摘要
目的建立应用高频彩色多普勒超声对乳腺肿块良恶性鉴别诊断的Fisher逐步判别分析模型。方法回顾性分析经病理证实的216例乳腺肿块患者,应用卡方检验比较良性组和恶性组的彩色多普勒超声影像特征,采用Fisher逐步判别法建立模型和验证模型的判别效果,并绘制ROC曲线进行评价。结果 216例乳腺肿块患者中良性144例,恶性72例。卡方检验结果显示良性组和恶性组的10个影像特征均有统计学差异。利用逐步法建立Fisher判别分析模型,最终得到一个包括包膜、毛刺或蟹足、回声均匀与否和后方回声共4个变量的判别函数,通过自身验证法,交互验证法和外部数据验证法对该判别模型进行验证,灵敏度分别是93.1%、90.3%、88%,特异度分别是96.5%、95.8%、92.7%,符合率分别是95.4%、94.0%、90.9%,说明该模型预测效果较为理想。ROC分析曲线下面积达到0.988,也说明模型的诊断价值较高。结论判别分析有助于高频彩色多普勒超声对乳腺肿块良恶性的鉴别诊断,可为临床医生的诊断提供参考,为乳腺癌的早期发现和筛选提供预警信息。
Objective To establish a Fisher stepwise discriminant analysis model for differential diagnosis of benign and malignant breast masses by high frequency and color doppler ultrasound. Methods 216 cases of breast masses diagnosed patho- logically were analyzed retrospectively. The ultrasonic imaging features of benign and malignant masses were recorded and com- pared by X2 test. Fisher stepwise discriminant analysis was used to build model and the effects of model were tested. The ROC curve was drawn and the areas under curves were calculated. Results There are 144 benign masses, and 72 malignant masses from 216 cases.x2 test showed that the differences of 10 features of ultrasonic imaging between benign and malignant groups were all statistically significant. Fisher stepwise discriminant analysis showed that envelope,burr or crab foot, echo and posterior echo are the significant independent variable of the discriminant function, as tested by the methods of internal data, cross-valida- ted and external data, the sensitivities were 93.1%, 90. 3 %, 88 % , the specificities were 96. 5 % , 95.8 % , 92.7 %, and coinci- dence rates were 95.4% ,94. 0% ,90. 9%. And the area under the ROC curve was 0. 988. All of these showed that the model have a high diagnosis value. Conclusion The discriminant analysis model can be helpful for differentiation of benign or malig- nant breast masses by high frequency and color doppler ultrasound. It can provide a reference and warning information for the screening and early detection of breast cancer by clinical doctors.
出处
《中国卫生统计》
CSCD
北大核心
2014年第1期13-15,共3页
Chinese Journal of Health Statistics
关键词
乳腺肿块
超声影像特征
判别分析
Breast masses
Ultrasonic imaging features
Discriminant analysis
