MRI radiomics based on deep learning 3D super-resolution reconstruction technology for predicting the efficacy of TACE combined with molecular targeted drugs in the treatment of unresectable hepatocellular carcinoma
MRI radiomics based on deep learning 3D super-resolution reconstruction technology for predicting the efficacy of TACE combined with molecular targeted drugs in the treatment of unresectable hepatocellular carcinoma
董亚宁 1朱菊芳 1毛珂 1翟晓阳 1段金辉 1韩东明1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 新乡医学院第一附属医院磁共振科,新乡 453100
折叠
摘要
目的 探讨基于深度学习的3D超分辨率重建技术的MRI影像组学在预测经动脉化疗栓塞术(transcatheter arterial chemoembolization,TACE)联合分子靶向药物治疗不可切除肝癌疗效中的价值.材料与方法 回顾性分析122例原发性肝癌(hepatocellular carcinoma,HCC)患者的资料,根据改良实体瘤疗效评价标准(modified response evaluation criteria in solid tumors,mRECIST)分为客观反应组(完全缓解+部分缓解)68例和无客观反应组(疾病进展+病情稳定)54例.基于生成对抗网络的3D超分辨率重建技术将MRI增强动脉早期图像分辨率提高至原来的2倍.以8∶2的比例随机分为训练集及验证集,于重建前后图像分别勾画感兴趣区体积进行影像组学分析筛选影像组学特征并计算影像组学评分.采用logistic回归建立重建前及重建后影像组学模型并筛选临床特征建立临床模型.采用受试者工作特征(receiver operating characteristic,ROC)曲线评估模型性能,采用DeLong检验比较曲线下面积(area under the curve,AUC).采用决策曲线(decision curve analysis,DCA)分析各模型的临床价值.结果 logistic回归分析表明肿瘤直径[比值比(odds ratio,OR)值=1.311,95%置信区间(confidence interval,CI)=1.112～1.547,P<0.001]与动脉期强化(OR值=9.466,95%CI=2.489～36.001,P<0.001)是联合治疗疗效的独立预测因素.重建后影像组学模型预测性能最佳,训练集与验证集AUC分别为0.883(95%CI:0.814～0.952)、0.844(95%CI:0.656～0.999),高于重建前影像组学模型0.847(95%CI:0.765～0.928)、0.753(95%CI:0.554～0.953)及临床模型0.834(95%CI:0.754～0.914)、0.760(95%CI:0.564-0.956).然而,训练集及验证集中各个模型的AUC差异不具有统计学意义(P均>0.05).DCA表明训练集在大于0.34、验证集在0.36～0.59以及大于0.71阈值范围内,重建后影像组学模型临床净收益最大.结论 采用基于生成对抗网络的3D超分辨率重建技术的MRI影像组学在预测TACE联合分子靶向药物治疗不可切除肝癌疗效中具有一定的应用价值.
Abstract
Objective:To explore the value of MRI radiomics based on deep learning three-dimensional(3D)super-resolution reconstruction technology in predicting the efficacy of transcatheter arterial chemoembolization(TACE)combined with molecularly targeted drugs in treating unresectable hepatocellular carcinoma(HCC).Materials and Methods:A retrospective analysis was conducted on data from 122 patients with primary HCC,divided into an objective response group(complete remission+partial remission,n=68)and a non-objective response group(progressive disease+stable disease,n=54)according to the modified response evaluation criteria in solid tumors(mRECIST).A 3D super-resolution reconstruction technique based on generative adversarial networks was used to double the resolution of MRI-enhanced arterial early images.The dataset was randomly divided into training and validation sets in an 8∶2 ratio.Radiomic features were extracted from volume of interests delineated on both pre-and post-reconstructed images,and subsequently,radiomic scores were calculated.Logistic regression classifiers were used to establish radiomic models for both pre-and post-reconstructed images.Multivariable logistic regression was employed to screen clinical characteristics and establish a clinical model.Model performance was evaluated using receiver operating characteristic(ROC)curves,with area under the curve(AUC)compared via DeLong's test.Decision curve analysis(DCA)was used to assess the clinical value of each model.Results:Logistic regression analysis identified tumor diameter[odds ratio(OR)=1.311,95%confidence interval(CI)=1.112-1.547,P<0.001]and arterial phase enhancement(OR=9.466,95%CI=2.489-36.001,P<0.001)as independent predictors of treatment efficacy for HCC.The post-reconstruction radiomic model exhibited the best predictive performance,with an AUC of 0.883(95%CI:0.814-0.952)in the training set and 0.844(95%CI:0.656-0.999)in the validation set.These results surpassed those of the pre-reconstruction radiomic model,which had AUC values of 0.847(95%CI:0.765-0.928)and 0.753(95%CI:0.554-0.953),respectively,and the clinical model,with AUC values of 0.834(95%CI:0.754-0.914)and 0.760(95%CI:0.564-0.956),respectively.However,the differences in AUC among the models in both the training and validation sets are not statistically significant(P values all>0.05).DCA indicated that the post-reconstruction radiomic model had the greatest net clinical benefit in the training set above a threshold of 0.34 and in the validation set between 0.36-0.59 and above 0.71.Conclusions:The application of MRI radiomics enhanced by 3D super-resolution reconstruction technology based on generative adversarial networks shows promise in predicting the efficacy of TACE combined with molecularly targeted therapy for unresectable HCC.
维普
万方数据