| 于佳妮,贾西平,马先盛,谢 莉,沈 威,李 涛,赵嘉培,李 浩,杨幸华,黄伟新,董红琳,贺灵慧,刘 晨,江潭耀,吕媛浩,罗 倩,燕铁斌.基于神经网络的ICF康复组合评定量化标准功能分级算法模型构建及其验证[J].中国康复医学杂志,2022,(10):1347~1353 |
| 基于神经网络的ICF康复组合评定量化标准功能分级算法模型构建及其验证 点此下载全文 |
| 于佳妮 贾西平 马先盛 谢 莉 沈 威 李 涛 赵嘉培 李 浩 杨幸华 黄伟新 董红琳 贺灵慧 刘 晨 江潭耀 吕媛浩 罗 倩 燕铁斌 |
| 广州中医药大学第二附属医院，广州，510120 |
| 基金项目：教育部创新促教基金项目（2018A01026）；国家自然科学基金面上项目（82272614）；国家自然科学基金青年项目（82104964） |
| DOI：10.3969/j.issn.1001-1242.2022.10.009 |
| 摘要点击次数: 825 |
| 全文下载次数: 453 |
| 摘要： |
| 摘要
目的：利用人工智能神经网络方式构建ICF康复组合（ICF-RS）评定量化标准总体及三个维度（身体功能、活动、参与）功能分级的算法模型，为应用ICF-RS评定量化标准进行数据分析及功能分级提供解决方案。
方法：本研究利用中文版ICF-RS评定量化标准，通过多中心合作，采用分层比例抽样的方法收集了6家已开展ICF-RS评定量化标准临床应用的康复医学科住院患者ICF-RS数据，以多个专家对同一患者的方式获取ICF-RS评定量化标准三个维度及整体功能状况的等级评价结果。借助于神经网络算法构建ICF-RS评定量化标准的各维度及整体功能分级模型，采用计算机k折交叉验证法选择最优模型参数，并计算模型预测受试者工作特征曲线[receiver operating characteristic(ROC) curve]下面积（area under ROC curve，AUC）、准确率（accuracy，ACC）和F1分数（F1-score）。模型构建后，再通过专家和患者一对一的方式收集100例ICF-RS数据用于对已建立的模型进行临床再测试，通过计算ROC、AUC、ACC和F1对模型性能进行评价。
结果：共收集584例住院患者的ICF-RS数据，其中484例数据用于构建及验证模型，100例数据用于测试模型的预测性能。根据k折交叉验证法结果显示，身体功能维度、活动维度、参与维度及整体ICF-RS功能分级模型的AUC分别是89.00%、92.00%、87.00%和87.00%，ACC分别达到75.19%、78.10%、72.91%和73.53%，F1分别是73.68%、77.04%%、69.28%、58.95%。在模型建立后将重新收集到的100例ICF-RS数据输入模型计算，发现各模型ROC曲线良好，AUC分别是89.04%、91.81%、86.85%、86.89%，ACC分别是64.00%、72.00%、61.00%、65.00%，F1分别是48.30%、59.95%、64.06%、49.35%。
结论：基于神经网络建立的ICF-RS整体及各维度功能分级算法模型对ICF-RS数据的功能等级预测准确率良好，预测价值较高，具有良好的临床应用价值。 |
| 关键词：国际功能、残疾和健康分类康复组合 机器学习 神经网络 功能分级 |
| Establishment and verification of ICF-RS assessment and quantification standard functional grading model based on neural network Download Fulltext |
|
| The Second Affiliated Hospital of Guangzhou University of Chinese Medicine，Guangzhou，510120 |
| Fund Project: |
| Abstract: |
| Abstract
Objective： To establish the functional classification models of ICF-RS and three dimensions （body functions， activities and participation） based on neural network， and to provide a solution for data analysis and functional evaluation by ICF-RS assessment and quantification standard.
Method： The Chinese version of ICF-RS assessment and quantitative criteria was used to collect ICF-RS data of inpatients in rehabilitation medicine departments by stratified proportional sampling method from 6 different medical institutions through multi-center cooperation， and the functional classification results of three dimensions and overall of the same patient were obtained by multiple experts. At the same time， the functional classification model of ICF-RS for each dimension and overall were established by using neural network. K-fold cross validation was used to select the optimal model parameters and calculate the model Area under ROC curve（AUC）， Accuracy and F1-score. After models were built， 100 ICF-RS data collected by experts and patients in a one-to-one manner were input into the models. The performance of models were tested by Receiver Operating Characteristic（ROC） curve， AUC， ACC and F1.
Result： We collected a total of 584 inpatients' ICF-RS data，in which 484 ICF-RS data were used to establish and validate the model， and 100 ICF-RS data were used to test the predictive performance of the models. According to the cross-validation results， the AUC of body functions dimension， activities dimension， participation dimension and overall ICF-RS functional classification model were 89.00%、92.00%、87.00% and 87.00%； ACC were 75.19%， 78.10%， 72.91% and 73.53%； F1 were 73.68%、97.04%、69.28% and 58.95%， respectively. After the models were established， 100 ICF-RS data required for model testing were input into the models. ROC curves of all models were good and AUC were 89.04%， 91.81%， 86.85%， 86.89%； ACC were 64.00%， 72.00%， 61.00%， 65.00%； F1 were 48.30%， 59.95%， 64.06%， 49.35%， respectively.
Conclusion： The ICF-RS and three dimensions functional classification model based on neural network have good accuracy and high predictive value. The models have good clinical application value. |
| Keywords:ICF rehabilitation set machine learning neural network functional classification |
|
| 查看全文 查看/发表评论 |
