乳腺实性肿块超声良恶性鉴别诊断的Logistic回归模型

目的建立乳腺实性肿块良恶性鉴别诊断超声特征的Logistic回归模型。方法对病理证实的90例93个乳腺实性肿块（良性组及恶性组）的超声特征进行回顾性分析，对比乳腺良恶性肿块声像图特点，通过多因素回归分析建立二分类Logistic回归模型，筛选有助于鉴别乳腺良恶性病变的主要相关超声特征，绘制ROC曲线并计算曲线下面积。结果良性组55例57个乳腺实性肿块及恶性组35例36个乳腺实性肿块声像图特征及检出率为：（1）形态规则47个（82．5％）及9个（25．O％）；（2）边界清晰50个（87．7％）及4个（11．1％）；（3）内部回声均匀52个（91．2％）及3个（8．3％）；（4）可见微钙化1个（1．8％）及8个（22．2％）；（5）有侧方回声失落26个（45．6％）及3个（8．3％）；（6）腋窝淋巴结肿大1个（1．8％）及19个（52．8％）；（7）肿块纵横比≥1者3个（5．3％）及28个（77．8％）；（8）血流RI≥0．7者1个（1／25，4．0％）及31个（31／34，91．2％）。多因素回归分析显示最后进入Logistic模型的5个特征分别为边界、内部回声、肿块纵横比、血流Adler分级和血流RI。ROC曲线下面积为0．988，标准误为0．005，95％可信区间（0．978，0．998）。结论以超声特征的乳腺实性肿块良恶性鉴别的Logistic回归模型有助于对乳腺实性肿块的良恶性进行鉴别诊断。

A Logistic regression model for differential diagnosis of benign and malignant solid breast masses by unltrasound

Objective To establish a logistic regression model for uhrasongraphic differential diag- nosis of benign and malignant solid breast masses. Methods A total of ninety three cases of solid breast masses in ninety patients （ fifty seven benign masses in fifty-five patients, thirty-six malignant masses in thirty-five patients）diagnosed by pathology were analyzed retrospectively. The Uhrasonographic characteristics of benign and malignant masses were compared. The ultrasonographic parameters of breast benign or malignant masses were analyzed by multiple stepwise binary logistic regression analysis, ROC curve was drawn and the areas under curves were calculated. Results The benign breast masses showed a smooth shape （47/57, 82.5% ）, clear edge（50/57,87.7% ）, homogeneous internal echoes（52/57, 91.2% ）, micro calcifications （ 1/57, 1.8% ）, lateral wall echo dorp-out （26/57, 45.6% ）, axillart lymph nodes diameter ≥10 mm （ 1/57, 1.8% ）, Longitudinal-Transverse axis ratio≥ 1 （3/57, 5.3% ）, and RI of blood flow≥0.7 （ 1/25, 4.0% ）. The malignant breast masses showed a smooth shape（9/36, 25. % ）, clear edge（4/36, 11.1% ）, homogeneous internal echoes（3/36, 8.3% ）, micro calcifications （8/36, 22.2% ）, lateral wall echo dorpout（3/36, 8.3% ）, axillart lymph nodes diameter 310 ram（ 19/36, 52.8% ）, longitudinal-transverse axis ratio≥1 （28/36, 77.8% ）, and RI of blood flow≥0.7 （31/34, 91.2% ）. Five ultrasonic features were finally applied into the Logistic regression model includingedge, Echogenieity, Longitudinal-Transverse axis ratio, Adler grade of blood flow and the RI of blood flow. The area under the R0C curve was 0. 988, the standard error was 0. 005, and the 95% confidence interval was between 0. 978 and 0. 998. Conclusion The logistic regression model can be helpful for classification of benign and malignant breast masses.