基于MRI影像组学机器学习模型预测干燥综合征唇腺活检病理结果的价值  

The value of predicting the pathological results of labial gland biopsy in Sjögren’s syndrome based on MRI radiomics machine learning models

作　　者：梁云平 瞿航 王苇[1] 顾越 周怡 赵义[1] LIANG Yunping;QU Hang;WANG Wei;GU Yue;ZHOU Yi;ZHAO Yi(Department of Medical Imaging,the Affiliated Hospital of Yangzhou University,Yangzhou,Jiangsu Province 225000,China)

机构地区：[1]扬州大学附属医院医学影像科,江苏扬州225000

出　　处：《实用放射学杂志》2024年第10期1592-1596,共5页Journal of Practical Radiology

摘　　要：目的探讨基于唇腺MRI影像组学机器学习模型预测干燥综合征(SS)唇腺活检病理结果的价值。方法回顾性分析178例可疑SS患者唇腺MRI资料,唇腺活检阳性97例、阴性81例,采用分层随机抽样,按4︰1的比例分为训练集(143例)和测试集(35例)。在T2WI水相下唇腺最大层面手动勾画感兴趣区(ROI)并提取影像组学特征(104个)。使用最小绝对收缩和选择算子(LASSO)进行特征筛选,使用筛选所得特征集构建Extra Trees、LightGBM和Gradient Boosting分类器模型。采用受试者工作特征(ROC)曲线评估模型的预测效能,并采用DeLong检验比较模型间曲线下面积(AUC)的差异。采用决策曲线分析(DCA)评价模型指导活检的临床应用价值。结果经LASSO筛选,获得5个最佳影像组学特征。Extra Trees、LightGBM和Gradient Boosting模型训练集和测试集AUC分别为1.000、0.807、0.960和0.655、0.779、0.639。DeLong检验显示3个模型测试集AUC差异无统计学意义。DCA显示LightGBM模型指导活检在更宽的风险阈值范围内具有更高的临床净收益,优于其他模型。结论基于唇腺T2WI水相影像组学特征LightGBM模型预测SS唇腺活检病理结果准确率较高,指导活检可以获得较高的临床收益,具有潜在临床应用价值。Objective To investigate the value of predicting the pathological results of labial gland biopsy in Sjögren’s syndrome(SS)based on the labial gland MRI radiomics machine learning models.Methods The labial gland MRI data of 178 suspected SS patients were analyzed retrospectively,and the labial gland biopsy pathology results were positive in 97 cases and negative in 81 cases.The samples were divided into training set(143 cases)and test set(35 cases)using a randomized stratified sampling according to the ratio of 4︰1.The region of interest(ROI)was manually outlined at the maximal level of the lower labial gland in T2WI water phase and radiomics features(104)were extracted.Feature screening was performed using the least absolute shrinkage and selection operator(LASSO),and the selected features set was used to construct Extra Trees,LightGBM,and Gradient Boosting classifier models.The predictive efficacy of the models was evaluated using the receiver operating characteristic(ROC)curve,and the DeLong test was used to compare the differences in the area under the curve(AUC)between the models.Decision curve analysis(DCA)was used to evaluate the clinical application value of the models in guiding biopsy.Results After LASSO screening,five optimal radiomics features were obtained.The AUC of Extra Trees,LightGBM,and Gradient Boosting models on the training and test sets were as follows 1.000,0.807,0.960 and 0.655,0.779,0.639,respectively.The DeLong test showed no statistically significant difference in AUC among the three models in the test set.DCA showed that the LightGBM model of guided biopsy had a higher clinical net benefit over a wider range of risk thresholds than other models.Conclusion Based on the radiomics features of the labial gland T2WI water phase,the LightGBM model has a high accuracy in predicting the pathological results of labial gland biopsy in SS,and guiding biopsy can obtain high clinical benefits,which has potential clinical application value.

关 键 词：干燥综合征 唇腺 磁共振成像 影像组学 

分 类 号：R593[医药卫生—内科学] R445.2[医药卫生—临床医学] R445