基于DW-MRI纹理分析建立预测宫颈鳞状细胞癌放化疗敏感性模型的研究

Model Study of Texture Analysis Based on DW-MRI to Predict the Chemoradiosensitivity in Cervical Cancer

目的建立基于磁共振弥散加权成像(DW-MRI)图像的纹理分析预测宫颈鳞状细胞癌放化疗敏感性的数学模型并进一步评价其临床价值。方法收集2016年3月至2017年11月秦皇岛市第一医院收治的30例经病理诊断为宫颈鳞状细胞癌并准备接受同步放化疗的患者为研究对象。在治疗前进行MRI检测,得到其弥散加权成像(DWI)和表观弥散系数(ADC)图像。对于每组图像均由两个影像医师在肿瘤最大层面进行肿瘤的感兴趣区勾画并用软件Mazda version 4.6分析,提取出一系列纹理参数。采用一致性相关系数和动态范围矩阵方法验证参数的可重复性和冗余度,之后采用t检验筛选有意义的纹理参数。将筛选得到的参数进行LASSO-Logistic回归分析,建立治疗敏感性预测模型。并进一步应用受试者工作特征曲线(ROC曲线)评估该模型的预测准确性。结果抗拒组DWI图像中的偏度、和熵、熵、差异熵、差值方差高于敏感组,抗拒组ADC图像中的峰度、熵、对比度、和方差高于敏感组,和均值低于敏感组(P<0.05),且具有较强的可重复性。LASSO-Logistic回归分析结果显示和熵以及和均值纳入预测概率计算公式中。ROC曲线评估结果显示ROC曲线下的面积为0.786 7(P<0.05)。结论基于DW-MRI的纹理分析方法在预测宫颈鳞状细胞癌放化疗敏感性方面有较高的准确性,有一定的临床价值。

Objective To construct and evaluate the accuracy of the predictive model on the chemoradiosensitivity of cervical cancer by performing texture analysis based on the picture of diffusion-weighted magnetic resonance imaging（DW-MRI）.Methods We retrospectively analyzed 30 patients with pathological diagnosis of cervical cancer admitted in Qinghuangdao First Hospital from Mar.2016 to Nov.2017 and collected the imaging pictures of DW-MRI and apparent diffusion coefficient（ADC） before their treatment.The region of interest was delineated by two radiologists for each sequence.The Mazda version 4.6 was applied to obtain the texture parameters from ROC and then statistically filtered to identify a subset of reproducible and non-redundant parameters that were used to construct the predictive model.LASSO-Logistic regression model along with 10-fold cross validation was performed to select the texture parameters associated with the outcome of chemo-radiotherapy and constructed a predictive signature.Then the predictive efficiency of the signature was evaluated using receiver operating characteristic（ROC） curves.Results The skewness,entropy,differential entropy,difference variance of DWI images in the resistive group were higher than those in the sensitive group,and the kurtosis,entropy,contrast and variance in ADC images in the resistive group were higher than those in the sensitive group,and the mean value was lower than that in the sensitive group（P 0.05）,and there was strong reproducibility.LASSO-Logistic regression analysis showed that entropy and mean value were included in the prediction probability formula.The evaluation results showed that the area under ROC curve was 0.786 7（P 0.05）.Conclusion Texture analysis based on DW-MRI could be used to help predict the chemoradiosensitivity of cervical cancer with high accuracy and a certain clinical value.