【摘要】 目的 建立基于乳腺MRI区分乳腺导管原位癌(DCIS)与导管原位癌伴微浸润(DCISM)的影像组学模型并验证其价值。方法 回顾性分析87例(DCISM 32例，DCIS 55例)女性，按7∶3分为训练组和验证组。分别搜集MRI图像及临床、病理和影像学资料。采用3D Slicer软件手动勾画乳腺癌灶的3D感兴趣区(ROI)并提取特征，使用最大相关性最小冗余性算法和最小绝对收缩和选择算法(LASSO)回归选择特征并建立影像组学模型。根据临床和影像学特征建立临床模型。基于影像组学评分(Radscore)和临床模型建立联合模型。结果 区分DCIS是否伴微浸润的独立影响因子包含时间-信号强度曲线(TIC)和表观扩散系数(ADC)。尽管临床模型和影像组学模型的曲线下面积(AUC)没有显著差异(分别为0.760和0.830),结合Radscore和临床影像学特征的联合模型表现了鉴别DCIS与DCISM的良好效能(AUC为0.840)。结论 结合Radscore和临床影像学特征的联合模型可为鉴别DCIS与DCISM提供新的手段。更多还原

【Abstract】 Objective To build and verify a nomogram for diagnosis microinvasion of DCIS ground on magnetic resonance imaging(MRI). Methods A total of 87(32 of DCIS with microinvasion and 55 of DCIS) women were retrospectively analyzed and were divided into training cohort and testing cohort on the basis of 7∶ 3. Clinical,pathological and MRI imaging features were collected. We delineated region of interest manually on primary lesion on MRI,and used RedundancyMaximum Relevance and Least absolute shrinkage and selection operator to select the features and build the radiomics model. Radiomics score(Radscore) were obtained from radiomics model. Clinical model was built on the basis of the clinical and radiological features. The combined nomogram was built based on radscore and clinical radiological features. We compared the diagnostic efficiency and clinical adaptability of different models.Results Time signal intensity curve and apparent diffusion coefficient were independent risk factors for diagnosis microinvasion of DCIS. The combined nomogram incorporating radscore and clinical radiological features showed a good calibration for diagnosis microinvasion of DCIS(0. 860of AUC). Although our result showed no significant difference with clinical model and radiomics model(0. 760 and 0. 830of AUC).Conclusion Our result shows that the combined nomogram built with MRI and clinical radiological features has potential to identify the microinvasion of DCIS.更多还原