| 胡钦瑞,李扬,谢达森,等.基于机器学习的黄斑前膜患病状态识别模型构建及诊断效能验证[J].中国临床保健杂志,2025,28(5):693-699. |
| 基于机器学习的黄斑前膜患病状态识别模型构建及诊断效能验证 |
| Construction of a macular pucker disease state recognition model based on machine learning and verification of its diagnostic efficacy |
| 投稿时间：2025-07-08 |
| DOI：10.3969/J.issn.1672-6790.2025.05.022 |
| 中文关键词: 黄斑变性 机器学习 诊断技术,眼科 影响因素分析 |
| 英文关键词: Macular degeneration Machine learning Diagnostic techniques,ophthalmological Root cause analysis 〖FL |
| 基金项目:福建省自然科学基金项目(2023J011584)；福建省厦门市自然科学基金面上项目(3502Z20227290)；福建省宁德市自然科学基金项目(2022J01) |
| 作者 | 单位 | E-mail | | 胡钦瑞 | 厦门大学附属厦门眼科中心,福建厦门 361004
厦门市眼部疾病临床医学研究中心,福建厦门 361013 | zhangzhaode1973@163.com | | 李扬 | 厦门大学附属厦门眼科中心,福建厦门 361004
厦门市眼部疾病重点实验室,福建厦门 361013 | zhangzhaode1973@163.com | | 谢达森 | 福建省眼表与角膜病重点实验室,福建厦门 361102
厦门市眼表与角膜疾病重点实验室,福建厦门 361013 | zhangzhaode1973@163.com | | 王晓颖 | 厦门大学附属厦门眼科中心,福建厦门 361004
厦门大学附属厦门眼科中心转化医学研究所,福建厦门 361001 | zhangzhaode1973@163.com | | 陈小妍 | 厦门大学附属厦门眼科中心,福建厦门 361004 | zhangzhaode1973@163.com | | 王斌 | 厦门大学附属厦门眼科中心,福建厦门 361004
厦门市眼部疾病临床医学研究中心,福建厦门 361013 | zhangzhaode1973@163.com | | 张招德 | 宁德师范学院附属宁德市医院,福建宁德 352100 | zhangzhaode1973@163.com |
|
| 摘要点击次数: 56 |
| 全文下载次数: 28 |
| 中文摘要: |
| 目的 建立机器学习模型识别黄斑前膜患病状态,并进行可解释性分析,解析可能与黄斑前膜患病状态显著相关的因素。方法 基于福建省眼病流行病学横断面调查数据,纳入全面的健康和社会人口学信息。研究过程包括数据预处理、特征选择和模型训练,创建并评估多种机器学习模型,包括决策树、随机森林、梯度提升决策树、Light Gradient Boosting Machine(LightGBM)和回归模型。使用Python中的Shapley Additive exPlanations(SHAP,Shapley加性解释)方法进行详细分析,以提高最佳模型的可解释性。结果 分析覆盖了8 211例参与者,共33个变量。在测试的机器学习模型中,LightGBM模型表现最佳,准确度为0.913 9,受试者工作特征曲线下面积为0.94。SHAP可解释性分析发现,收缩压是与患病状态关联性最强的特征,其次是视力、散光、地理位置、年龄和性别。结论 机器学习模型可有效识别黄斑前膜患病状态,其中收缩压和视力等关键因素是中老年人群黄斑疾病的重要关联指标。 |
| 英文摘要: |
| Objective To establish a machine learning model for identifying the state of macular pucker and perform interpretability analysis to identify factors that may be significantly related to the state of macular pucker.Methods Based on the cross-sectional survey data on the epidemiology of eye diseases in Fujian Province,comprehensive health and socio-demographic information was incorporated.The research process involved data preprocessing,feature selection,and model training.Multiple machine learning models were created and evaluated,including decision trees,random forests,gradient boosting decision trees,Light Gradient Boosting Machine (LightGBM),and regression models.Detailed analysis was conducted using the Shapley Additive exPlanations (SHAP) method in Python to enhance the interpretability of the optimal model.Results The analysis covered 8,211 participants with a total of 33 variables.Among the tested machine learning models,the LightGBM model performed the best,with an accuracy of 0.913 9 and an area under the receiver operating characteristic curve (AUC) of 0.94.The SHAP interpretability analysis revealed that systolic blood pressure was the feature most strongly associated with disease status,followed by visual acuity,astigmatism,geographical location,age,and gender.Conclusions Machine learning models can effectively identify the condition of macular pucker,with key factors such as systolic blood pressure and vision being important indicators associated with macular diseases in middle-aged and elderly populations. |
|
查看全文
|
| 关闭 |
|
|
|