目的 初步探讨基于表观扩散系数(ADC)影像学诊断围绝经期女性骨质疏松的价值.方法 回顾性选取2020年9月-2022年9月我院收治的围绝经期女性(170例)为研究对象,使用随机数表法将170例患者分为训练集(130例)和验证集(40例),根据是否发生骨质疏松将训练集患者分为骨质疏松组(72例)和非骨质疏松组(58例).比较两组患者的临床资料;构建影像组学和临床因素的Logistic回归模型,得出每个患者的影像组学得分(Rad-score)和临床得分(Clinic-score);进一步筛选变量通过广义线性回归模型,并构建联合预测模型,得出联合得分(Combine-score);通过受试者工作特征曲线(ROC)、Hosmer-Lemeshow检验和决策曲线分析(DCA)对模型进行验证.结果 Clinic-score=0.515×ADC+0.211×SIR+0.681 ×BMD+0.373×NMID+0.428×PINP+0.602×C TX+0.586 ×维生素+0.862 ×血清钙+0.945 ×血清磷+0.709×E2.Rad-score=-1.356×original_ngtd m_Busyness+0.926 × wavelet.LHL_firstorder_Median+2.815 × wavelet.LH H_ngtdm_Busyness-0.719 × log.sigma.3.0.mm.3D_gldm_DependenceVariance-1.528.Combine-score=1.686 × Rad-score+1.861 × ADC+1.916 × S IR+0.371×BMD+0.213×NMID+0.539×PINP+0.931× CTX+1.174 × 维生素+0.759 × 血清钙+0.493 ×血清磷+0.899 ×E2.骨质疏松组和非骨质疏松的Rad-score、Clinic-score及Combine-score 比较差异均有统计学意义(P<0.05).临床预测模型分别与影像组学模型、联合预测模型的AUC在训练集和验证集中比较差异均有统计学意义(P<0.05).验证结果显示各模型在训练集和验证集中拟合均较好(P>0.05).结论 与单纯临床预测模型和影像组学模型相比,联合预测模型的鉴别能力最优,在诊断围绝经期女性骨质疏松中具有较高的诊断价值.
Objective To explore the value of imaging diagnosis of osteoporosis in peri-menopausal women based on apparent diffusion coefficient(ADC).Methods Perimenopausal women(170 cases)admitted to our hospital from September 2020 to September 2022 were retrospectively selected as the study objects.170 patients were divided into a training set(130 cases)and a validation set(40 cases)using random number table method.According to whether osteoporosis occurred,the patients in the training set were divided into an osteoporosis group(72 cases)and a non-osteoporosis group(58 cases).The clinical data of the two groups were compared.Logistic regression models of imaging and clinical factors were constructed to obtain the Rad-score and clinical-score of each patient.The variables are further screened through the generalized linear regression model,and the joint prediction model is constructed to obtain the combination-score.The model was verified by receiver operating characteristic curve(ROC),Hosmer-Lemeshow test and decision curve analysis(DCA).Results The clinical score(Clinic-score)was 0.515 × ADC+0.211 × SIR+0.681 × BMD+0.373 × NMID+0.428 × PINP+0.602 × CTX+0.586 × vitamin+0.862 × serum calcium+0.945 × serum phosphorus+0.709 × E2.Imaging score(Rad-score)=-1.356×original_ngtdm_Busyness+0.926×wavelet.LHL_firstorder_Median+2.815×wavelet.LHH_ngtdm_Busyness-0.719×log.sigma.3.0.mm.3D_gldm_DependenceVariance-1.528.The combined score(Combine-score)was 1.686 × Rad-score+1.861 × ADC+1.916 × SIR+0.371 × BMD+0.213 × NMID+0.539 × PINP+0.931 × CTX+1.174 × vitamin+0.759 × serum calcium+0.493 × serum phosphorus+0.899 × E2.There were significant differences in Rad score,Clinic score and Combine score between osteoporosis group and non-osteoporosis group(P<0.05).The AUC of clinical prediction model was significantly different from that of imaging omics model and combined prediction model in training set and validation set(P<0.05).The verification results showed that all models fit well in both the training set and the verification set(P>0.05).Conclusion Compared with the simple clinical prediction model and the imaging omics model,the combined prediction model has the best differential ability,and has higher diagnostic value in the diagnosis of perimenopausal female osteoporosis.