基于Gd-EOB-DTPA增强MRI纹理分析与血清学指标预测肝细胞癌微血管侵犯

The value of Gd-EOB-DTPA enhanced magnetic resonance texture analysis combined with serological indicators in predicting microvascular invasion in patients with hepatocellular carcinoma

目的:探讨基于Gd-EOB-DTPA增强肝胆期MRI的纹理分析技术联合血清学指标对肝细胞癌(HCC)微血管侵犯(MVI)的预测价值。方法:回顾性分析2018年12月-2022年6月在本院经手术病理确诊为HCC并在术前1个月内完成肝脏Gd-EOB-DTPA增强扫描(动脉期、门静脉期、延迟期及肝胆期)的206例患者的临床和MRI资料。记录术前甲胎蛋白(AFP)、谷丙转氨酶(ALT)、谷草转氨酶(AST)和PIVKA-Ⅱ四项血清学指标。按照病理检查结果分为MVI组和非MVI组。按照9:1的比例将所有患者分为训练集(MVI组91例,非MVI组94例)和验证集(MVI组9例,非MVI组12例)。分析两组患者基于肝胆期图像提取的HCC病灶的纹理特征及血清学指标的差异。纹理特征主要包括图像像素值的一阶特征和描述肿瘤内部及表面纹理的灰阶分布特征。采用多因素logistic回归分析筛选HCC发生MVI的独立危险因素,采用ROC曲线分析评估各单一变量及联合模型对MVI的预测效能。结果:纹理分析结果显示,与非MVI组相比,MVI组的GLCM_Energy、SumAverg、DifVarnc和CE_Entropy显著增高,而Teta-2显著降低,差异均有统计学意义(P<0.05)。血清学指标方面,MVI组的谷丙转氨酶(ALT)、甲胎蛋白(AFP)和异常凝血酶原(PIVKA-Ⅱ)水平均显著高于非MVI组,差异均有统计学意义(P<0.05)。Logistic回归分析结果显示GLCM_Energy(AUC=0.771),CE_Entropy(AUC=0.719)和PIVKA-Ⅱ(AUC=0.769)对MVI具有一定的预测效能,三个指标联合后具有更好的预测效能(AUC=0.888)。验证组中数据的逻辑回归分析结果显示2项纹理特征(GLCM_Energy、CE_Entropy)和血清学指标PIVKA-Ⅱ可预测MVI(P<0.001),AUC分别为0.688、0.684和0.758;而且将3项指标联合后可观察到更好的预测效能(AUC=0.860)。结论:基于Gd-EOB-DTPA增强肝胆期MRI提取的纹理特征(GLCM_Energy和CE_Entropy)联合血清学指标PIVKA-Ⅱ在术前预测HCC伴有MVI方面具有一定的价值。

Objective:To explore the predictive value of texture analysis technology combined with serological indicators based on Gd-EOB-DTPA enhanced hepatobiliary MRI for microvascular invasion(MVI)in hepatocellular carcinoma(HCC).Methods:The clinical and MRI data of 206 patients with HCC confirmed by surgery and pathology in our hospital from December 2018 to June 2022 were retrospectively analyzed.All patients underwent liver Gd-EOB-DTPA enhanced four-phase(arterial phase,portal phase,delayed phase and hepatobiliary phase)MRI scans within one month before surgery.The results of four serological indicators including preoperative alpha fetoprotein(AFP),alanine aminotransferase(ALT),aspartate aminotransferase(AST),and PIVKA-Ⅱwere recorded.According to the pathological examination results,the patients were divided into MVI group and non-MVI group.All patients were divided into training set(91 in the MVI group and 94 in the non-MVI group)and validation set(9 in the MVI group and 12 in the non-MVI group)according to the ratio of 9:1.The differences in texture features of HCC extracted from hepatobiliary phase images and serological indicators between MVI group and non-MVI group were analyzed.Texture features mainly include first-order features of image pixel values and grayscale distribution features that describe the internal and surface textures of tumors.Multiple logistic regression analysis was used to select independent risk factors for MVI in HCC,and ROC curves were plotted for the predictive efficacy of each single variable and combined model for MVI.Results:The results of texture analysis showed that GLCM_Energy,SumAverg,DifVarnc and CE_Entropy were significantly higher and Teta-2 was significantly lower in the MVI group compared with the non-MVI group,with statistically significant differences(all P<0.05).The levels of the four serological indexes(ALT,AFP and PIVKA-Ⅱ)were significantly higher in the MVI group than those in the non-MVI group,with statistically significant differences(all P<0.05).Logistic regression showed that GLCM_Energy(AUC=0.771),CE_Entropy(AUC=0.719)and PIVKA-Ⅱ(AUC=0.769)were predictors for the occurrence of MVI,and the combination of the three parameters had better predictive efficacy(AUC=0.888)than single parameter in the training set.In the validation group,the texture feature of GLCM_Energy and CE_Entropy,and serological indicator of PIVKA-Ⅱcould also be used to predict MVI with AUC of 0.688,0.684 and 0.758(all P<0.001);and when these three indicators were combined,higher predictive efficacy can be observed(AUC=0.860).Conclusion:The combination of texture features(GLCM_Energy and CE_Entropy)from GD-EOB-DTPA enhanced hepatobiliary images and serological marker of PIVKA-Ⅱhas high predictive value for the presence of MVI in HCC before surgery.