摘要
目的通过基于多参数MRI影像组学的列线图对膝关节半月板的损伤进行风险度评估。方法选取2020年11月至2021年11月在浙江大学医学院附属杭州市第一人民医院行右膝关节MRI检查的患者86例,按照7:3的比例进行分层抽样,将86例患者随机分为训练组59例,验证组27例。根据MRI诊断标准将所有患者分为撕裂组50例和未撕裂组36例。所有图像在联影智能科研平台上进行影像组学特征提取并筛选,得到影像组学特征参数分数(Rad-score),同时建立影像组学特征模型。通过纳入Rad-score及患者的临床资料,用二元logistic回归建立联合模型,得到联合模型的列线图,实现模型可视化。同时另选取2020年9月至2021年12月在浙江大学医学院附属杭州市肿瘤医院行左膝关节MRI检查的患者30例对模型进行外部验证,优化模型。绘制ROC曲线评价临床特征模型、影像组学特征模型及联合模型对膝关节半月板损伤风险的评估效能。结果联合模型对膝关节半月板损伤风险有较好的评估效能,训练组AUC=0.916,95%CI:0.815~0.973,灵敏度为0.703,特异度为0.955,准确度为0.797;验证组AUC=0.907,95%CI:0.731~0.984,灵敏度为0.846,特异度为0.929,准确度为0.889。通过外院30例膝关节MRI图像进行联合模型的外部验证显示,模型预测结果准确度为0.733(22/30),联合模型预测结果与膝关节半月板实际损伤呈中等的一致性(Kappa=0.467,P<0.01)。结论通过基于膝关节多参数MRI的Rad-score及临床特征资料建立的联合模型所得到的列线图,对膝关节半月板损伤的风险度评估具有一定价值。
Objective To construct a risk assessment Nomogram for knee meniscus injury based on multiparameter MRI.Methods The magnetic resonance images of 86 patients with right knee injury who underwent MRI examination in Affiliated Hangzhou First People's Hospital,Zhejiang University School of Medicine from November 2020 to November 2021 were retrospectively analyzed.There were 50 cases of knee meniscus tear(tear group)and 36 cases without meniscus tear(non-tear group).Patients were randomly assigned in the training cohort(n=59)or the verification cohort(n=27)by stratified sampling at a 7∶3 ratio.The radiomic features were extracted and screened on the United Imaging Intelligence Scientific Research Platform and the Rad-score were calculated.Then,the radiomic feature model was established.The combined model was established by incorporating the Rad-score and clinical features of patients using logistic regression analysis.The Nomogram was constructed to visualize the model.Thirty cases enrolled from Affiliated Hangzhou Cancer Hospital,Zhejiang University school of Medicine were included to verify and optimize the model.Receiver operating characteristic curves(ROC curves)were generated to evaluate the assessment efficiency of clinical feature model,radiomic feature model and the combined model for meniscus injury of the knee.Results The combined model had better efficiency in assessing the risk for meniscus injury of the knee.In the training cohort(AUC=0.916,95%CI:0.815-0.973),the sensitivity,specificity and the accuracy were 0.703,0.955 and 0.797 respectively.In the validation cohort(AUC=0.907,95%CI:0.731-0.984),the sensitivity,specificity and the accuracy was 0.846,0.929,and 0.889 respectively.The external verification using 30 cases from other hospital showed that the accuracy of the prediction of the combined model was 0.733(22/30),which was moderately consistent with the actual injury of the knee meniscus(Kappa=0.467,P<0.01).Conclusion A Nomogram has been constructed based on Rad-score of multi-parameter magnetic resonance imaging and clinical features,which may be used for risk assessment of meniscus injury of the knee clinically.
作者
甄涛
胡大成
周健
沈起钧
ZHEN Tao;HU Dacheng;ZHOU Jian;SHEN Qijun(Department of Radiology,Affiliated Hangzhou First People's Hospital,Zhejiang University School of Medicine,Hangzhou 310006,China;不详)
出处
《浙江医学》
CAS
2022年第14期1506-1512,共7页
Zhejiang Medical Journal
基金
浙江省卫生科技计划项目(2021KY240)
浙江大学临床科学研究基金(YYJJ2019Z06)
杭州市生物医药和健康产业发展扶持科技专项(2021WJCY028)。
