机器学习分类器预测Gd-EOB-DTPA增强MRI肝胆期肝脏强化不足
该回顾性研究纳入214例慢性肝病或肝硬化患者,利用机器学习分类器结合特征选择算法预测Gd-EOB-DTPA增强MRI肝胆期肝脏强化不足。结果显示,使用总胆红素和白蛋白的SVM模型预测性能优异(AUC=0.93),高于传统逻辑回归模型(AUC=0.92)及基于MELD-Na(AUC=0.90)和Child-Pugh评分(AUC=0.89)的模型。机器学习方法可在MRI检查前预测肝胆期强化不足。
要点
- 总胆红素和白蛋白是预测肝胆期强化不足的关键特征。
- SVM模型预测性能最佳(AUC=0.93),优于传统评分系统。
- 可在MRI检查前识别可能强化不足的患者,辅助优化扫描方案。
核心结论
- 基于机器学习的分类器(如SVM)和特征选择算法可在MRI检查前预测肝胆期肝脏强化不足。
研究设计与样本
- 回顾性研究纳入214例慢性肝病或肝硬化患者,均接受Gd-EOB-DTPA增强MRI。
- 收集多种肝功能检查、Child-Pugh评分和MELD-Na评分作为候选预测因子。
- 采用随机欠采样提升算法评估特征重要性,并构建决策树、K近邻、随机森林和SVM分类器模型。
- 通过计算受试者工作特征曲线下面积(AUC)分析预测模型性能。
关键影像表现/征象
- 肝胆期强化不足通过肝-门静脉信号强度比和5级视觉分级评估。
诊断标准与关键数值
- SVM使用总胆红素和白蛋白预测肝胆期强化不足的AUC为0.93(95% CI: 0.93-0.94)。
- 传统逻辑回归模型AUC为0.92(95% CI: 0.92-0.93)。
- 基于MELD-Na的模型AUC为0.90(95% CI: 0.89-0.91),基于Child-Pugh评分的模型AUC为0.89(95% CI: 0.88-0.90)。
临床应用与阅片要点
- 对于慢性肝病或肝硬化患者,可利用总胆红素和白蛋白构建的机器学习模型在MRI检查前预测肝胆期强化不足风险。
- 预测强化不足高风险患者可考虑调整扫描方案或强化方式,以避免因强化不足影响诊断。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
PURPOSE: The purpose of this study was to reveal the usefulness of machine learning classifier and feature selection algorithms for prediction of insufficient hepatic enhancement in the HBP.
METHODS: We retrospectively assessed 214 patients with chronic liver disease or liver cirrhosis who underwent MRI enhanced with Gd-EOB-DTPA. Various liver function tests, Child-Pugh score (CPS) and Model for End-stage Liver Disease Sodium (MELD-Na) score were collected as candidate predictors for insufficient hepatic enhancement. Insufficient hepatic enhancement was assessed using liver-to-portal vein signal intensity ratio and 5-level visual grading. The clinico-laboratory findings were compared using Student's t-test and Mann-Whitney U test. Relationships between the laboratory tests and insufficient hepatic enhancement were assessed using Pearson's and Spearman's rank correlation coefficient. Feature importance was assessed by Random UnderSampling boosting algorithms. The predictive models were constructed using decision tree(DT), k-nearest neighbor(KNN), random forest(RF), and support-vector machine(SVM) classifier algorithms. The performances of the prediction models were analyzed by calculating the area under the receiver operating characteristic curve(AUC).
RESULTS: Among four machine learning classifier algorithms using various feature combinations, SVM using total bilirubin(TB) and albumin(Alb) showed excellent predictive ability for insufficient hepatic enhancement(AUC = 0.93, [95% CI: 0.93-0.94]) and higher AUC value than conventional logistic regression(LR) model (AUC = 0.92, [95% CI; 0.92-0.93], predictive models using the MELD-Na (AUC = 0.90 [95% CI: 0.89-0.91]) and CPS (AUC = 0.89 [95% CI: 0.88-0.90]).
CONCLUSION: Machine learning-based classifier (i.e. SVM) and feature selection algorithms can be used to predict insufficient hepatic enhancement in the HBP before performing MRI.
引用
- Ko JS, Byun J, Park S, et al. Prediction of insufficient hepatic enhancement during the Hepatobiliary phase of Gd-EOB DTPA-enhanced MRI using machine learning classifier and feature selection algorithms. Abdom Radiol (NY). 2021. PMID: 34647145. doi:10.1007/s00261-021-03308-0
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。