基于MRI影像组学鉴别浸润性与非浸润性乳腺癌

MRI radiomics in identifying invasive and non-invasive breast cancer

目的探讨基于MRI影像组学鉴别浸润性与非浸润性乳腺癌的价值。方法回顾性分析100例接受乳腺常规MR和动态增强扫描的乳腺癌患者(75例浸润性、25例非浸润性),将其分为训练组(n=70)和验证组(n=30)。提取病灶纹理特征,采用最小绝对缩减和变量选择算子(LASSO)回归对训练组纹理特征进行降维,建立影像组学标签。比较浸润性与非浸润性乳腺癌临床、病理及影像学特征,以多因素Logistic回归分析建立影像组学模型,采用ROC曲线评价模型的诊断效能。结果共提取3132个影像学特征,经LASSO回归降维获得19个价值较高者,建立影像组学标签。浸润性与非浸润性乳腺癌之间,训练组和验证组毛刺、基底细胞角蛋白(CK5/6)、瘤细胞增殖因子(Ki-67)和影像组学标签差异均有统计学意义(P均<0.05),训练组时间-强度曲线(TIC)类型差异有统计学意义(P<0.05),验证组TIC类型差异无统计学意义(P>0.05)。训练组CK5/6、Ki-67和影像组学标签为浸润性乳腺癌的独立危险因素(P均<0.05);以其构建影像组学模型,在训练组和验证组鉴别浸润性乳腺癌的AUC分别为0.97和0.85,均优于CK5/6、Ki-67和影像组学标签。结论基于MRI影像组学模型鉴别浸润性与非浸润性乳腺癌效果较好。

Objective To investigate the value of radiomics based on MRI for differentiating invasive and non-invasive breast cancer.Methods Data of 100 patients with breast cancer who underwent conventional MR and dynamic contrast-enhanced scanning(75 cases of invasive and 25 of non-invasive)were retrospectively analyzed.The patients were divided into training group(n=70)and validation group(n=30).The texture features of lesions were extracted,least absolute shrinkage and selection operator(LASSO)regression was used to reduce the dimension of texture features in training group to establish imaging label.The clinical,pathological and imaging features of invasive and non-invasive breast cancer were compared.Then radiomics model was established with multiple Logistic regression analysis,and ROC curve was used to evaluate the diagnostic efficiency of this model.Results A total of 3132 imaging features were extracted,and 19 more valuable features were obtained with LASSO regression to build the radiomics label.Significant differences of burr,basal cell keratin(CK5/6),tumor cell proliferation factor(Ki-67)and the radiomics signatures were found between invasive and non-invasive breast cancer in training group and validation group(all P<0.05).Time-intensity curve(TIC)types of invasive and non-invasive breast cancer were significantly different in training group(P<0.05),but not in the validation group(P>0.05).CK5/6,Ki-67 and imaging radiomics signatures were independent risk factors for invasive breast cancer in training group(all P<0.05),and were used to build radiomics model,and the AUC of the model for differentiating invasive breast cancer was 0.97 and 0.85 in training group and validation group,respectively,superior to CK5/6,Ki-67 and radiomics signatures.Conclusion Radiomics model based on MRI had rather good diagnostic value for differentiating invasive and non-invasive breast cancer.