CT texture analysis can be a potential tool to differentiate gastrointestinal stromal tumors without KIT exon 11 mutation
CT texture analysis can be a potential tool to differentiate gastrointestinal stromal tumors without KIT exon 11 mutation
复制标题
DOI：
10.1016/j.ejrad.2018.07.025
复制
发表时间：
2018-10-01
影响因子：
3.3
通讯作者：
Zhao, Xinming
中科院分区：
文献类型：
--
作者：
Xu, Fei;Ma, Xiaohong;Zhao, Xinming
目的：探讨CT织构分析对胃肠道间质瘤（gist）的鉴别价值。材料和方法：本研究由69名gist研究组和17名gist验证组组成。收集并分析患者的临床资料。进行了二维和三维织构分析。在研究组中评估质地参数，在验证组中进行验证。分析了纹理参数在单个感兴趣区域（single- roi）、双感兴趣区域（double-ROI）和整个感兴趣体积（whole- voi）上的重复性。采用logistic回归模型分析GIST基因型的独立预测因子。训练支持向量机（SVM）分类器，并计算6重交叉验证ROC曲线。由放射科医师给出增强CT图像上各病变的主观异质性评分，并评估相应异质性评分的差异。结果：非胃位置、较低的CD34_stain水平和较高的质地参数标准差（stdDeviation）与研究组未发生KIT外显子11突变的gist相关。结合stdDeviation、解剖位置和CD34_stain水平得到的交叉验证SVM分类器对GIST基因型的预测效率为中等到良好（AUC = 0.864-0.904）。stdDeviation是无KIT外显子11突变的GIST的独立预测因子，与研究组GIST基因型有中等相关性（AUC = 0.726-0.750）。经验证组验证，stdDeviation具有良好的性能（AUC = 0.904 ~ 0.962）。双roi改进了单roi的性能，降低了单roi的分段选择带来的方差，并显示了roi与全voi之间的良好一致性。主观异质性评分在GIST基因型之间无统计学差异。结论：CT结构分析可以在增强CT图像上帮助区分KIT外显子11突变与KIT外显子11突变的gist。
Objective: To evaluate CT texture analysis as a tool to differentiate gastrointestinal stromal tumors (GISTs) without KIT exon 11 mutation.Materials and methods: This study consisted of a study group of 69 GISTs and a validation group of 17 GISTs. Clinical information of the patients were collected and analyzed. Two-dimensional and three-dimensional texture analysis was performed. The textural parameters were evaluated in the study group and were validated in the validation group. The repeatability of the textural parameters on the single region of interest (single-ROI), double-ROI, and whole volume of interest (whole-VOI) was analyzed. The independent predictor for the GIST genotypes was analyzed with logistic regression models. The support vector machine (SVM) classifiers were also trained and 6-fold cross validation ROC curves were computed. Subjective heterogeneity scores of each lesion on enhanced CT images were given by radiologists and the corresponding difference of the heterogeneity rating was evaluated.Results: The non-gastric location, lower CD34_stain level and higher textural parameter standard Deviation (stdDeviation) were associated with the GISTs without KIT exon 11 mutation in the study group. The cross validation SVM classifiers achieved with combination of stdDeviation, anatomic location and CD34_stain level demonstrated medium to good prediction efficiency (AUC = 0.864-0.904) regarding the GIST genotypes. The stdDeviation was an independent predictor of GISTs without KIT exon 11 mutation, and had a medium correlation with the GIST genotypes in the study group (AUC = 0.726-0.750). The stdDeviation showed good performance (AUC = 0.904-0.962) when validated in the validation group. The double-ROIs improved the performances of single-ROIs, decreasing the variances of single-ROIs brought by section-selection, and demonstrating excellent agreements between ROIs and whole-VOI. Subjective heterogeneity scores had no statistically significant differences between GIST genotypes.Conclusion: CT texture analysis can potentially help to differentiate GISTs without KIT exon 11 mutation from those GISTs with KIT exon 11 mutation on enhanced CT images.