);上海交通大学医学院地高大双百人计划(20152224);贵州省科技支撑计划,黔科合支撑【2023】一般196;中国残疾人联合会资助项目(2021CDPFAT-45)。

摘　　要：目的构建基于3D打印矫形器三点力学数据与多种机器学习算法的青少年特发性脊柱侧弯(adolescent idiopathic scoliosis,AIS)Cobb角预测模型,以提供一种创新、无辐射的AIS早期临床筛查和监测方法。方法采集AIS患者的临床数据及3D打印矫形器的力学数据,构建包含性别、年龄、疾病类型、体重和Risser评分等特征的综合数据集。使用随机森林、支持向量回归、梯度提升回归机、极限梯度提升、轻量级梯度提升机和类别提升6种算法构建并评估Cobb角预测模型性能。结果梯度提升回归机模型在多项评估指标上表现最佳,精确率达到0.937、召回率为0.818、F1分数为0.949、曲线下面积(area under the curve,AUC)为0.843,在验证集中该模型的预测值准确率达到0.942,与实际Cobb值拟合较好。结论基于力学数据和机器学习的Cobb角预测模型有效避免了早期临床筛查中传统全脊柱X线片检查的辐射风险,实现了AIS患者的非侵入性评估,提高了筛查和监测的安全性和效率,为临床医生提供了有力的辅助决策工具,具有重要的临床意义。Objective A Cobb angle prediction model for adolescent idiopathic scoliosis(AIS)based on threepoint mechanical data from three-dimensional(3D)-printed orthotics and various machine learning algorithms was developed,so as to provide an innovative,radiation-free method for early clinical screening and monitoring of AIS.Methods Clinical data from AIS patients and mechanical data from 3D-printed orthotics were collected to construct a comprehensive dataset with features such as gender,age,disease type,weight,and Risser score.Six algorithms,namely,random forest,support vector regression,gradient boosting regressor,extreme gradient boosting,lightgbm,and catboost,were used to construct and evaluate the performance of Cobb angle prediction models.Results The gradient boosting regressor model had the best performance on several evaluation metrics,achieving a precision rate of 0.937,recall rate of 0.818,F1-score of 0.949,and an area under curve(AUC)value of 0.843.In the validation set,the model’s predictions reached an accuracy rate of 0.942,fitting well with the actual Cobb values.Conclusions The Cobb angle prediction model based on mechanical data and machine learning effectively avoids the radiation risks associated with traditional full-spine X-ray examinations in early clinical screening.It provides a non-invasive assessment for AIS patients,enhancing the safety and efficiency of screening and monitoring,and offering a powerful decision-making tool for clinicians,with a great clinical significance.

关 键 词：特发性脊柱侧弯 3D打印矫形器 Cobb角预测 机器学习 非辐射评估 

分 类 号：R318.01[医药卫生—生物医学工程]