机构地区:[1]滨州医学院附属医院放射科,山东滨州256603 [2]滨州医学院医学影像学院,山东烟台264003
出 处:《实用放射学杂志》2023年第10期1606-1610,共5页Journal of Practical Radiology
基 金:山东省大学生创新创业训练计划基金项目(S202210440171);滨州医学院科研计划与科研启动基金项目(BY2022KJ09)。
摘 要:目的探讨机器学习对乳腺癌瘤内、瘤周多模态MRI影像组学特征预测Ki-67表达的诊断效能。方法回顾分析经病理证实的202例乳腺癌患者多模态MRI图像。应用3D Slicer软件手动勾画感兴趣区(ROI),提取影像组学特征。按7︰3随机分为训练组(n=141)和验证组(n=61),采用方差阈值法和最小绝对收缩和选择算子(LASSO)算法筛选影像组学特征;用逻辑回归(LR)、支持向量机(SVM)、决策树(DT)和随机森林(RF)构建模型。采用受试者工作特征(ROC)曲线评估不同模型预测效能。结果对基于单序列4种学习分类器模型的诊断效能比较,LR模型预测效能最佳;LR算法在基于多参数预测中,MRI对比增强(CE)+T_(2)序列、CE+表观扩散系数(ADC)序列、T_(2)WI+ADC序列、CE+T_(2)WI+ADC序列在验证组曲线下面积(AUC)分别为0.713、0.929、0.837、0.855,基于CE+ADC序列LR模型AUC高于其他模型,有统计学差异(P<0.05)(DeLong检验)。结论采用多模态MRI(CE、T_(2)WI和ADC)影像组学的联合,可以提高单一参数影像模型预测乳腺癌Ki-67表达,其中使用LR构建的CE+ADC模型诊断效能最优,有望成为临床预测Ki-67的一种无创手段。Objective To explore the diagnostic efficacy of machine learning in predicting Ki-67 expression of breast carcinoma by using intratumoral and peritumoral radiomics based on multi-modal MRI.Methods The multi-modal MRI images of 202 breast carcinoma patients confirmed by pathology were analyzed retrospectively.3D Slicer software was used to manually sketch the region of interest(ROI)and extract the radiomics features.All cases were randomly divided into training set(n=141)and testing set(n=61)according to the ratio of 7︰3.The variance threshold method and the least absolute shrinkage and selection operator(LASSO)algorithm were used to screen the radiomics characteristics.Four machine learning algorithms including logistic regression(LR),support vector machine(SVM),decision tree(DT),and random forest(RF)were used to construct the prediction model.The prediction efficiency of the different models was evaluated by drawing receiver operating characteristic(ROC)curve.Results Comparing the diagnostic effectiveness of four learning classifier models based on single sequence,LR model had the best predictive effectiveness;In the multi parameter prediction based on LR algorithm,the area under the curve(AUC)of MRI contrast enhanced(CE)+T_(2) sequence,CE+apparent diffusion coefficient(ADC)sequence,T_(2)WI+ADC sequence,CE+T_(2)WI+ADC sequence in the training set was 0.713,0.929,0.837,0.855,respectively,and the AUC of LR model based on CE+ADC sequence was significantly higher than that of other models,and the difference was statistically significant(P<0.05)(DeLong test).Conclusion The combination of multi-modal MRI(CE,T_(2)WI and ADC)radiomics can improve the performance of single parameter imaging model in predicting Ki-67 expression of breast carcinoma.Among them,the CE+ADC model constructed by LR has the best diagnostic efficiency,which is expected to become a non-invasive means for predicting Ki-67 in clinical practice.
关 键 词:乳腺癌 KI-67 磁共振成像 影像组学 机器学习
分 类 号:R737.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学] TP181[自动化与计算机技术—控制理论与控制工程]
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...