机器学习模型k近邻算法分析脑电图对主观性耳鸣的诊断价值  

The diagnostic value of machine learning model k-nearest neighbor algorithm to analyze EEG for subjective tinnitus

作　　者：王荣国[1] 高洁[1] 宋晓飞[1] 屈永涛[1] WANG Rongguo;GAO Jie;SONG Xiaofei;QU Yongtao(Department of Otolaryngology,Hebei General Hospital,Shijiazhuang 050000,Hebei,China)

机构地区：[1]河北省人民医院耳鼻喉科,河北石家庄050000

出　　处：《中南医学科学杂志》2023年第5期696-698,共3页Medical Science Journal of Central South China

基　　金：河北省卫生健康委员会项目(20210800)。

摘　　要：目的探讨机器学习模型k近邻算法分析脑电图对主观性耳鸣的诊断价值。方法纳入主观性耳鸣患者87例(耳鸣组),健康受试者91例(对照组)。使用MATLAB和EEGLAB工具箱、小波包变换和样本熵相结合的方法分析两组δ、θ、α1、α2、β1、β2、β3、γ频段在耳鸣发生网络相关7个区域的样本熵差异。对耳鸣脑电图特征数据使用Python的scikit-learn包进行k近邻算法分析,使用准确率、召回率、精确度和F1得分评估k近邻算法对主观性耳鸣的诊断价值。结果两组样本熵在左听觉、左额叶、中央、右顶叶和左顶叶等区域差异有显著性(P<0.05)。耳鸣组δ、α2和β1节律平均熵大于对照组,θ、α1、β2、β3和γ节律平均熵小于对照组(P<0.05)。耳鸣组和对照组样本熵在FC5、C1、CP1和P4单通道中差异有显著性(P<0.05)。k近邻算法对主观性耳鸣的诊断准确率为91.98%,召回率为90.24%,准确率为96.28%,F1得分为93.12%。结论机器学习模型k近邻算法分析脑电图结果可以辅助临床医生对耳鸣进行诊断。Aim To evaluate the diagnostic value of machine learning model k-nearest neighbor algorithm in analyzing EEG for subjective tinnitus.Methods 87 subjective tinnitus patients(tinnitus group)and 91 healthy subjects(control group)were included.A combination of MATLAB and EEGLAB toolboxes,wavelet packet transform,and sample entropy were used to analyze sample entropy differences ofδ,θ,α1,α2,β1,β2,β3,γfrequency band in 7 regions related to tinnitus occurrence network.The characteristic data of tinnitus electroencephalogram were analyzed by using Python s scikit-learn package for k-neighbor algorithm analysis,and analyzing accuracy,recall,accuracy,and F1 score for subjective tinnitus by using k-neighbor algorithm to evaluate the diagnostic value.Results There was a significant difference in entropy between the two groups of samples in left auditory,left frontal,central,right parietal,and left parietal lobes(P<0.05).Tinnitus groupδ,α2 andβ1 average entropy of rhythm was greater than that of the control group,andθ,α1,β2,β3 andγaverage entropy of the rhythm was lower than that of the control group(P<0.05).There was a significant difference in sample entropy between the tinnitus group and the control group in FC5,C1,CP1,and P4 single channels(P<0.05).The k-nearest neighbor algorithm has a diagnostic accuracy of 91.98%,a recall rate of 90.24%,an accuracy rate of 96.28%,and an F1 score of 93.12%for subjective tinnitus.Conclusion Machine learning model k-nearest neighbor algorithm analysis of EEG can assist clinical doctors in diagnosing tinnitus.

关 键 词：K近邻算法 脑电图 主观性耳鸣 样本熵 小波包变换 

分 类 号：R764.45[医药卫生—耳鼻咽喉科]