基于临床-影像组学建立子宫内膜癌微卫星不稳定性的评估模型

目的  探讨应用临床因素联合影像组学建立子宫内膜癌（EC）微卫星不稳定性（MSI）评估模型的价值。方法  回顾性分析2018年6月~2022年1月手术前经MR影像学检查及手术后病理学检测确诊为EC的患者68例，收集患者影像学及临床病理资料，根据患者微卫星稳定情况将患者分为不稳定组（n=27）和稳定组（n=41）。临床模型构建采用Logistic回归分析对临床因素进行筛选，影像组学模型构建采用3DSlicer软件勾画病灶感兴趣区并提取影像组学特征，利用最小绝对收缩和选择算子算法进行特征降维。绘制ROC曲线对影像组学模型、临床模型和临床-影像联合模型进行预测效能评估，并使用Delong检验比较3种模型的预测效能是否具有统计学差异。结果  Logistic回归分析显示，错配修复蛋白MutL同源物1、减数分裂后分离蛋白表达和肿瘤分化程度、肌层侵犯深度是EC MSI的临床危险因素。筛选6个影像组学特征用于构建影像组学模型（P < 0.05）。经ROC曲线分析，临床-影像联合模型在EC MSI中具有较好的预测及评估性能（P < 0.05），临床模型、影像组学模型及临床-影像联合模型AUC分别为0.871、0.932、0.981。Delong检验结果显示临床-影像联合模型和影像组学模型与临床模型比较，差异有统计学意义（Z=1.933、2.735，P=0.046、0.006）。结论  应用临床因素联合MR影像组学特征建立的评估模型对EC MSI具有较好的预测价值。

Establishment of microsatellite instability assessment model for endometrial carcinoma by clinic-imaging omics

  Objective  To investigate the value of microsatellite instability (MSI) assessment model of endometrial carcinoma (EC) established by clinical factors combined with imaging omics.  Methods  Sixty- eight patients with EC diagnosed by preoperative MR imaging and postoperative pathological examination from June 2018 to January 2022 were retrospectively analyzed, with imaging and clinicopathologic data collected. Patients were divided into unstable group (n=27) and stable group (n=41) according to microsatellite stability. Logistic regression analysis was used to screen the clinical factors. ITK-SNAP software was used to delineate the region of interest of the lesion and extract the image omics characteristics. The minimum absolute contraction and selection operator algorithm were used for feature dimension reduction. ROC curve was plotted to assess the predictive power of the image-omics model, the clinical model, and the combined clinical-imaging model, and Delong test was used to compare the predictive power of the three models.  Results  Logistic regression analysis showed that mismatch repair protein MutL homolog 1, post-meiotic segregation protein 2 expression, tumor differentiation and depth of myometrial invasion were clinical risk factors of EC MSI. Six image-omics features were screened for construction of imageomics models (P < 0.05). Through ROC curve analysis, the combined clinic-image model had good prediction and evaluation performance in EC MSI, and the AUCs of the clinical model, the image-omics model, and the combined clinic-image model were 0.871, 0.932 and 0.981, respectively. The results of Delong test showed that the difference between the clinical imaging combined model and the imaging omics model and the clinical model was statistically significant (Z=1.933, 2.735, P=0.046, 0.006).  Conclusion  The assessment model established by clinical factors combined with MR imaging omics characteristics has a good predictive value for EC MSI.