基于机器学习的高强度聚焦超声消融子宫肌瘤剂量预测模型研究  

Study on Dose Prediction Model for High-Intersity Focused Ultrasound(HIFU)Ablation of Uterine Fibroids on Machine Learning

作　　者：胡文昊 文银刚[2] 徐凡[3] 唐英[3] 杜永洪[1] Hu Wenhao;Wen Yin gang;Xu Fan;Tang Ying;Du Yonghong(State Key Laboratory of Ultrasound in Medicine and Engineering,College of Biomedical Engineering,Chongqing Medical University,Chongqing Key Laboratory of Biomedical Engineering,Chongqing 400016,China;Ultrasound Medical National Engineering Research Center,Chongqing 401121,China;The Second Affiliated Medical College of North Sichuan Medical College,The Department of Gynecology of Nanchong Central Hospital,Nanchong,Sichuan 637000,China)

机构地区：[1]超声医学工程国家重点实验室,重庆医科大学生物医学工程学院、重庆市生物医学工程学重点实验室,重庆市400016 [2]超声医疗国家工程研究中心,重庆市401121 [3]川北医学院第二临床医学院,南充市中心医院妇科,四川省南充市637000

出　　处：《中国超声医学杂志》2023年第11期1280-1283,共4页Chinese Journal of Ultrasound in Medicine

基　　金：重庆市技术创新与应用发展专项面上项目(No.cstc2019jscx-msxmX0255)。

摘　　要：目的构建基于机器学习的高强度聚焦超声(HIFU)消融子宫肌瘤剂量预测模型。方法收集接受HIFU消融子宫肌瘤患者678例的临床资料,通过Spearman相关性分析和随机森林计算重要性得分相结合的方法筛选特征,将特征纳入多层感知器(MLP)和极限梯度提升(XGBoost)模型中建立消融剂量预测模型,并评估预测模型的性能。结果结合Spearman相关性分析和随机森林的结果,筛选出子宫位置、肌瘤类型、肌瘤位置、肌瘤长径等13类变量。分别建立MLP和XGBoost的剂量预测模型,其训练集上的平均绝对误差(MAE)、均方根误差(RMSE)、决定系数(R2)分别为0.163、0.195、0.951和0.029、0.088、0.993,MLP和XGBoost模型在测试集上的MAE、RMSE、R2分别为0.158、0.191、0.925和0.032、0.091、0.985。结论MLP和XGBoost模型对HIFU消融子宫肌瘤的剂量均有较好的预测能力,能辅助临床医师为患者制定更合理的子宫肌瘤治疗方案。Objective To construct a dose prediction model for high-intensity focused ultrasound(HIFU)ablation of uterine fibroids based on machine learning.Methods Clinical data of 678 patients with uterine fibroid underwent HIFU ablation were collected,Spearman correlation analysis and random forest importance score were used to select features,and the features were incorporated into Multilayer Perceptron(MLP)and Extreme Gradient Boosting(XGBoost)models to establish a ablation dose prediction model,and the performance of the prediction model was evaluated.Results Combined with Spearman correlation analysis with Random forest results,13 variables such as uterus position,fibroids type,fibroids location and fibroids long diameter were screened.The dose prediction models of MLP and XBGoost were established respectively.The Mean Absolute Error(MAE),Root Mean Square Error(RMSE),Best Coefficient of Determination(R°)of MLP and XGboost Models on the training Set were O.163,0.195,0.951 and 0.029,0.088,0.993,respectively.The MAE,RMSE,R²of MLP and XGBoost Models on the training sets were 0.158,0.191,0.925 and 0.032,0.091,0.985,respectively.ConclusionssBoth MLPand XGBoost models have good predictive ability for the dose of HIFU ablation of uterine fibroids,and can assist clinicians to develop a more reasonable treatment plan for patients with uterine fibroids.

关 键 词：高强度聚焦超声 子宫肌瘤 机器学习 剂量预测 

分 类 号：R445.1[医药卫生—影像医学与核医学] R737.33[医药卫生—诊断学]