抑郁症脑电信号特征及分类研究

发布时间：2018-04-17 10:15

本文选题：抑郁症 + 功率谱　；参考：《北京工业大学》2014年硕士论文

【摘要】：抑郁症是发病率较高的情感障碍精神疾病。人脑无刺激状态下释放出的脑电信号能反映大脑的不同状态，其中包括异常状态，已是临床上精神疾病检查的常用工具。首先，本研究希望在抑郁症病人的脑电信号中发现与正常人异常的特征，以便深入了解抑郁症的疾病机理，以及对诊断和治疗效果评估有所帮助。接着用3种著名模式识别算法，使用本研究发现的特征对脑电信号进行分类研究，以检验这些特征的分类效果。本研究对正常人，，抑郁症未用药病人和抑郁症用药病人这3种人群采样，形成3组被试。其中正常组14人，抑郁症未用药组11人，抑郁症用药组11人。这3组被试在年龄，性别与受教育程度上匹配。在被试者没有刺激的并且清醒的状态下，采集8分钟的脑电信号。本研究中深入研究三组被试在脑电信号的绝对功率，相对功率比，左右不对称性和安静过程差异性各个频带间的显著差异，发现以下有意义的结果：绝对功率的alpha1，alpha2频带；左右不对称性的theta频带；安静过程差异性绝对功率方面的alpha2频带是素质依赖性特征。绝对功率的delta频带；相对功率比的alpha1，theta频带；左右不对称性的alpha1，alpha2，beta1，delta频带；安静过程差异性方面的theta，delta频带是状态依赖性特征。另外，绝对功率的beta1频带，相对功率比的beta1频带是可以反映抗抑郁剂副作用的特征。本研究还使用抑郁症脑电信号绝对功率，相对功率比，左右不对称性，安静过程差异性4个特征进行了分类研究。每种特征下又分为7个频带与全频带的综合。分类算法分别是C4.5决策树算法，朴素贝叶斯算法和k最近邻算法。从3组分类结果来看，其中C4.5决策树算法最高分类正确率达到70.00%，朴素贝叶斯算法最高能达到83.33%，k最近邻算法最高能达到70.37%。并且对抑郁症素质依赖性和状态依赖性特征做了分类检验。
[Abstract]:Depression is a high incidence of emotional disorders mental illness.The electroencephalogram (EEG) released by the human brain without stimulation can reflect the different states of the brain, including abnormal state. It is a common tool for the clinical examination of mental illness.First of all, this study hopes to find abnormal features with normal people in EEG signals of depression patients, so as to understand the disease mechanism of depression and to evaluate the diagnosis and treatment effect.Then three famous pattern recognition algorithms are used to study the classification of EEG signals using the features found in this study to test the classification effect of these features.In this study, 3 groups of subjects were collected from normal subjects, untreated depression patients and depressive patients.There were 14 cases in normal group, 11 cases in untreated depression group and 11 cases in depressive drug group.The three groups matched age, gender and educational attainment.When the subjects were not stimulated and awake, 8-minute EEG signals were collected.In this study, we studied the significant differences in the absolute power, relative power ratio, left and right asymmetry and quiet process difference between the three groups of subjects in different frequency bands, and found the following significant results: the absolute power of alpha1 / alpha2 band;The theta band with left and right asymmetry and the alpha2 band with absolute power difference in quiet process are quality-dependent.The delta frequency band of absolute power; the alpha1 / theta band of relative power ratio; the left and right asymmetrical alpha1 / alpha2B / beta / delta band; the deltas frequency band of the quiet process difference are state-dependent.In addition, the beta1 band of absolute power and the beta1 band of relative power ratio can reflect the side effects of antidepressants.In this study, four characteristics of depression EEG, such as absolute power, relative power ratio, left and right asymmetry and quiet process difference, were used to classify the EEG signals.Each feature is divided into seven bands and full band synthesis.The classification algorithms are C4.5 decision tree algorithm, naive Bayes algorithm and k-nearest neighbor algorithm.From the results of three groups of classification results, the C4.5 decision tree algorithm can achieve the highest classification accuracy of 70.00g and the naive Bayesian algorithm can reach the highest 83.33k nearest neighbor algorithm up to 70.37.At the same time, the quality dependence and state-dependent characteristics of depression were tested by classification.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R749.4

【参考文献】