癫痫发作瞬态带宽特征自适应检测方法

癫痫是常见的神经系统疾病之一。癫痫发作的识别通常采用脑电测量记录中的癫痫发作起始点,以辅助医生进行诊断并对患者的发作状态报警。利用脑电信号的瞬态参数提出了一种自适应带宽特征,可用于提高癫痫发作检测精度。首先,利用经验模态分解(EMD)求得脑电信号的本征模态函数(IMF),并计算特定阶次IMF的解析信号;其次,利用该解析信号求解瞬时幅值与瞬时频率,对EEG信号的带宽特征添加权重,得到可用于癫痫检测的自适应带宽特征(Adaptive Bandwidth);最后,利用该特征完成癫痫发作检测。采用长达118 h 49 min的癫痫患者临床脑电数据进行实验,实验结果表明,自适应带宽特征的敏感性、特异性、准确性参数均比原特征取得明显提高。自适应带宽特征可提高癫痫发作检测精度并降低时间延迟,便于及时采取治疗措施,为临床检测提供了重要依据。

Adaptive detection method based on instantaneous bandwidth feature for seizure onset

Seizure is one of the most common neurological system diseases. Seizure onset is usually identified by the starting point of seizure onset in the EEG measurement record, which can help doctor to conduct the seizure diagnosis and the alarm of the patient state. In order to improve the performance of seizure onset detection algorithm, this paper proposes an adaptive bandwidth feature based on EEG signal instantaneous parameters, which can be used to improve seizure onset detection accuracy. Firstly, the intrinsic mode functions (IMFs) of the EEG signal is calculated using empirical mode decomposition (EMD). Then, the Hilbert transform is conducted on a specific order IMF to get the analytical signal. The analytical signal is used to calculate the instantaneous amplitude and frequency. The weights are introduced in the bandwidth features of the EEG signal, and the adaptive bandwidth features used for seizure detection are obtained. Finally, the seizure onset detection is achieved using these adaptive bandwidth features. In order to verify the proposed method, the clinic EEG data of the epilepsy patients with a length of 118 hours and 49 minutes were used to conduct experiment, the experiment results show that the sensitivity, specificity and accuracy of the adaptive bandwidth features are improved obviously compared with the original features; the adaptive bandwidth features improve the seizure onset detection accuracy and reduce the delay time, so that timely treatment can be taken, and the proposed method provides an significant basis for seizure clinic detection.