基于稀疏表示的脑电棘波检测算法研究

EEG Spike Detection Using Sparse Representation

提出了一种基于稀疏表示的脑电棘波检测算法,首先以高斯函数及其一、二阶导数为原子的生成函数构建了一个冗余多成份字典,再应用匹配追踪算法获取脑电信号在此字典下的M项稀疏逼近,由该逼近的导数信息与原子的结构参数可准确度量瞬时波形的形态结构特征,进而提出基于形态结构匹配的棘波检测算法,克服了Gabor字典不能识别周期化棘波序列的缺点,同时能够有效去除背景节律与伪迹的影响,检测结果表明该算法针对临床EEG信号的检测率高达93.9%,正确率高达88.0%.

An approach is proposed to automatically detect EEG spikes,based on sparse representation of signals.Firstly,Gaussian function and its first and second derivations are used as the generating functions to construct the redundant multi-component dictionary.Secondly,the M-term sparse approximation is obtained using matching pursuit method in our dictionary.Various morphological structure features of transients can be extracted accurately,utilizing the derivative information of the sparse approximation and the structure parameters of atoms. Finally, a detection algorithm is presented, based on morphological StlUcture match. It can overcome the Gabor dictionary＇s shortcomings that can＇ t detect the series of spike-wave complexes, moreover, can reject back- ground transients and artifacts effectively. The experimental results indicate that our detection technique yields high sensitivity of 93.9 percent and selectivity of 88.0 percent evaluated on clinical EEGs.