腹盆部CT股骨近端影像组学模型行骨密度筛查的可行性  

The feasibility of bone mineral density screening using a proximal femur radiomics model derived from abdomen-pelvic CT scans

作　　者：杜长谕 刘义军[1] 王诗耕 童小雨 范勇 魏巍 陈安良[1] 何健 DU Changyu;LIU Yijun;WANG Shigeng;TONG Xiaoyu;FAN Yong;WEI Wei;CHEN Anliang;HE Jian(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Dalian,Liaoning Province 116011,China)

机构地区：[1]大连医科大学附属第一医院放射科,辽宁大连116011

出　　处：《实用放射学杂志》2025年第2期310-314,共5页Journal of Practical Radiology

摘　　要：目的基于腹盆部CT股骨近端图像构建自动骨密度(BMD)评估模型,分析其在机会性骨质疏松症(OP)筛查中的应用价值。方法回顾性选取行腹盆部平扫CT检查的患者351例,按7︰3的比例随机分为训练集(n=245)和测试集(n=106)。将所有图像传至定量CT(QCT)后处理工作站,测量左侧股骨近端BMD,根据QCT BMD T值将患者分为骨质疏松(T值≤-2.5)、骨量减少(-2.5<T值<-1)和骨质正常(T值≥-1)。使用自动分割模型分割左侧股骨近端,并分别采用随机森林(RF)和逻辑回归(LR)分类器构建2个三分类的BMD评估影像组学模型。构建受试者工作特征(ROC)曲线,并计算曲线下面积(AUC)、敏感度、特异度等指标评估2个模型的诊断性能,采用DeLong检验比较模型间差异。结果测试集中RF模型和LR模型鉴别骨质疏松的AUC分别为0.953和0.954,鉴别骨量减少的AUC分别为0.894和0.870,鉴别骨质正常的AUC分别为0.975和0.982;模型性能比较结果显示,训练集和测试集中RF模型和LR模型鉴别3种骨质状态能力均无统计学差异(P>0.05)。结论基于腹盆部平扫CT构建的RF和LR影像组学模型均可用于机会性BMD筛查,具有较高的诊断效能。Objective To develop an automated bone mineral density(BMD)assessment model based on proximal femur images from abdomen-pelvic CT scans and to analyze its application value in opportunistic osteoporosis(OP)screening.Methods A retrospective selection was conducted on 351 patients who underwent abdomen-pelvic plain CT examination.The patients were randomly divided into training set(n=245)and test set(n=106)in a ratio of 7:3.All images were transferred to a quantitative computed tomography(QCT)post-processing workstation to measure the BMD of the left proximal femur.According to the QCT BMD T-score,the patients were divided into osteoporosis(T-score≤-2.5),osteopenia(-2.5<T-score<-1)and normal bone density(T-score≥-1).The left proximal femur was dissected using an automatic segmentation model,and two three-class BMD assessment radiomics models were constructed using random forest(RF)and logistic regression(LR)classifiers,respectively.The receiver operating characteristic(ROC)curves were generated,and the area under the curve(AUC),sensitivity,specificity and other metrics were calculated to evaluate the diagnostic performance of the two models.The DeLong test was used to compare differences between the models.Results In the test set,the AUC of the RF and LR models for identifying osteoporosis were 0.953 and 0.954,respectively.The AUC for identifying osteopenia were 0.894 and 0.870,and the AUC for identifying normal bone density were 0.975 and 0.982,respectively.The comparison of model performance showed no statistically significant differences between the RF and LR models in identifying the three bone states in both the training and test sets(P>0.05).Conclusion Both the RF and LR radiomics models,constructed based on abdomen-pelvic plain CT scans,can be used for opportunistic BMD screening with high diagnostic efficiency.

关 键 词：骨质疏松症 影像组学 机会性筛查 髋部 计算机体层成像 

分 类 号：R681[医药卫生—骨科学] R445[医药卫生—外科学] R814.42[医药卫生—临床医学]