aEEG信号图像重构及基于集成SVM的分类研究
发布时间：2018-11-12 10:26
【摘要】：新生儿振幅整合脑电(aEEG)是一种有效的新生儿脑功能状态长期监测技术。因其成本低,操作简单以及可长期监测等优点,aEEG已经越来越多地应用于新生儿重症监护病房中。设计一个自动判别方案或算法实现对aEEG信号的自动判别非常重要,这不仅能够使医生从识别脑功能异常这项繁重任务中解放出来,从而专注于解决这些脑部疾病,而且对aEEG这项技术的使用和推广具有深远意义。本研究主要工作是设计一套自动化的aEEG信号判读方案。该方案从数据中提取其有效特征表示并训练出模型,之后通过该模型来预测未知aEEG样本的类别。论文首次提出一种aEEG信号重构的方法,通过该方法获得aEEG信号幅值频率等高图,该图比原始aEEG信号更能直观反映局部振幅变化特点。本论文从不同角度刻画aEEG信号并提取了 4类特征,包含图像特征、线性特征、直方图特征以及复杂度特征。其中,论文将LBP图像算子引入aEEG信号幅值频率等高图分析中并得到其二维图像特征,该特征比一般基于幅值的一维特征更能刻画aEEG信号的特点;下边界作为医生判读aEEG重要依据但医学上没有具体定义,论文重新定义aEEG下边界并引入打分系统对其量化;同时,还第一次引入自排列熵来量化aEEG信号复杂度。最后,论文提出一种基于支持向量机的集成方法称为Hybrid-SVM,用于实现aEEG信号的自动分类。为证明其有效性和准确性,论文在包含276个aEEG样本的数据集上进行验证。实验结果表明,图像特征可以有效刻画正异常aEEG信号特征,并且可以不同程度地提升各常用分类器的分类性能。与原始的支持向量机算法以及其它集成方法相比,本文的aEEG信号识别方法综合性能最优,其中识别准确率达到95.68%。基于集成方法Hybrid-SVM的aEEG信号分类方法有助于临床检测新生儿脑部异常。
[Abstract]:Amplitude integrated EEG (aEEG) is an effective long-term monitoring technique for neonatal brain function status. Because of its advantages of low cost, simple operation and long-term monitoring, aEEG has been used more and more in neonatal intensive care unit. It is very important to design an automatic discriminant scheme or algorithm to identify aEEG signals automatically, which not only frees doctors from the heavy task of identifying abnormal brain function, but also focuses on solving these brain diseases. Moreover, it is of great significance to use and popularize aEEG technology. The main work of this study is to design a set of automatic aEEG signal interpretation scheme. The scheme extracts its effective feature representation from the data and trains the model. Then the model is used to predict the class of unknown aEEG samples. In this paper, a method of aEEG signal reconstruction is proposed for the first time. The amplitude and frequency contour diagram of aEEG signal is obtained by this method, which can reflect the characteristic of local amplitude change more intuitively than the original aEEG signal. This paper describes aEEG signals from different angles and extracts four kinds of features, including image features, linear features, histogram features and complexity features. In this paper, the LBP image operator is introduced into the amplitude and frequency contour graph analysis of aEEG signal and its two-dimensional image feature is obtained. This feature is more capable of characterizing the aEEG signal than the general one-dimensional feature based on the amplitude. The lower boundary is an important basis for doctors to interpret aEEG, but there is no specific definition in medicine. This paper redefines the lower boundary of aEEG and introduces a scoring system to quantify it. At the same time, the self-permutation entropy is introduced for the first time to quantify the complexity of aEEG signals. Finally, an ensemble method based on support vector machine (SVM) called Hybrid-SVM, is proposed to realize the automatic classification of aEEG signals. To prove its validity and accuracy, this paper is validated on a dataset containing 276 aEEG samples. The experimental results show that the image features can effectively describe the positive abnormal aEEG signal features, and can improve the classification performance of the commonly used classifiers in varying degrees. Compared with the original support vector machine (SVM) algorithm and other ensemble methods, the aEEG signal recognition method in this paper has the best comprehensive performance, and the recognition accuracy is 95.68%. The method of aEEG signal classification based on Hybrid-SVM is helpful for clinical detection of neonatal brain abnormalities.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R722.1;TP391.41
本文编号：2326844
[Abstract]:Amplitude integrated EEG (aEEG) is an effective long-term monitoring technique for neonatal brain function status. Because of its advantages of low cost, simple operation and long-term monitoring, aEEG has been used more and more in neonatal intensive care unit. It is very important to design an automatic discriminant scheme or algorithm to identify aEEG signals automatically, which not only frees doctors from the heavy task of identifying abnormal brain function, but also focuses on solving these brain diseases. Moreover, it is of great significance to use and popularize aEEG technology. The main work of this study is to design a set of automatic aEEG signal interpretation scheme. The scheme extracts its effective feature representation from the data and trains the model. Then the model is used to predict the class of unknown aEEG samples. In this paper, a method of aEEG signal reconstruction is proposed for the first time. The amplitude and frequency contour diagram of aEEG signal is obtained by this method, which can reflect the characteristic of local amplitude change more intuitively than the original aEEG signal. This paper describes aEEG signals from different angles and extracts four kinds of features, including image features, linear features, histogram features and complexity features. In this paper, the LBP image operator is introduced into the amplitude and frequency contour graph analysis of aEEG signal and its two-dimensional image feature is obtained. This feature is more capable of characterizing the aEEG signal than the general one-dimensional feature based on the amplitude. The lower boundary is an important basis for doctors to interpret aEEG, but there is no specific definition in medicine. This paper redefines the lower boundary of aEEG and introduces a scoring system to quantify it. At the same time, the self-permutation entropy is introduced for the first time to quantify the complexity of aEEG signals. Finally, an ensemble method based on support vector machine (SVM) called Hybrid-SVM, is proposed to realize the automatic classification of aEEG signals. To prove its validity and accuracy, this paper is validated on a dataset containing 276 aEEG samples. The experimental results show that the image features can effectively describe the positive abnormal aEEG signal features, and can improve the classification performance of the commonly used classifiers in varying degrees. Compared with the original support vector machine (SVM) algorithm and other ensemble methods, the aEEG signal recognition method in this paper has the best comprehensive performance, and the recognition accuracy is 95.68%. The method of aEEG signal classification based on Hybrid-SVM is helpful for clinical detection of neonatal brain abnormalities.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R722.1;TP391.41
【参考文献】
相关期刊论文 前5条
1 谢平;魏秀利;杜义浩;陈晓玲;;基于自排序熵的表面肌电信号特征提取方法[J];模式识别与人工智能;2014年06期
2 王宝宏;时文玲;王克煊;李燕;罗瑶;;振幅整合脑电图在新生儿缺氧缺血性脑病中的诊断价值[J];中国儿童保健杂志;2013年10期
3 慈春燕;李文;卢宪梅;;早产儿早期振幅整合脑电图特点的分析[J];山东大学学报(医学版);2012年09期
4 王绍宾;王一抗;王志中;施亿峗;邵肖梅;;基于非线性动力学的振幅整合新生儿脑电图分析[J];生物医学工程学杂志;2009年06期
5 张丹丹;丁海艳;刘云峰;周丛乐;丁海曙;叶大田;;振幅整合脑电图的实现及其在新生儿脑功能监护中的应用前景[J];中国医疗器械信息;2008年03期
相关硕士学位论文 前1条
1 王愈;新生儿振幅整合脑电自动识别研究[D];华东师范大学;2015年
，本文编号：2326844
本文链接：https://www.wllwen.com/yixuelunwen/eklw/2326844.html
最近更新
教材专著
