摘要
目的基于超分辨显微成像(URM)微血管参数构建随机森林模型,探讨其鉴别淋巴瘤与淋巴结反应性增生的临床价值。方法前瞻性纳入经病理确诊的浅表淋巴结病变患者33例,其中淋巴瘤组17例(68个淋巴结病变),淋巴结反应性增生组16例(63个淋巴结病变)。应用URM定量分析微血管密度、弯曲度、流速最大值、平均值及标准差,以及微血管密度占比、复杂度、灌注指数,比较两组上述参数的差异。绘制受试者工作特征(ROC)曲线分析各URM微血管参数鉴别淋巴瘤与淋巴结反应性增生的诊断效能。将纳入淋巴结病变按7∶3比例随机分为训练集和测试集,基于URM参数采用随机森林算法构建鉴别淋巴瘤与淋巴结反应性增生的机器学习模型,使用混淆矩阵和ROC曲线评估模型性能,对各参数进行重要性排序并筛选关键特征参数。结果淋巴瘤组微血管弯曲度最大值、平均值、标准差和微血管复杂度均高于淋巴结反应性增生组,微血管流速标准差、最大值均低于淋巴结反应性增生组,差异均有统计学意义(均P<0.05)。ROC曲线分析显示,微血管弯曲度最大值鉴别淋巴瘤与淋巴结反应性增生的曲线下面积(AUC)为0.822,高于其他URM微血管参数。基于URM微血管参数构建鉴别淋巴瘤与淋巴结反应性增生的随机森林模型,该模型在测试集的准确度、召回率、特异度及F1分数分别为84.62%、91.67%、84.62%、0.88,AUC为0.853。模型各参数重要性排序显示,微血管弯曲度最大值和微血管复杂度的贡献度较高;模型筛选出鉴别淋巴瘤与淋巴结反应性增生的关键特征参数为微血管弯曲度最大值和微血管复杂度。结论URM通过量化微血管形态学及血流动力学参数,可为鉴别淋巴瘤与淋巴结反应性增生提供参考,基于此构建的随机森林模型可进一步提高鉴别诊断效能。
Objective To construct a random forest model based on ultra-resolution microscopy imaging(URM)microvascular parameters,and to explore its clinical value in differentiating lymphoma from reactive lymphoid hyperplasia.Methods A total of 33 patients with superficial lymph node lesions confirmed by pathology were prospectively enrolled,including 17 cases in the lymphoma group(68 lymph node lesions)and 16 cases in the reactive lymphoid hyperplasia group(63 lymph node lesions).URM was used to quantitatively analyze the microvascular density,tortuosity,flow velocity(maximum,mean,and standard deviation),microvascular density ratio,complexity,and perfusion index,the differences in above parameters between the two groups were compared.Receiver operating characteristic(ROC)curves were drawn to analyze the diagnostic efficacy of URM microvascular parameters in differentiating lymphoma form reactive lymphoid hyperplasia.Lymph node lesions were randomly divided into training and testing sets(7∶3 ratio).A machine learning model for differentiating lymphoma from reactive lymphoid hyperplasia was constructed based on URM microvascular parameters with a random forest algorithm,and model performance was evaluated by a confusion matrix and the ROC curves.Key feature parameters were screened based on importance ranking.Results The maximum,mean,and standard deviation of microvascular tortuosity and microvascular complexity in the lymphoma group were significantly higher than those in the reactive lymphoid hyperplasia group,while the standard deviation and maximum of microvascular flow velocity were significantly lower(all P<0.05).ROC curve analysis showed that the maximum microvascular tortuosity achieved an area under the curve(AUC)of 0.822 for differentiating lymphoma from reactive lymphoid hyperplasia,which was higher than other URM microvascular parameters.A random forest model for differentiating lymphoma from reactive lymph node hyperplasia based on URM microvascular parameters was constructed.The accuracy,recall rate,specificity,and F1 score of this model in the test set were 84.62%,91.67%,84.62%,and 0.88,respectively,and the AUC was 0.853.Based on importance ranking,maximum microvascular tortuosity and microvascular complexity demonstrated the highest contributions.The model selected maximum microvascular tortuosity and microvascular complexity as the key feature parameters for differentiating lymphoma from reactive lymphoid hyperplasia.Conclusion URM provides reference for differentiating lymphoma from reactive lymphoid hyperplasia by quantifying morphological and hemodynamic parameters of microvessels.The random forest model constructed based on URM microvascular parameters can further enhance the diagnostic performance.
作者
蒋亨
杨君
华兴
刘政
李泞珊
JIANG Heng;YANG Jun;HUA Xing;LIU Zheng;LI Ningshan(Department of Ultrasound,the Second Affiliated Hospital of Army Medical University,PLA,Chongqing 400037,China)
出处
《临床超声医学杂志》
2025年第10期793-798,共6页
Journal of Clinical Ultrasound in Medicine
基金
国家自然科学基金青年科学基金项目(82102077)
重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0056)
陆军军医大学第二附属医院青年博士人才孵化项目(2023YQB039)。
关键词
超分辨显微成像
微血管参数
淋巴瘤
淋巴结反应性增生
随机森林模型
Ultra-resolution microscopy imaging
Microvascular parameters
Lymphoma
Reactive lymphoid hyperplasia
Random forest model
