| 尚宏霖,詹钰琪,莫少颖,范誉铧,杨云竣,赵海,王伟.基于MR T2WI及弥散加权成像影像组学构建机器学习模型预测直肠癌周围神经侵犯[J].中国医学影像技术,2025,41(4):616~621 |
| 基于MR T2WI及弥散加权成像影像组学构建机器学习模型预测直肠癌周围神经侵犯 |
| Machine learning model based on MR T2WI and diffusion-weighted imaging radiomics for predicting perineural invasion of rectal cancer |
| 投稿时间：2024-11-01 修订日期：2025-01-30 |
| DOI：10.13929/j.issn.1003-3289.2025.04.023 |
| 中文关键词: 直肠肿瘤 肿瘤侵袭性 磁共振成像 影像组学 机器学习 |
| 英文关键词:rectal neoplasms neoplasm invasiveness magnetic resonance imaging radiomics machine learning |
| 基金项目: |
|
| 摘要点击次数: 641 |
| 全文下载次数: 113 |
| 中文摘要: |
| 目的 观察基于MR T2WI及弥散加权成像(DWI)影像组学构建的机器学习(ML)模型预测直肠癌周围神经侵犯(PNI)的价值。方法 回顾性收集343例直肠癌,按8 ∶ 2比例划分训练集 与测试集 。以单、多因素逻辑回归(LR)分析临床相关资料,筛选直肠癌PNI的独立预测因素并构建临床模型。基于术前T2WI及DWI提取及筛选最佳影像组学特征,分别以极度随机树、多层感知器、轻量梯度提升机、极限梯度提升、支持向量机(SVM)、LR、K-邻近法及随机森林算法构建ML模型,筛选其中最优者,联合临床相关因素构建临床-影像组学ML模型;评估各模型预测效能及临床价值。结果 患者年龄为直肠癌PNI的独立预测因素(OR=0.988,P＜0.001),以之构建的临床模型在训练集和测试集的曲线下面积(AUC)分别为0.435及0.458。8种ML模型SVM模型最优,其在训练集和测试集的AUC分别为 0.887及0.854;临床-影像组学ML模型在训练集和测试集的AUC分别为0.887及0.860,与SVM模型差异均无统计学意义(P均＞0.05)。决策曲线分析显示,阈值为0.20~0.45时,SVM模型的临床净收益高于其他模型。结论 基于T2WI及DWI影像组学构建的SVM模型可有效预测直肠癌PNI。 |
| 英文摘要: |
| Objective To observe the value of machine learning model based on MR T2WI and diffusion weighted imaging (DWI) radiomics for predicting perineural invasion (PNI) of rectal cancer. Methods Totally 343 patients with rectal cancer were retrospectively collected and divided into training set (n=275, 92 PNI and 183 PNI ) and test set (n=68, 23 PNI and 45 PNI ) at the ratio of 8 ∶ 2. Univariate and multivariate logistic regression (LR) were used to analyze clinical data and screen the independent predictors of PNI in rectal cancer, so as to construct a clinical model. The best radiomics features were extracted and screened based on preoperative T2WI and DWI. Then extremely randomized trees, multilayer perceptron, light gradient boosting machine, extreme gradient boosting, support vector machine (SVM), LR, K-nearest neighbor and random forest algorithms were used to construct ML models, respectively, and the optimal ML model was selected to establish a clinical-radiomics ML model combined with clinical relevant independent predictors. The predictive efficacy and clinical value of each model were evaluated. Results Patients’ age was the independent predictor of PNI of rectal cancer (OR=0.988, P＜0.001), and the area under the curve (AUC) of the clinical model constructed based on it was 0.435 and 0.458 in training and test sets, respectively. SVM model was the best one among 8 ML models, with AUC in training and test set of 0.887 and 0.854, respectively. The AUC of clinical-radiomics ML model in training and test sets was 0.887 and 0.860, respectively, not different with AUC of SVM model (both P＞0.05). Decision curve analysis showed that when the threshold value was 0.20—0.45, clinical net benefit of SVM model was higher than that of other models. Conclusion SVM model based on T2WI and DWI radiomics could effectively predict PNI of rectal cancer. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
