摘要
目的:探讨基于T_(1)WI六种不同机器学习分类算法模型对新生儿急性胆红素脑病(ABE)的诊断价值。方法:收集36例ABE患儿(ABE组)和49例非ABE新生儿(非ABE组)的临床和MR资料,基于Radcloud平台对T_(1)WI苍白球进行标注、影像组学特征提取及降维,使用5折交叉验证方法,采用支持向量机、决策树、线性判别分析、随机森林、逻辑回归(LR)、多层感知器(MLP)等6种不同机器学习分类算法进行预测模型构建及验证,采用受试者操作特征(ROC)曲线下面积(AUC)评估不同机器学习模型预测ABE的诊断效能,并与传统视觉阅片、基于苍白球(G)与壳核(P)信号比值(G/P值)半定量评估方法比较。结果:基于T_(1)WI图像构建的机器学习模型可用于预测ABE,以MLP、LR模型为最优并具有一定的诊断效能AUC分别达0.844、0.835,高于G/P值半定量评估(AUC为0.807)及传统视觉诊断(AUC为0.755)。结论:基于T_(1)WI放射组学机器学习模型能够更好的有效识别ABE,有望为临床早期识别ABE提供更可靠参考价值。
Objective:To explore the diagnostic value of six different machine learning classification algorithm models based on T_(1)-weighted imaging(T_(1)WI)for neonatal acute bilirubin encephalopathy(ABE).Methods:Clinical and MRI data from 36 ABE neonates(ABE group)and 49 non-ABE neonates(non-ABE group)were collected.T_(1)WI images of the globus pallidus were annotated sing the Radcloud platform,followed by radiomic feature extraction and dimensionality reduction.Six machine learning algorithms-support vector machine(SVM),decision tree(DT),linear discriminant analysis(LDA),random forest(RF),logistic regression(LR),and multilayer perceptron(MLP)-were employed to construct and validate prediction models using 5-fold cross-validation.Receiver operating characteristic(ROC)curves were plotted to evaluate the diagnostic performance of these models,which was then compared with the traditional visual diagnosis and semi-quantitative evaluation based on the globus pallidus(G)/putamen(P)signal intensity ratio(G/P values).Results:The T_(1)WI-based machine learning models demonstrated potential for predicting ABE,with the MLP and LR models showing optimal diagnostic performance(AUC=0.844 and 0.835,respectively),outperforming the G/P values method(AUC=0.807)and traditional visual diagnosis(AUC=0.755).Conclusion:T_(1)WI-based radiomic machine learning models can effectively identify ABE and may provide a more reliable reference for early clinical detection of ABE.
作者
林春琴
林钱森
陈杰云
王建新
杨彦
黄海云
LIN Chun-qin;LIN Qian-sen;CHEN Jie-yun(Department of Pediatrics,Quanzhou First Hospital Affiliated to Fujian Medical University,Fujian 362000,China)
出处
《放射学实践》
北大核心
2026年第3期295-300,共6页
Radiologic Practice
基金
福建省自然科学基金面上项目(2022J011463)
泉州市科技计划项目(2024NY002,2023NS054)。
关键词
急性胆红素脑病
磁共振成像
机器学习
Acute bilirubin encephalopathy
Magnetic resonance imaging
Machine learning
