【摘要】 目的利用影像组学与集成学习进行肝脏脂肪变性分级研究。方法回顾性分析2018年6月至8月于上海市第一人民医院进行MR上腹部mDixon成像序列扫描的成人患者资料,将患者的MRI数据利用影像组学特征提取方法和机器学习进行建模,研究采用3项指标对三种集成学习分类算法(AdaBoost、GBDT与XGBoost)的性能进行评估,包括准确率、精确率、召回率。结果 XGBoost算法性能最佳,分类准确率达到81.9%;五项特征重要性之和大于19%,即在总体肝脏脂肪变形程度轻中度分类模型之中所占权重接近1/5。结论影像组学与集成学习方法为脂肪变性分级提供了一种较为可靠的辅助诊断手段,对轻中度脂肪变性的研究也能够为患者脂质代谢相关疾病的临床干预或治疗时机提供一定的参考价值。更多还原

【Abstract】 Objective To study the classification of hepatic steatosis using radiomics and ensemble learning. Methods A retrospective study was conducted on the datum of adult patients who underwent abdomen MRI with mDixon sequence scanning from June 2018 to August 2018 in Shanghai General Hospital. The MRI data of patients were modeled by using the method of radiomics feature extraction and machine learning. Three indexes were used to evaluate the performance of three ensemble learning classification algorithms(AdaBoost, GBDT and XGBoost), including the rate of accuracy and recall. Results The XGBoost algorithm had the best performance, the classification accuracy was 81.9%, and the sum of the importance of the five features was greater than 19%, which meant that the weight of the total liver fat deformation degree classification model was close to 1/5. Conclusion The method of combining radiomics and ensemble learning provides a more reliable auxiliary diagnostic means for the classification of steatosis. The study of mild to moderate steatosis can also provide a certain reference value for clinical intervention or treatment opportunity of lipid metabolism related diseases.更多还原