基于CT增强动脉期图像的纹理分析对预测结直肠癌患者K-ras基因突变状态的价值  

The value of CT arterial phase texture analysis in predicting K-ras gene mutation status in colorectal cancer patients

作　　者：郭玉琢 刘爱连[1] 武敬君 刘宇卉 赵莹[1] 陈安良[1] 张钦和 李昕 吴艇帆 郭妍 GUO Yu-zhuo;LIU Ai-lian;WU Jing-jun;LIU Yu-hui;ZHAO Ying;CHEN An-liang;ZHANG Qin-he;LI Xin;WU Ting-fan;GUO Yan(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Dalian Liaoning 116011,China;College of Medical Imaging,Dalian Medical University,Dalian Liaoning 116044,China;GE Healthcare,Shanghai 200000,China)

机构地区：[1]大连医科大学附属第一医院放射科,辽宁大连116011 [2]大连医科大学医学影像学院,辽宁大连116044 [3]通用电气医疗,上海200000

出　　处：《中国临床医学影像杂志》2022年第6期419-424,共6页Journal of China Clinic Medical Imaging

基　　金：国家自然科学基金资助项目(项目编号:61971091)。

摘　　要：目的:探讨CT动脉期图像纹理分析在预测结直肠癌(Colorectal cancer,CRC)K-ras基因突变状态中的价值。方法:回顾性分析44例经术后病理证实的CRC患者(22例K-ras突变型,22例K-ras野生型)。所有CRC患者均经术后病理检查证实,且术前均行CT全腹增强扫描。在AW4.6工作站获取CT动脉期图像并导入ITK-SNAP软件,由两名观察者分别使用AK软件提取21个纹理参数(最大强度、均值偏差、均值、中强度等)。采用RF模型筛选出两名观察者勾画具有高度一致性的参数,采用独立样本T检验或Mann-Whitney U检验对两组纹理参数进行比较分析。采用Logistic回归方法对纹理参数进行联合。通过ROC分析测试纹理参数的诊断效能。通过Delong试验比较两个模型AUC之间差异。结果:增强扫描动脉期最小强度、百分位数5、百分位数10、百分位数15、百分位数20、百分位数25、分位数0.025、分位数0.25、方差、均匀性在两名观察者间差异有统计学意义(P<0.05),上述各纹理参数的AUC分别为0.778、0.748、0.711、0.686、0.661、0.787、0.789、0.661、0.748、0.806。结合所有组间差异显著的纹理参数,ROC曲线下面积提高到0.911(敏感性77.3%,特异性95.5%)。Delong试验表明两种模型AUC间差异无统计学意义(P>0.05)。结论:增强CT动脉期影像纹理分析能够潜在的、无创的预测CRC患者K-ras基因突变状态。Objective: To investigate the value of CT arterial phase image texture analysis in the prediction of K-ras gene mutation status in colorectal cancer(CRC). Methods: A retrospective analysis of 44 patients with CRC confirmed by postoperative pathology(22 cases of K-ras mutant type, 22 cases of K-ras wild type) was performed. All CRC patients were confirmed by postoperative pathological examinations, and CT scans and enhanced scans were performed before surgery. CT arterial phase images were obtained by AW4.6 workstation and imported into ITK-SNAP software. Two observers used AK(Artificial Intelligent Kit, AK) software to extract 21 texture parameters(maximum intensity, mean deviation, mean value, medium intensity,etc.). The RF model was used to screen out the highly consistent parameters outlined by two observers, and the independent sample T test or Mann-Whitney U test was used to compare and analyze the two sets of texture parameters. The Logistic regression method was used to combine texture parameters. The diagnostic performance of the test texture parameters was analyzed by receiver operating characteristic curve(ROC). The difference of AUC between the two models was compared by the Delong test. Results: In enhanced scans arterial phase, there were significant differences in minimum intensity, Percentile5,Percentile10, Percentile15, Percentile20, Percentile25, Quantity0.025, Quantity0.25, Variance, Uniformity between the two groups(P<0.05), and the AUC of each texture parameter was 0.778, 0.748, 0.711, 0.686, 0.661, 0.787, 0.789, 0.661, 0.748, 0.806.Combined with all texture parameters with significant differences between groups, the area under ROC curve was increased to0.911(sensitivity 77.3%, specificity 95.5%). Delong test showed that there was no significant difference in AUC between the two models(P>0.05). Conclusion: Texture analysis of enhanced CT arterial phase images can potentially and noninvasively predict the K-ras gene mutation status of CRC patients.

关 键 词：结直肠肿瘤 基因 ras 体层摄影术 X线计算机 

分 类 号：R735.35[医药卫生—肿瘤] R735.37[医药卫生—临床医学] R814.42