基于S变换和深信度网络的自动癫痫检测

发布时间：2018-04-20 13:08

本文选题：癫痫检测 + 限制波尔兹曼机(RBM)　；参考：《山东大学》2017年硕士论文

【摘要】：癫痫是一种慢性神经系统综合征,可以由遗传、多种神经系统疾病以及全身疾病引起,具有反复性和突发性。癫痫可见于各个年龄段,其发作的临床表现主要有突发意识丧失、全身强直、抽搐、躯体或者肢体电击样抖动等。微观上,癫痫发作是由大脑内部神经元异常放电活动导致的,临床上常用脑电图记录这种电活动,从而辅助癫痫疾病的治疗和研究。脑电图一方面包含了大量的生理和病理信息,为医生诊断和治疗癫痫疾病提供了有力的帮助;另一方面脑电图通常包含数十个小时的脑电数据,而且这些数据需要医务人员根据临床经验人工读取,因此分析脑电图是一项非常枯燥与耗时的工作,人工的主观性和疲劳会影响诊断的准确性。为了提高癫痫诊断与治疗的效率和准确性,信息技术的发展带来了自动癫痫检测。自动癫痫检测是运用计算机以及检测算法进行脑电图的读取和分析,从而大大减轻了人工读图的工作负担,提高了癫痫治疗效率。本文提出了一种基于深度学习的深信度网络(Deep Belief Network,DBN)进行自动癫痫检测的算法。深信度网络是深度学习中的经典算法模型,它可以看作由多个限制玻尔兹曼机(Restricted Boltzmann Machines,RBM)叠加而成。目前,DBN已经广泛应用到对象建模、特征提取、识别等领域。本文提出的算法框架如下:首先对原始脑电数据分段,运用S变换进行时频分析,从而在每段中提取一组特征模值;然后按照训练样本和测试样本进行线性归一化;接着把归一化的特征模值送入自定义结构的深信度网络中,通过逐层提取和抽象,最终得到两类输出,即发作和非发作;最后运用做差处理的方法,对两类输出结果进行后处理,从而提高分类的准确性。本文的实验数据来自德国弗莱堡医学院癫痫研究中心,采用了其中9位病人的34次癫痫发作对该方法进行评估。本文提出的方法将深度学习领域中的经典算法模型应用到癫痫检测方法,推动了癫痫检测技术的发展。
[Abstract]:Epilepsy is a kind of chronic nervous system syndrome, which can be caused by heredity, many nervous system diseases and systemic diseases. Epilepsy can be seen in all ages, the main clinical manifestations of seizures are sudden loss of consciousness, systemic ankylosis, convulsions, body or limb electrocution shaking and so on. Microscopically, epileptic seizures are caused by abnormal discharges of neurons in the brain, which are recorded by EEG in clinic to assist the treatment and research of epileptic diseases. On the one hand, EEG contains a lot of physiological and pathological information, which provides powerful help for doctors in the diagnosis and treatment of epilepsy; on the other hand, EEG usually contains tens of hours of EEG data. These data need to be read manually according to the clinical experience, so EEG analysis is a very boring and time-consuming work, artificial subjectivity and fatigue will affect the accuracy of diagnosis. In order to improve the efficiency and accuracy of epilepsy diagnosis and treatment, the development of information technology has brought automatic epilepsy detection. Automatic epilepsy detection is to use computer and detection algorithm to read and analyze EEG, thus greatly reducing the workload of manual reading map and improving the efficiency of epilepsy treatment. In this paper, an automatic epilepsy detection algorithm based on Deep Belief Network (DBN) is proposed. It is believed that the degree network is a classical algorithm model in depth learning, which can be considered as superposed by several restricted Boltzmann machines, restricted Boltzmann machines and RBMs. At present, DBN has been widely used in object modeling, feature extraction, recognition and other fields. The framework of this algorithm is as follows: firstly, the original EEG data are segmented, and then a set of eigenmodes are extracted from each segment by using S-transform, and then linear normalization is carried out according to the training sample and the test sample. Then the normalized eigenmode value is sent into the confidence network of the custom structure, and two types of output are obtained by extracting and abstracting them layer by layer, that is, attack and non-seizure. Finally, the method of differential processing is used. In order to improve the accuracy of classification, two kinds of output results are processed. The data were collected from the Epilepsy Research Center of the Freiburg School of Medicine in Germany. The method was evaluated with 34 seizures in 9 of the patients. In this paper, the classical algorithm model in the field of deep learning is applied to epilepsy detection, which promotes the development of epilepsy detection technology.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R742.1;TP181

【相似文献】