肿瘤预防与治疗2025，Vol.38Issue(9)：796-803,8.DOI:10.3969/j.issn.1674-0904.2025.09.006
基于磁共振成像的机器学习模型术前预测肿块型浸润性乳腺癌脉管侵犯
Machine Learning Model Based on Magnetic Resonance Imaging for Pre-operative Prediction of Vascular Invasion in Mass-Forming Invasive Breast Cancer
摘要
Abstract
Objective:To explore the value of machine learning models with different numbers of features based on high-resolution delayed contrast-enhanced MRI in preopera-tively predicting lymphovascular invasion(LVI)in patients with mass-type invasive breast cancer.Methods:The lesions of patients with invasive breast cancer were delineated using 3D Slicer software,and inter-observer consistency of the seg-mentation between two physicians was evaluated.Least absolute shrinkage and selection operator and recursive feature elimi-nation were employed to select three feature subsets comprising 8,16,and 24 features.Models were constructed using deci-sion tree(DT),support vector machine(SVM),and logistic regression algorithms based on these feature subsets.Predic-tive performance was evaluated using accuracy,Matthews correlation coefficient(MCC),F1-score,and AUC.Results:A total of 120 high-resolution delayed contrast-enhanced MRI images from patients with non-specific mass-type invasive breast cancer were collected and randomly divided into a training group(83 cases)and a validation group(37 cases)in a 6.9∶3.1 ratio.One hundred and seven features were extracted from the target regions,and 81 remained after consistency testing.Among the three models with different feature numbers,the SVM model based on 24 radiomic features demonstrated the best predictive performance in the validation set,with the highest AUC of 0.915(95％CI:0.658～0.957),accuracy of 0.838,F1 score of 0.769,and MCC of 0.697.The logistic regression model based on 8 features demonstrated relatively stable per-formance both in the training and validation sets,with an AUC of 0.790(95％CI:0.715～0.865)in the training set and an AUC of 0.762(95％CI:0.536～0.916)in the validation set.For the DT model,the AUC values showed no significant change with varying numbers of features in either the training or validation sets.The AUCs for DT models with 8,16,and 24 features were 0.927(95％CI:0.884～0.970),0.919(95％CI:0.871～0.965),and 0.919(95％CI:0.871～0.965)in the training set,and 0.785(95％CI:0.641～0.929),0.783(95％CI:0.638～0.928),and 0.783(95％CI:0.638～0.928)in the validation set,respectively.Conclusion:The SVM model based on high-resolution delayed contrast-en-hanced MRI shows promise for preoperative prediction of LVI status in breast cancer and may provide valuable guidance for clinical individualized treatment.关键词
磁共振/机器学习/乳腺癌/脉管侵犯Key words
Magnetic resonance imaging/Machine learning/Breast cancer/Lymphovascular invasion分类
医药卫生引用本文复制引用
王子娟,罗红兵,胡云涛..基于磁共振成像的机器学习模型术前预测肿块型浸润性乳腺癌脉管侵犯[J].肿瘤预防与治疗,2025,38(9):796-803,8.基金项目
北京医学奖励基金会(编号:YXJL-2023-0227-0066) This study was supported by grants from Beijing Medical Award Foundation(No.YXJL-2023-0227-0066). （编号:YXJL-2023-0227-0066）
