影像组学结合深度学习预测异柠檬酸脱氢酶野生型弥漫性星形细胞瘤的端粒酶逆转录酶启动子状态  

Radiomics Combined with Deep Learning in Predicting Telomerase Reverse Transcriptase Promoter Status in Isocitrate Dehydrogenase-Wildtype Diffuse Astrocytoma

作　　者：徐欣怡 张旺 张力强[1] 王琳玲 文明[1] XU Xinyi;ZHANG Wang;ZHANG Liqiang;WANG Linling;WEN Ming(Department of Radiology,the First Affiliated Hospital of Chongqing Medical University,Chongqing 400016,China;不详)

机构地区：[1]重庆医科大学附属第一医院放射科,重庆400016 [2]西南大学计算机与信息科学学院,重庆400715

出　　处：《中国医学影像学杂志》2024年第11期1097-1104,共8页Chinese Journal of Medical Imaging

摘　　要：目的 探讨基于MRI的影像组学和深度学习建立融合模型预测异柠檬酸脱氢酶野生型弥漫性星形细胞瘤的端粒酶逆转录酶启动子(TERTp)突变状态。资料与方法 回顾性分析2019年1月—2021年6月重庆医科大学附属第一医院癌症基因组图谱和癌症影像档案的175例异柠檬酸脱氢酶野生型弥漫性星形细胞瘤患者(训练组122例,测试组53例),评估其TERTp突变状态。在T1c和T2f图像上勾画水肿区和肿瘤区,使用SE-Net模型构建深度学习模型,提取不同区域(水肿区、肿瘤区和整体病变)的组学特征,并通过最小绝对收缩和选择算法筛选11个特征建立组学模型。最后,将影像组学模型、深度学习模型和包含伦勃朗视觉感受图像特征的临床模型结合为融合模型,使用校准曲线和决策曲线评估模型。结果 最终建立6个预测模型,临床模型训练组和测试组曲线下面积(AUC)分别为0.815(95%CI 0.738~0.892)和0.645(95%CI 0.494~0.796);深度学习模型训练组和测试组AUC分别为0.860(95%CI 0.798~0.922)和0.735(95%CI 0.614~0.856);融合组学模型比单独水肿或肿瘤区组学模型预测性能更优,训练组和测试组AUC分别为0.906(95%CI 0.856-0.955)和0.867(95%CI 0.769~0.964);融合模型预测性能最佳,训练组和测试组AUC分别为0.964(95%CI 0.929~1.000)和0.905(95%CI 0.818~0.991)。结论 影像组学结合深度学习的临床融合模型在预测异柠檬酸脱氢酶野生型弥漫性星形细胞瘤的TERTp突变状态方面表现良好。Purpose To investigate the fusion model based on MRI radiomics and deep learning to predict the telomerase reverse transcriptase promoter(TERTp)mutation status in isocitrate dehydrogenase-wildtype diffuse astrocytoma.Materials and Methods A retrospective analysis of 175 patients with isocitrate dehydrogenase-wildtype diffuse astrocytoma(122 in the training group and 53 in the test group)from January 2019 to June 2021 in the First Affiliated Hospital of Chongqing Medical University.The Cancer Genome Atlas and The Cancer Imaging Archive were performed to assess TERTp mutation status.The edema and tumor regions were outlined on T1c and T2f images,deep learning model were constructed using the SE-Net model,radiomics features of different regions(edema region,tumor region and overall lesion)were extracted,and 11 features were screened by the least absolute shrinkage and selection operator to build radiomics model.Finally,the radiomics model,deep learning model and clinical model containing Visually Accessible Rembrandt Images features were combined as fusion model,and the model was evaluated using calibration curves and decision curves.Results Six predictive models were eventually built,with an area under curve(AUC)of 0.815(95%CI 0.738-0.892)and 0.645(95%CI 0.494-0.796)for the training and test groups of the clinical model;the AUC for the training and test groups of the deep learning model was 0.860(95%CI 0.798-0.922)and 0.735(95%CI 0.614-0.856);the fusion radiomics model had better predictive performance than the edema or tumor region radiomics models alone,with AUC of 0.906(95%CI 0.856-0.955)and 0.867(95%CI 0.769-0.964)in the training and test groups;the fusion model showed the best performance,with AUC of 0.964(95%CI 0.929-1.000)and 0.905(95%CI 0.818-0.991)in the training and test groups.Conclusion The clinical fusion model of radiomics combined with deep learning performed well in predicting TERTp mutation status in isocitrate dehydrogenase-wildtype diffuse astrocytoma.

关 键 词：星形细胞瘤 磁共振成像 影像组学 深度学习 端粒酶逆转录酶启动子 异柠檬酸脱氢酶 

分 类 号：R739.41[医药卫生—肿瘤] R445.2[医药卫生—临床医学]