| PET/CT影像组学联合机器学习鉴别诊断原发性中枢神经系统淋巴瘤与脑转移瘤的价值 |
| 投稿时间：2025-02-07 修订日期：2025-02-21 点此下载全文 |
| 引用本文：付静瑜,张育溪,杨帆,彭岱云,柳江燕.PET/CT影像组学联合机器学习鉴别诊断原发性中枢神经系统淋巴瘤与脑转移瘤的价值[J].医学研究杂志,2025,54(7):96-102 |
| DOI：
10.11969/j.issn.1673-548X.2025.07.018 |
| 摘要点击次数: 3 |
| 全文下载次数: 4 |
|
| 基金项目:甘肃省科技厅联合基金资助项目(24JRRA922)；兰州大学第二医院萃英科技创新计划应用基础研究(CY2023-MS-B10) |
|
| 中文摘要:目的 本研究旨在探讨PET/CT影像组学联合机器学习鉴别诊断原发性中枢神经系统淋巴瘤(primary central nervous system lymphoma,PCNSL)与脑转移瘤(brain metastasis,BM)的价值。方法 选取2019年1月～2024年11月就诊于兰州大学第二医院核医学科的69例患者,共127个病灶(包括43个PCNSL病灶和84个BM病灶),按7∶3的比例分为训练集(n=88)和验证集(n=39)。使用3D slicer提取PET和CT的影像组学特征,采用最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)回归联合5折交叉验证进行特征降维,构建3种经典机器学习模型——逻辑回归(Logistic regression,LR)、决策树(decision tree,DT)和支持向量机(support vector machine,SVM),使用受试者工作特征(receiver operating characteristic,ROC)曲线和混淆矩阵评估各模型的诊断效能。结果 基于PET影像组学的SVM模型在验证集中表现最佳[曲线下面积(area under the curve,AUC)=0.917,敏感度为84.6%,特异性为92.3%)],优于CT模型(AUC=0.787,敏感度为73.1%,特异性为76.9%)。结论 基于PET影像组学代谢特征构建的支持向量机模型在PCNSL与BM的术前鉴别中展现出较高诊断价值,其作为无创辅助诊断方法可临床推广应用。 |
| 中文关键词:PET/CT 影像组学 原发性中枢神经系统淋巴瘤 脑转移瘤 |
| |
| Value of PET/CT Radiomics Combined with Machine Learning in Differentiating Primary Central Nervous System Lymphoma from Brain Metastases. |
|
|
| Abstract:Objective This study aimed to investigate the value of PET/CT radiomics combined with machine learning in differentiating primary central nervous system lymphoma (PCNSL) from brain metastases (BM). Methods Sixty-nine patients with 127 lesions (including 43 PCNSL lesions and 84 BM lesions) who attended the Department of Nuclear Medicine of Lanzhou University Second Hospital from January 2019 to November 2024 were selected and divided into a training set (n=88) and a validation set (n=39) in a 7∶3 ratio. Radiomics features of PET and CT were extracted using the 3D slicer, and feature dimensionality reduction was performed using least absolute shrinkage and selection operator(LASSO) regression combined with 5-fold cross-validation to construct three classical machine learning models-Logistic regression (LR), decision tree (DT), and support vector machine (SVM), and the diagnostic efficacy of each model was evaluated using receiver operating characteristic(ROC) curves and confusion matrices. Results The SVM model based on PET imaging performed best in the validation set [area under the curve(AUC)=0.917, sensitivity=84.6%, specificity=92.3%] and was significantly better than the CT model (AUC=0.787, sensitivity=73.1%, specificity=76.9%). Conclusion The support vector machine model constructed based on PET radiomics metabolic features demonstrated high diagnostic value in the preoperative differentiation between PCNSL and BM, showing potential for clinical application as a non-invasive auxiliary diagnostic tool. |
| keywords:PET/CT Radiomic Primary central nervous system lymphoma Brain metastases |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
