基于磁共振表观扩散系数图像的放射组学特征评估胃肠道间质瘤危险度

Evaluation of the Risk of Gastrointestinal Stromal Tumor Based on Radiomics Features of Magnetic Resonance Apparent Diffusion Coefficient Images

目的探讨基于MR表观扩散系数(ADC)图像提取的纹理特征参数评估胃肠道间质瘤(GIST)危险度的价值。资料与方法纳入术前行MRI扫描,且病理诊断为GIST的26例患者。对患者ADC图像进行放射组学纹理特征分析,半自动分割感兴趣区(ROI)。通过灰度共生矩阵、灰度游程矩阵及形状因子参数方法获取病灶ADC图像的放射组学特征。在显微镜下计数肿瘤组织细胞的核分裂数。根据不同核分裂数将肿瘤分为中-高危组(17例)和低危组(9例)。分析组间放射组学特征参数的差异。对组间差异有统计学意义的纹理参数进行Logistic回归分析以筛选出可用于鉴别肿瘤危险度的特征参数,并通过受试者工作特征(ROC)曲线评价该特征参数鉴别肿瘤危险度的效能。结果相关、灰度共生矩阵能量、灰度共生矩阵熵、最大三维直径、短距离强度在中-高危组与低危组间差异有统计学意义(P<0.05),对应ROC曲线下面积分别为0.804、0.797、0.752、0.908、0.784;5项纹理特征参数拟合后ROC曲线下面积为0.928。结论基于MRADC图像提取的放射组学特征纹理参数作为一种无创的方法对于评估GIST患者肿瘤病理危险度具有较高的价值。

Purpose To explore the value of texture feature parameters extracted from MR apparent diffusion coefficient(ADC)images to evaluate the risk of gastrointestinal stromal tumors(GIST).Materials and Methods Twenty-six patients with gastrointestinal stromal tumors confirmed by pathology who underwent preoperative MRI were enrolled in this study.The Radiomics of patients’ADC images were analyzed,and semi-automatically segmenting the region of interest(ROI).Using three parameters calculated methods,including GLCM,RLM and form factor parameters.According to different mitotic numbers,the tumors were divided into middle-high risk group and low risk group.The difference between the two groups was analyzed.Multivariate Logistic regression and ROC curve were used to test the diagnostic efficiency of texture feature parameters.Results The correlation,GLCM energy,GLCM entropy,maximum threedimensional diameter and short-distance intensity were statistically significant between the high-risk group and the low-risk group(P<0.05),and the corresponding areas under the ROC curve were 0.804,0.797,0.752,0.908,0.784;after fitting the five texture feature parameters,the area under the ROC curve was 0.928.Conclusion The texture parameters of radiology features extracted based on MR ADC images have a high value for evaluating the pathological risk of tumors in GIST patients as a noninvasive method.