探讨乳腺MRI鉴别乳腺小肿块(最大直径≤1 cm)良、恶性的诊断价值。

回顾性分析2012年9月至2018年12月在上海交通大学医学院附属新华医院就诊的160例乳腺小肿块患者的术前乳腺MRI检查图像，其中良性疾病组111例，恶性疾病组49例。比较2组病灶位置、形态学特征、时间-信号强度曲线(TIC)、T2WI信号特征和表观扩散系数(ADC)的差异。年龄的组间比较采用Student’s t检验，病灶大小和ADC值的组间比较采用Mann-Whitney U检验，计数资料(病灶位置、周围脂肪征、形态、边缘、内部强化和TIC类型)组间比较采用χ²检验，伴对侧同时恶性肿瘤比例和T2WI信号特征的组间比较采用Fisher确切概率法。计数资料亚组内两两比较采用Bonferroni校正。绘制年龄和ADC值的受试者工作特征曲线，多因素分析采用二元logistic回归，使用"Forward: LR"法，建立良、恶性疾病诊断模型，计算该模型的曲线下面积、敏感度、特异度、阳性预测值、阴性预测值和正确率，输出预测模型列线图。

(1)单因素分析：良性组和恶性组的周围脂肪征、病灶边缘、内部强化类型、TIC类型、T2WI信号特征及ADC值比较，差异均有统计学意义(χ²＝13.083、11.224、7.628、14.060、21.892；P<0.001、0.004、0.006、0.001、<0.001；Z＝-3.952，P<0.001)。(2)多因素分析：年龄>50岁、周围脂肪征阳性、T2WI上略低信号或周边高信号、ADC值≤1.22×10^-3mm²/s是提示乳腺癌的独立危险特征(OR＝6.728，95%CI：2.123～21.318，P＝0.001；OR＝5.545，95%CI：1.306～23.533，P＝0.020；OR＝31.110，95%CI：2.167～446.576，P＝0.011；OR＝13.794，95%CI: 2.096～90.790，P＝0.006；OR＝5.802，95%CI：1.350～24.938，P＝0.018)。(3)鉴别诊断模型：联合上述特征的诊断模型对乳腺小肿块鉴别诊断的敏感度、特异度、阳性预测值、阴性预测值及正确率分别为89.2%(33/37)、69.4%(59/85)、55.9%(33/59)、93.6%(59/63)和75.4%(92/122)。

乳腺良、恶性小肿块的MRI图像特征表现重叠，单一特征鉴别诊断较困难。联合患者年龄、周围脂肪征、T2WI信号特征和ADC值有助于提高诊断正确率。

To investigate the value of MRI in the differential diagnosis of breast masses with a maximum diameter ≤1 cm.

This retrospective study reviewed the preoperative MRI findings of 160 patients with small breast masses in the Xinhua Hospital, School of Medicine, Shanghai Jiao Tong University between September 2012 and December 2018. There were 49 patients with malignant lesions and 111 patients with benign lesions. The mass location, morphological features, time-signal intensity curve (TIC), T2-weighted image features(T2WI) and apparent diffusion coefficient (ADC) of the masses were compared. Student’s t test was used to compare patient age between two groups. The Mann-Whitney U test was employed to compare lesion size and ADC value between two groups. The count data (mass location, surrounding fat sign, shape, margin, internal enhancement and TIC type) were compared by χ² test. The variables like proportion of concurrent contralateral cancer and T2WI features were compared by Fisher exact test. The pairwise comparison in the subgroups was adjusted by Bonferroni. Receiver operating characteristic (ROC) curve of the age and ADC value was drawn. A binary logistic regression was performed for multivariate analysis using the " Forward: LR" method to establish a diagnostic model for benign and malignant diseases. The area under the ROC curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value and accuracy rate of this model were calculated. The nomogram of the model was plotted.

(1) Univariate analysis showed that there were significant differences in surrounding fat sign, margin of mass, internal enhancement, TIC type, T2-weighted image features and ADC values between the benign group and malignant group (χ²=13.083, 11.224, 7.628, 14.060, 21.892; P<0.001, 0.004, 0.006, 0.001, <0.001; Z=-3.952, P<0.001). (2) Multivariate analysis showed that age>50 years old, positive peripheral fat sign, slightly hypointense or peripheral hyperintense signal on T2-weighted images, and ADC value ≤1.22×10^-3 mm²/s were independent risk factors for malignancy (OR=6.728, 95%CI: 2.123~21.318, P=0.001; OR=5.545, 95%CI: 1.306~23.533, P=0.020; OR=31.110, 95%CI: 2.167~446.576, P=0.011; OR=13.794, 95%CI: 2.096~90.790, P=0.006; OR=5.802, 95%CI: 1.350~24.938, P=0.018). (3) Differential diagnosis model: The sensitivity, specificity, positive predictive value, negative predictive value and accuracy rate of the diagnostic model combining the above-mentioned characteristics for the differential diagnosis of small breast masses was 89.2%(33/37), 69.4%(59/85), 55.9%(33/59), 93.6%(59/63) and 75.4%(92/122), respectively.

Benign and malignant small breast masses overlap a lot in MRI features. It is challenging to make a differential diagnosis based on single feature. The combination of age, surrounding fat sign, T2-weighted image features and ADC values helps to increase the accuracy of diagnosis.