计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (20): 119-128.DOI: 10.3778/j.issn.1002-8331.2206-0275

• 模式识别与人工智能 • 上一篇    下一篇

复杂网络与GA-PSO算法下的癫痫脑电识别

王若凡,王浩东,石连栓   

  1. 天津职业技术师范大学 信息技术工程学院,天津 300022
  • 出版日期:2023-10-15 发布日期:2023-10-15

EEG Recognition of Epilepsy Based on Complex Network and GA-PSO

WANG Ruofan, WANG Haodong, SHI Lianshuan   

  1. School of Information Technology Engineering, Tianjin University of Technology and Education, Tianjin 300022, China
  • Online:2023-10-15 Published:2023-10-15

摘要: 癫痫(epilepsy,EP)是大脑神经元异常放电的一种慢性脑部神经疾病,临床上大多由专业神经科医生在癫痫患者的脑电图(electroencephalogram,EEG)信号上进行人工视觉检测分析,容易受医生的主观影响。因此,筛选有效的脑电特征对EP进行自动识别显得尤为重要。对多通道EEG信号构建相位同步(phase synchronization index,PSI)脑网络,从网络节点和结构功能角度提取15个拓扑特征,并结合支持向量机(support vector machine,SVM)分类器实现癫痫脑电自动识别。分析发现delta(δ)频带下EP脑网络连接显著增加,网络传输效率与小世界属性降低,分类效果最好。进一步将特征进行频带内与频带间交叉组合,并应用粒子群优化算法(particle swarm optimization,PSO)和引入遗传算法(genetic algorithm,GA)的改进算法GA-PSO筛选特征组合,研究发现GA-PSO算法优于PSO算法,EP识别准确率更高,且在四频带交叉下达到0.933 5。结果表明,引入GA算法使得GA-PSO算法保持种群多样性,避免出现过早收敛从而陷入局部最优,可提高种群的搜索能力,从而快速、稳定地筛选得到最优特征组合,有效对EP脑电信号进行识别。

关键词: 癫痫, 脑电, 复杂网络, 特征选择, 粒子群算法, 支持向量机

Abstract: Epilepsy(EP) is a chronic brain neurological disease with abnormal discharge of brain neurons. Clinically, most professional neurologists perform manual visual detection and analysis on electroencephalogram(EEG) signals, which is easily influenced by individual subjective opinion. Therefore, it is particularly significant to select effective EEG features for automatic recognition of epileptic seizure. The phase synchronization(PSI) brain network is constructed for the multi-channel EEG signals, and 15 topological features are extracted from the perspective of network nodes and structure functions, and support vector machine(SVM) classifier is used to realize the automatic recognition of epileptic EEG. It is found that the connection of EP brain network increases significantly in the delta (δ) frequency band, the network transmission efficiency and small world attribute decrease, and the classification performance is the best. The features are further combined with intra-band and inter-band, then the particle swarm optimization(PSO) and improved algorithm GA-PSO (with genetic algorithm(GA) introduced) are applied to determine the feature combinations. It is found that the GA-PSO algorithm performs better than PSO with a higher recognition accuracy of 0.933 5 under the four cross-frequency. The results show that the introduction of GA algorithm makes the GA-PSO algorithm maintain the diversity of the population, avoid premature convergence and fall into local optimization, and improve the search ability of the population, so as to quickly and stably screen the optimal feature combination, which can effectively recognize EEG signals of epilepsy patients.

Key words: epilepsy, electroencephalogram, complex network, feature selection, particle swarm optimization, support vector machine