基于机器学习的卒中后抑郁影响因素分析

发布时间：2018-02-14 16:48

本文关键词： 脑卒中卒中后抑郁机器学习随机森林模型单规则算法集成学习　出处：《中医杂志》2017年17期 　论文类型：期刊论文

【摘要】：目的通过机器学习判断脑卒中患者发生抑郁的影响因素。方法从病历系统中提取符合纳入条件的688例脑卒中患者的病历资料,包括年龄、性别、脉象、面色、舌质、舌苔、中医药干预方式、体重指数(BMI)、血压、血糖、血甘油三酯、血总胆固醇、吸烟史、饮酒史、抑郁家族史、影像学卒中病灶部位及最终抑郁判断结果。采用单规则(1R)算法进行学习并判断提取信息中影响卒中后患者是否抑郁的危险因素,继而将所搜集病例分为训练数据集(500例)和测试数据集(188例),并使用随机森林模型形成最优判别结果。结果通过单规则算法得出脑卒中后是否抑郁最重要的影响因素为卒中病灶所在部位,其中计算机推测卒中病灶位于额叶及颞叶者最易发生卒中后抑郁,基底节、脑干、小脑、延髓、枕叶的病灶则不易引起抑郁,其准确分类率达到88.95%(612/688例)。对前500例训练数据集进行随机森林模型判别,其抑郁判断的正确率为98.2%;188例测试集判断结果正确率达99.47%;将688例患者资料运用随机森林模型进行学习,总的正确率为98.84%。重要性测度结果显示,病灶位置、中医药干预手段及抑郁家族史是脑卒中后是否发生抑郁最重要指标的前3位。结论病灶位于额颞叶的脑卒中患者以及具有抑郁史的患者更容易发生卒中后抑郁。
[Abstract]:Objective to determine the influencing factors of depression in stroke patients by machine learning. Methods the medical records of 688 stroke patients, including age, sex, pulse, facial color, tongue quality and tongue coating, were extracted from the medical record system. Chinese medicine intervention, body mass index BMIA, blood pressure, blood glucose, serum triglyceride, blood total cholesterol, smoking history, drinking history, family history of depression, Imaging stroke focus location and final depression judgment results. Single rule 1R algorithm was used to study and determine the risk factors of extracting information to affect the depression of post-stroke patients. Then, the collected cases were divided into training data set (500 cases) and test data set (188 cases), and the random forest model was used to form the optimal discriminant result. Results the most important effect of depression after stroke was obtained by a single rule algorithm. The location of the lesion was the location of the stroke. The computer speculated that the lesions located in the frontal lobe and temporal lobe were most likely to develop post-stroke depression, and the lesions in basal ganglia, brainstem, cerebellum, medulla oblongata and occipital lobe were not likely to cause depression. The accurate classification rate reached 88.95% 612 / 688 cases. Random forest model was used to judge the data set of the first 500 cases of training, and the correct rate of judging depression was 98.2%. The correct rate of judging the result of the test set was 99.470.The data of 688 patients were studied with the random forest model. The overall accuracy was 98.84. The importance measure showed that the location of the lesion, Chinese medicine intervention and family history of depression are the first three most important indicators of depression after stroke. Conclusion patients with focal lesions located in frontotemporal lobe and patients with history of depression are more likely to develop post-stroke depression.
【作者单位】：广州中医药大学针灸康复临床医学院;郑州市中心医院康复医学部;广州中医药大学热带医学研究所;
【基金】：国家重点基础研究发展计划(“973”计划)(2012CB518504)
【分类号】：R277.7

【相似文献】