基于全肿瘤MRI筛选胰腺导管腺癌SMAD4表达影像-生物学标志物

目的 利用包括影像组学特征在内的全肿瘤MRI评价，结合组织病理学和SMAD4表达分析，寻找预测胰腺导管腺癌（Pancreatic ductal adenocarcinoma, PDAC）SMAD4表达的影像-生物学标记物。 方法 60例经手术病理证实的PDAC纳入研究。所有患者均在术前于本院同一1.5T MRI接受腹部平扫及动态增强检查，并对PDAC进行包括主观评价和影像组学分析的全肿瘤影像学评价。对PDAC进行病理评价，并进行SMAD4表达分析。对影像学主观评价指标和全肿瘤定量评价进行统计分析，同时利用最大相关最小冗余（maximum relevance minimum redundancy，mRMR）算法筛选影像组学特征并进行多因素融合建模，获取可能预测SMAD4表达的影像-生物学标记物。 结果 SMAD4阳性表达者11例（18.3%）、阴性表达者49例（81.7%）。不同SMAD4表达组间仅性别差异有统计学意义（P=0.012），SMAD4阴性表达组男性比例明显高于女性，余临床资料及影像学评价指标差异均无统计学意义。经基于SMAD4表达的全肿瘤影像组学特征单因素筛选，6个影像组学特征进入多因素影像组学特征融合建模。所得最佳子集模型为门脉期_wavelet-HHL_NGTDM_Contrast+T1WI_wavelet-HHL_GLDM_HighGrayLevelEmphasis，BIC为33.4，ROC曲线下面积为0.92（P＜0.001），预测敏感度为90.9%、特异度为81.6%，组内预测准确度为83.3%。 结论 包括影像组学特征在内的全肿瘤MRI评价有望无创获取胰腺导管腺癌SMAD4表达的影像-生物学标记物，具有较好的诊断效能。

Imaging biomarkers of SMAD4 expression in pancreatic ductal adenocarcinoma: a preliminary study with whole-tumor MRI evaluation

Objective To investigate the relationship between imaging features from whole-tumor evaluation with MRI as well as radiomics and SMAD4 expression in pancreatic ductal adenocarcinoma (PDAC) and to determine imaging biomarkers of SMAD4. Methods A total of 60 patients with pathologically confirmed PDAC were included in the study. All the patients received pre-operative abdominal contrast-enhanced MRI examination with the same equipment at our institute. Whole-tumor evaluation including subjective evaluation and radiomics analysis was performed. Sections of specimens were reviewed and SMAD4 expression was evaluated. Univariate analysis, as well as radiomics feature screening and multi-factor fusion modeling, were conducted to find the imaging biomarkers that could predict SMAD4 expression. Results There were 11 cases (18.3%) with positive expression of SMAD4, and negative expression of SMAD4 was found in 49 cases (81.7%). Only the gender difference between the two groups was statistically significant. The proportion of males was significantly higher than that of females in the negative expression group. There was no significant difference in qualitative features and whole-tumor quantitative features. After single-factor screening of whole tumor radiomics features, 6 features entered into multi-factor fusion modeling. Portal phase_wavelet-HHL_NGTDM_Contrast+T1WI_wavelet-HHL_GLDM_HighGrayLevelEmphasis were enrolled as the best subset model, with the BIC of 33.4, and the area under ROC curve of 0.92 (P<0.001) for predicting the expression of SMAD4. The prediction sensitivity and specificity was 90.9% and 81.6%, with the intra-group prediction accuracy of 83.3%. Conclusion Whole-tumor evaluation with radiomics analysis is expected to noninvasively obtain the imaging biological markers of SMAD4 expression in pancreatic ductal adenocarcinoma with good performance.