基于CT图像纹理分析预测胸腺上皮性肿瘤Masaoka-Koga分期的研究

Application of CT texture analysis in predicting Masaoka-Koga staging of thymic epithelial tumors

目的探讨CT图像纹理分析在预测胸腺上皮性肿瘤(TETs)Masaoka-Koga分期中的价值。方法回顾性分析经手术病理证实的114例TETs患者的CT图像,根据Masaoka-Koga分期将TETs分为早期(Ⅰ~Ⅱ期)与进展期(Ⅲ~Ⅳ期)。提取CT平扫及增强图像上肿瘤各期的纹理参数,对纹理参数进行降维后求得加权Rad-score值,使用受试者工作特征(ROC)曲线分析纹理特征的预测效能。结果114例TETs患者中包括男58例,女56例;早期61例,进展期53例。基于CT图像纹理分析,在鉴别早期和进展期TETs时CT平扫参数V_(14)(Kurtosis)、V_(462)(Correlation_angle90_offset9)、V_(481)(Sum Entropy)等参数具有统计学意义(均P<0.05),动脉期V_(16)(Histogram Energy)、V_(32)(Percentile70)、V_(481)(Sum Entropy)等参数具有统计学意义(均P<0.05);静脉期V_(148)(GLCM Entropy_offset3_SD)、V_(469)(Inertia_angle135_offset9)、V_(966)(LRE_offset9)等参数具有统计学意义(均P<0.001);Rad-score值在鉴别早期和进展期TETs方面:CT平扫、动脉期和静脉期的AUC值分别为0.692(0.595,0.789)、0.878(0.813,0.944)和0.685(0.603,0.767)。结论CT图像纹理分析在术前预测TETs的Masaoka-Koga分期中具有潜在价值,有助于对TETs患者进行个性化诊疗。

Objective To investigate the value of CT texture analysis in predicting Masaoka-Koga staging of thymic epithelial tumors(TETs).Methods CT images of 114 patients(58 men,56 women)with histologically confirmed TETs in early(61)and advanced(53)stages from 2009 to 2015 were retrospectively analyzed.The texture parameters of each stage of tumor on unenhanced and enhanced CT were extracted and the weighted RAD-score values were obtained after screening and dimensionality reduction.Receiver operating characteristic(ROC)curve was used to analyze the prediction efficiency of texture features.Results The weighted parameter RAD-score values were significantly different between early and advanced TETs for V_(14)(Kurtosis,P<0.001),V_(462)(Correlation_angle90_offset9,P<0.001),and V_(481)(Sum Entropy,P=0.001)on unenhanced CT with area under the ROC curve(AUC)of 0.692(95% confidence interval:0.595-0.789).In the arterial phase,V16(Histogram Energy,P<0.001),V_(32)(Percentile70,P=0.032),and V_(481)(Sum Entropy,P<0.001)had highest tumor stage prediction efficiency for with AUC of 0.878(0.813-0.944).In the venous phase,V_(148)(GLCM Enteropy_offset3_SD,P<0.001),V469(Inertia_angle135_offset9,P<0.001),and V_(966)(LRE_offset9,P<0.001)were efficient in predicting the tumor stage with AUC of 0.685(0.603-0.767).Conclusion CT texture analysis is useful in predicting the Masaoka-Koga stage for personalized treatment planning in patients with TETs.The arterial phase CT texture parameters have the highest predictive efficiency.