基于增强CT的胃肠道间质瘤纹理分析对术前危险度分级的预测价值

The value of texture analysis based on enhanced CT imaging in predicting risk classification of Gastrointestinal Stromal Tumor

目的探讨增强CT纹理分析预测胃肠道间质瘤(GISTs)危险度分级的可行性。方法选取本院55例经手术病理证实的胃肠道间质瘤的影像学资料。基于增强CT扫描成像的直方图纹理分析及灰度共生矩阵纹理分析提取与肿瘤异质性的相关的纹理参数,包括平均值(mean)、方差(Variance)、偏度(skewness)、峰度(kurtosis)、10%分位像素值、50%分位像素值、90%分位像素值、能量(energy)、自相关(correlation)、对比度(Contrast)及熵值(entropy)纹理参数。根据术后病理结果分成极低/低危组、中危组、高危组。分析不同组纹理参数差异及纹理参数与肿瘤危险度分级、直径、Ki-67的相关性。采用ROC曲线分析纹理参数鉴别极低/低危组与中-高危组间质瘤的价值。结果均值、偏度、方差、10%分位像素值、50%分位像素值、90%分位像素值、对比度及自相关在不同危险度GISTs间差异均无统计学意义(均P>0.05);峰度、能量及熵在三组间存在差异,差异均有统计学意义(均P<0.05)。峰度及熵与肿瘤长径、Ki-67指数及肿瘤危险度分级呈正相关,能量与三者呈负相关关系,相关性具有统计学意义(P<0.05)。ROC曲线分析显示,峰度0.63为阈值时,诊断的敏感度为71.8%,特异度为63.7%;能量0.20为阈值时,敏感度为79.4%,特异度为70.1%;熵界1.10为阈值时,敏感度为85.3%,特异度为76.8%。结论基于增强CT的纹理分析可在胃肠道胶质瘤术前危险度分级中发挥重要作用,其中以能量和熵值参数价值最高。

Objective To investigate the feasibility of predicting the risk of gastrointestinal stromal tumors(GISTs) with CT enhanced texture analysis. Methods 55 cases with gastrointestinal stromal tumors confirmed by surgical pathology were included in our study. Parameters related to tumor heterogeneity obased on Histogram texture analysis and grayscale co-occurrence matrix texture analysis were extracted, including mean, Variance, skewness, kurtosis kurtosis, 10% pixel values, 50% pixel values, 90% pixel values, energy, correlation, contrast and entropy. According to postoperative pathological results, GISTs were divided into very low/low risk group, medium risk group, and high risk group. The differences in texture parameters and the correlation between texture parameters and tumor grade, diameter and Ki-67 were analyzed. Diagnostic performance of texture parameters were analyzed for predicting very low/low risk group and medium-high risk group with ROC curve. Results There was no significant difference in mean, skewness, variance, 10% pixel values, 50% pixel values, 90% pixel values, contrast and autocorrelation between different risk GISTs(all P>0.05). There were significant differences in kurtosis, energy and entropy between the three groups(all P<0.05). The kurtosis and entropy were positively correlated with tumor diameter, Ki-67 index and tumor risk grade(all P<0.05). There was a negative correlation between energy and tumor diameter, Ki-67 index and tumor risk grade(all P<0.05). The results of ROC curve analysis showed that the sensitivity was 71.8% and the specificity was 63.7% when the cut-off value of kurtosis was 0.63;the sensitivity was 79.4% and the specificity was 70.1% when the cut-off value of energy was 0.20;The sensitivity was 85.3% and the specificity was 76.8% when the cut-off value of entropy was 1.10. Conclusion Texture analysis based on enhanced CT may play an important role in assessing preoperative risk classification of GISTs, of which energy and entropy parameters are the most valuable.