术前影像学指标预测乳腺癌模型的构建与评价  

Establishment and evaluation of a model using preoperative imagine indexes for breast cancer

作　　者：张静 甘旭 陈微微[2] ZHANG Jing;GAN Xu;CHEN Wei-wei(Department of Ultrasonic Diagnosis,Yuyao Hospital of Traditional Chinese Medicine,Yuyao 315400,P.R.China;Department of Ultrasonic Diagnosis,Yuyao People's Hospital,Yuyao 315400,P.R.China)

机构地区：[1]余姚市中医医院超声诊断科,浙江余姚315400 [2]余姚市人民医院超声诊断科,浙江余姚315400

出　　处：《中华肿瘤防治杂志》2020年第20期1664-1668,共5页Chinese Journal of Cancer Prevention and Treatment

摘　　要：目的早期诊断是影响乳腺癌预后的重要因素。影像学检查能够为乳腺癌诊断提供有价值的数据。本研究根据术前影像学变量建立乳腺癌预测模型,探讨不同的影像学指标对良恶性乳腺疾病的诊断价值。方法选择2013-01-01-2015-06-30因乳腺疾病在余姚市中医医院接受手术治疗的180例患者(其中病理确诊乳腺癌123例)以及在余姚市人民医院接受手术治疗的125例患者(其中病理确诊乳腺癌92例)。患者术前均接受完善的影像学检查。根据患者术前影像学特征通过Logistic回归建立预测乳腺癌发生风险的模型,然后通过受试者工作特征(receiver operating characteristic,ROC)曲线评价其预测能力,并获得最佳截断值。结果通过比较所有变量在乳腺良恶性患者间的分布,发现8个影像学指标差异有统计学意义,P<0.05;分别是乳腺X射线摄影(mammography,MMG)毛刺症、MMG恶性钙化、超声肿块边缘不清晰、超声肿块纵横比>1、核磁共振成像(magnetic resonance imaging,MRI)时间-信号曲线Ⅲ型、MRI不均匀强化、肿块直径>5cm和乳腺影像报告和数据系统(breast imaging reporting and data system,BI-RADS)≥4分。将上述8个变量纳入Logistic多因素回归分析,结果发现MMG恶性钙化(OR=3.821,95%CI为1.73~8.378,P=0.001)、超声肿块纵横比>1(OR=2.184,95%CI为1.152~4.141,P=0.017)、MRI不均匀强化(OR=3.040,95%CI为1.594~5.798,P=0.001)、BI-RADS≥4分(OR=13.547,95%CI为6.868~26.723,P<0.001)是与乳腺癌有关联的危险性因素;而肿块直径>5cm能够减少危险性,OR=0.428,95%CI为0.269~0.680,P=0.005。此外本模型的C-index值为0.875(95%CI为0.831~0.919),提示对乳腺癌的预测能力较好。通过ROC曲线得出:当模型的预测概率值为0.872时,对乳腺癌的预测为最佳,此时模型对乳腺良恶性鉴别诊断的敏感性、特异性分别为81.1%、80.5%。结论依据影像学检查结果建立的预测模型可以准确预测个体患者存在乳腺癌的�OBJECTIVE Increasingly evidences have shown that the early diagnosis plays an important role in the treatment of breast cancer.Among various diagnostic methods,imaging technology can provide valuable information for breast cancer patients.We built a model to predict the risk of breast cancer based on the preoperative imagine indexes and explored the diagnostic value of different imaging indexes for benign and malignant breast diseases.METHODS From January 1 st,2013 to June 30 th,2015,a total of 180 patients(123 cases of breast cancer confirmed by pathology)in our hospital and 125 patients(92 cases of breast cancer confirmed by pathology)in Yuyao People’s Hospital were selected for surgical treatment due to breast disease.All the patients received complete imaging examination(X-ray,color Doppler ultrasound and magnetic resonance imaging)before operation.According to the preoperative imaging characteristics,a model predicting breast cancer risk was built by logistic regression,then the predictive ability was evaluated by the receiver operating characteristic(ROC)curve and the optimal truncation value is obtained and best cutoff value of the model was also obtained.RESULTS By comparing the distribution of all variables among benign and malignant breast patients,there were 8 imaging variables with statistical significance(P<0.05).They were burr sign(mammography,MMG),malignant calcification(MMG),unclear margin of mass(ultrasound,US),aspect ratio >1(US),typeⅢ of time-signal intensity curve(magnetic resonance imaging,MRI),uneven enhancement(MRI),tumor diameter>5 cm,and breast imaging reporting and data system(BI-RADS)≥4.After including the above 8 variables into multivariable logistic regression analysis,we found malignant calcification(MMG;OR=3.821,95%CI:1.743-8.378,P=0.001),aspect ratio > 1(US;OR=2.184,95%CI:1.152-4.141,P=0.017),uneven enhancement(MRI;OR=3.040,95%CI:1.594-5.798,P=0.001),and BI-RADS≥4(OR=13.547,95%CI:6.868-26.723,P<0.001)were risk factors associated with breast cancer,and tumor diameter>5 cm(OR

关 键 词：乳腺癌 诊断 影像技术 多因素分析 

分 类 号：R737.9[医药卫生—肿瘤]