[1]于秋彤,周睿涵,王捷,等.基于机器学习的脑卒中特征SHAP分析及高危因素机制探讨[J].中国医学物理学杂志,2026,43(9):1275-1282.[doi:DOI:10.3969/j.issn.1005-202X.2026.09.016]
　YU Qiutong,ZHOU Ruihan,WANG Jie,et al.Machine learning-based prediction of stroke risk with SHAP-based feature analysis and mechanism analysis of high-risk factors[J].Chinese Journal of Medical Physics,2026,43(9):1275-1282.[doi:DOI:10.3969/j.issn.1005-202X.2026.09.016]
点击复制

基于机器学习的脑卒中特征SHAP分析及高危因素机制探讨()

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
43卷
期数:
2026年第9期
页码:
1275-1282
栏目:
医学人工智能
出版日期:
2026-09-30

文章信息/Info

Title:
Machine learning-based prediction of stroke risk with SHAP-based feature analysis and mechanism analysis of high-risk factors
文章编号:
1005-202X（2026）09-1275-08
作者:
于秋彤1; 周睿涵1; 王捷2; 余善成2
1.南京医科大学医学影像学院， 江苏 南京 211166；2.南京医科大学生物医学工程与信息学院， 江苏 南京 211166
Author(s):
YU Qiutong1;  ZHOU Ruihan1;  WANG Jie2;  YU Shancheng2
1. School of Medical Imaging, Nanjing Medical University, Nanjing 211166, China 2. School of Biomedical Engineering and Informatics, Nanjing Medical University, Nanjing 211166, China
关键词:
脑卒中; 机器学习; 风险因素; 数据分析
Keywords:
Keywords: stroke machine learning risk factor data analysis
分类号:
R318；R743
DOI:
DOI:10.3969/j.issn.1005-202X.2026.09.016
文献标志码:
A
摘要:
目的：评估脑卒中风险预测中各个特征的重要性及其对预测结果的影响。方法：采用预处理后的公开数据集（brain_stroke），分别训练LGBM、Gradient Boosting、CatBoost和XGBoost模型。基于表现良好的模型，利用沙普利（SHAP）值进行特征分析排序，探讨主要风险因素。结果：4种模型的准确率均达到96%以上。全样本分析显示：吸烟状况是最重要的预测因素，曾经吸烟比正在吸烟的患病风险更大；行为和环境因素在脑卒中风险评估中发挥着关键作用，传统生理因素的重要性排在后面；曾经吸烟、年龄大、居住在城市、男性、私营企业工作、未婚、高血糖、高身体质量指数、高血压、心脏病等特征类别更倾向于推动预测结果为患病。结论：传统生理因素在脑卒中患病风险中的作用已经得到广泛认可，但是本研究发现行为与环境因素也发挥着重要作用，其重要性甚至高于传统生理因素。在预防策略制定和医疗资源合理分配时应更加注重行为与环境因素，从而有效降低脑卒中发病率。
Abstract:
Abstract: Objective To evaluate the significance of each feature in stroke risk prediction and its impact on the prediction results. Methods Four models, including LGBM, Gradient Boosting, CatBoost, and XGBoost, were trained on the preprocessed public dataset (brain_stroke). Based on the well-performing models, the SHapley Additive exPlanations (SHAP) values were utilized for feature analysis and ranking, thereby exploring the main risk factors. Results The accuracy of all 4 models exceeded 96%. The analysis on the entire sample identified smoking status as the most important predictive factor. Former smokers had a higher stroke risk than current smokers. Behavioral and environmental factors played a pivotal role in stroke risk assessment, while traditional physiological factors ranked in the bottom. Features such as prior smoking history, older age, urban residence, male, working in a private enterprise, unmarried status, elevated blood glucose, high body mass index, hypertension, and heart disease contributed to a higher predicted stroke risk. Conclusion Traditional physiological factors have been recognized as contributors to stroke risk. The findings in this study reveal that behavioral and environmental factors also play a significant role, sometimes even more important than conventional physiological ones. Greater attention should therefore be given to these behavioral and environmental factors in developing prevention strategies and allocating healthcare resources, thereby effectively reducing stroke incidence.

相似文献/References:

[1]高丹丹,吴宝明,何庆华,等.融合视听觉引导肌电反馈与生物电刺激的新型神经康复仪的研制[J].中国医学物理学杂志,2014,31(04):5085.[doi:10.3969/j.issn.1005-202X.2014.04.024]
[2]王美娇,李莎,岳海振,等. Model Analytics辅助的智能放疗计划建模[J].中国医学物理学杂志,2017,34(9):870.[doi:DOI:10.3969/j.issn.1005-202X.2017.09.002]
　[J].Chinese Journal of Medical Physics,2017,34(9):870.[doi:DOI:10.3969/j.issn.1005-202X.2017.09.002]
[3]周文,王瑜,肖红兵,等. 基于KPCA算法的阿尔茨海默症辅助诊断[J].中国医学物理学杂志,2018,35(4):404.[doi:DOI:10.3969/j.issn.1005-202X.2018.04.007]
　ZHOU Wen,WANG Yu,XIAO Hongbing,et al. Assisted diagnosis of Alzheimer’s disease based on KPCA algorithm[J].Chinese Journal of Medical Physics,2018,35(9):404.[doi:DOI:10.3969/j.issn.1005-202X.2018.04.007]
[4]余锦娟,林勇. 基于机器学习的骨质疏松性骨折预测研究[J].中国医学物理学杂志,2018,35(11):1329.[doi:DOI:10.3969/j.issn.1005-202X.2018.11.017]
　YU Jinjuan,LIN Yong. Prediction of osteoporotic fractures based on machine learning[J].Chinese Journal of Medical Physics,2018,35(9):1329.[doi:DOI:10.3969/j.issn.1005-202X.2018.11.017]
[5]刘渊,程玉玉,贺睿敏,等.基于机器学习的鼻咽癌转移淋巴结鉴别模型[J].中国医学物理学杂志,2019,36(11):1350.[doi:DOI:10.3969/j.issn.1005-202X.2019.11.020]
　LIU Yuan,CHENG Yuyu,HE Ruimin,et al.Machine learning-based classification model of lymph node metastasis in nasopharyngeal carcinoma[J].Chinese Journal of Medical Physics,2019,36(9):1350.[doi:DOI:10.3969/j.issn.1005-202X.2019.11.020]
[6]孙凯,姚旭峰,马风玲,等.基于机器学习的血细胞分类研究进展[J].中国医学物理学杂志,2020,37(1):127.[doi:DOI:10.3969/j.issn.1005-202X.2020.01.023]
　SUN Kai,YAO Xufeng,et al.Blood cell classification based on machine learning[J].Chinese Journal of Medical Physics,2020,37(9):127.[doi:DOI:10.3969/j.issn.1005-202X.2020.01.023]
[7]雷炳业,潘嘉瑜,吴逢春,等.基于机器学习的神经精神疾病辅助诊断研究进展[J].中国医学物理学杂志,2020,37(2):257.[doi:DOI:10.3969/j.issn.1005-202X.2020.02.022]
　LEI Bingye,PAN Jiayu,et al.Advances in auxiliary diagnosis of neuropsychiatric diseases based on machine learning[J].Chinese Journal of Medical Physics,2020,37(9):257.[doi:DOI:10.3969/j.issn.1005-202X.2020.02.022]
[8]李彩,范炤.基于机器学习的阿尔兹海默症分类预测[J].中国医学物理学杂志,2020,37(3):379.[doi:DOI:10.3969/j.issn.1005-202X.2020.03.023]
　LI Cai,FAN Zhao.Classification and prediction of Alzheimer’s disease based on machine learning[J].Chinese Journal of Medical Physics,2020,37(9):379.[doi:DOI:10.3969/j.issn.1005-202X.2020.03.023]
[9]闫凤,牛振洋,费振乐,等.机器学习在肺癌VMAT计划中对危及器官剂量预测的可行性[J].中国医学物理学杂志,2020,37(7):934.[doi:DOI:10.3969/j.issn.1005-202X.2020.07.025]
　YAN Feng,NIU Zhenyang,FEI Zhenle,et al.Feasibility of machine learning in OAR dosimetric prediction in VMAT plan for lung cancer[J].Chinese Journal of Medical Physics,2020,37(9):934.[doi:DOI:10.3969/j.issn.1005-202X.2020.07.025]
[10]黄仕雄,杨松华,王亮,等.基于预测剂量引导的宫颈癌自动计划研究[J].中国医学物理学杂志,2020,37(9):1101.[doi:10.3969/j.issn.1005-202X.2020.09.004]
　HUANG Shixiong,YANG Songhua,WANG Liang,et al.Automatic planning of radiotherapy for cervical carcinoma based on dose prediction[J].Chinese Journal of Medical Physics,2020,37(9):1101.[doi:10.3969/j.issn.1005-202X.2020.09.004]
[11]陈爱德,胡亚恒,魏文存.基于机器学习的多模态MRI对脑卒中病灶特征的预测模型[J].中国医学物理学杂志,2026,43(2):276.[doi:DOI:10.3969/j.issn.1005-202X.2026.02.019]
　CHEN Aide,HU Yaheng,WEI Wencun.Machine learning-based predictive model for stroke lesion characteristics using multimodal MRI[J].Chinese Journal of Medical Physics,2026,43(9):276.[doi:DOI:10.3969/j.issn.1005-202X.2026.02.019]

备注/Memo

备注/Memo:
【收稿日期】2026-03-21 【基金项目】全国高等学校大学物理改革研究项目（2024PR057）；江苏省大学生创新创业训练计划（X202510312114）；江苏省教育科学规划课题（C/2023/01/22）；南京医科大学教育研究课题（2025YB061） 【作者简介】于秋彤，研究方向：基于可解释机器学习模型的脑卒中风险预测与因素分析，E-mail: 3265932582@qq.com 【通信作者】余善成，博士，副教授，研究方向：人工智能的医学应用，E-mail: ysc5458@njmu.edu.cn
更新日期/Last Update: 2026-09-30