Electroencephalogram (EEG) is a fundamental and unique tool for exploring human brain activity in general and epileptic mechanism in particular. It offers significant information about epileptic seizures source known as epileptogenic area. However, it is often complicated to detect critical changes in EEG signals by visual examination, since this signal aspect of epileptic persons seems to be normal out of the seizure. Thus, the challenge is to design such a robust and automatic system to detect these unseen changes and use them for diagnosis. In this research, we apply the Artificial Metaplasticity Multi-Layer Perceptron (AMMLP) together with discrete wavelet transform (DWT) to Bonn EEG signals for seizure detection goal. Significant features were then extracted from the well-known EEG brainwaves. Aiming to decrease the computational time and improve classification accuracy, we performed a features ranking and selection employing the Relief algorithm. The obtained AMMLP classification accuracy of 98.97% proved the effctiveness of the applied approach. Our results were compared to recent researches results on the same database, proving to be superior or at least an interesting alternative for seizures detection within EEG signals.

中文翻译：

---

改进的用于癫痫发作检测的人工神经网络

脑电图 (EEG) 是一种基本且独特的工具，用于探索人类大脑的一般活动，特别是癫痫机制。它提供了有关癫痫发作源的重要信息，称为致癫痫区。然而，通过视觉检查来检测 EEG 信号的关键变化通常很复杂，因为癫痫患者的这种信号方面在癫痫发作时似乎是正常的。因此，挑战在于设计这样一个强大且自动的系统来检测这些看不见的变化并将它们用于诊断。在这项研究中，我们将人工超塑性多层感知器 (AMMLP) 与离散小波变换 (DWT) 一起应用于波恩 EEG 信号以实现癫痫检测目标。然后从众所周知的 EEG 脑电波中提取重要特征。为了减少计算时间并提高分类精度，我们使用 Relief 算法进行了特征排序和选择。获得的 AMMLP 分类准确率为 98.97%，证明了所应用方法的有效性。我们的结果与最近在同一数据库上的研究结果进行了比较，证明对于 EEG 信号中的癫痫发作检测来说是优越的或至少是一个有趣的替代方案。