张正杰,程云章,王庆国,张娴,张鹏善,黄陈.增强动脉期CT影像组学特征联合临床术前预测胃癌脉管浸润[J].中国医学影像技术,2024,40(4):553~558
增强动脉期CT影像组学特征联合临床术前预测胃癌脉管浸润
Enhanced arterial phase CT radiomics features combined with clinic for preoperative predicting gastric cancer lymphovascular invasion
投稿时间:2023-08-06  修订日期:2024-01-29
DOI:10.13929/j.issn.1003-3289.2024.04.016
中文关键词:  胃肿瘤  肿瘤转移  体层摄影术,X线计算机  影像组学
英文关键词:stomach neoplasms  neoplasm metastasis  tomography, X-ray computed  radiomics
基金项目:国家自然科学基金(82203751)。
作者单位E-mail
张正杰 上海交通大学医学院附属第一人民医院胃肠外科, 上海 200080
上海理工大学上海介入医疗器械工程技术研究中心, 上海 200093 
 
程云章 上海理工大学上海介入医疗器械工程技术研究中心, 上海 200093  
王庆国 上海交通大学医学院附属第一人民医院放射科, 上海 200080  
张娴 上海交通大学医学院附属第一人民医院胃肠外科, 上海 200080
上海理工大学上海介入医疗器械工程技术研究中心, 上海 200093 
 
张鹏善 上海交通大学医学院附属第一人民医院胃肠外科, 上海 200080  
黄陈 上海交通大学医学院附属第一人民医院胃肠外科, 上海 200080 richard-hc@hotmail.com 
摘要点击次数: 331
全文下载次数: 283
中文摘要:
      目的 观察增强动脉期CT影像组学特征联合临床术前预测胃癌脉管浸润(LVI)的价值。方法 回顾性纳入298例胃癌患者,根据是否伴LVI将其分为阳性组(n=155)及阴性组(n=143),并按7 ∶ 3比例分为训练集(n=208)及测试集(n=90)。基于增强动脉期CT图提取病灶影像组学特征,采用logistic回归分析筛选胃癌LVI的临床影响因素;分别采用支持向量机(SVM)、逻辑回归(LR)、随机森林(RF)及极端梯度提升树(XGBoost)建立影像组学模型、临床模型及临床-影像组学模型,评估各模型预测胃癌LVI的效能。结果 以SVM、LR、RF及XGBoost建立的影像组学模型预测训练集胃癌LVI的曲线下面积(AUC)分别为0.896、0.821、1.000及1.000,其在测试集的AUC分别为0.744、0.801、0.740及0.747。基于4种机器学习建立的临床模型在训练集的AUC均为0.810,在测试集均为0.840。基于SVM、LR、RF及XGBoost建立的临床-影像组学模型预测训练集胃癌LVI的AUC分别为0.920、0.900、1.000及1.000,其在测试集的AUC分别为0.900、0.890、0.840及0.790。测试集中,基于SVM、LR及RF的临床-影像组学模型的AUC均大于影像组学模型和临床模型(P均<0.05)。结论 增强动脉期CT影像组学联合临床有助于术前预测胃癌LVI。
英文摘要:
      Objective To observe the value of enhanced arterial phase CT radiomics features combined with clinic for preoperative predicting gastric cancer lymphovascular invasion (LVI). Methods Data of 298 patients with gastric cancer were retrospectively analyzed. The patients were divided into positive group (n=155) and negative group (n=143) based on with LVI or not, also into training set (n=208) and test set (n=90) at a ratio of 7 ∶ 3. The radiomics features of tumor ROI were extracted based on enhanced arterial phase CT images, and clinical impact factors of gastric cancer LVI were screened with logistic regression analysis. Then radiomics model, clinical model and clinical-radiomics model were established using support vector machine (SVM), logistic regression (LR), random forest (RF) and extreme gradient boosting (XGBoost), respectively, and the efficacy of each model for predicting gastric cancer LVI was evaluated. Results The area under the curve (AUC) of the radiomics model based on SVM, LR, RF and XGBoost for predicting gastric cancer LVI in training set was 0.896, 0.821, 1.000 and 1.000, respectively, while was 0.744, 0.801, 0.740 and 0.747 in test set, respectively. The AUC of clinical model based on SVM, LR, RF and XGBoost were all 0.810 in training set, while were all 0.840 in test set. The AUC of clinical-radiomics model based on SVM, LR, RF and XGBoost in training set was 0.920, 0.900, 1.000 and 1.000, respectively, in test set was 0.900, 0.890, 0.840 and 0.790, respectively. In test set, the AUC of clinical-radiomics model based on SVM, LR and RF were all higher than that of radiomics model and clinical model (all P<0.05). Conclusion Enhanced arterial phase CT radiomics features combined with clinic was helpful for predicting gastric cancer LVI before surgical operation.
查看全文  查看/发表评论  下载PDF阅读器