In this paper, we explore the use of multiscale bubble entropy and power metric for feature extraction procedure and extend it with MLA and stability analysis to design a reliable multichannel seizure detection technique. First, we represent the multichannel EEG signal in 2D matrix form and then apply AM FM model to exploit the decomposed form of EEG. Thereafter, we construct the complexity coefficient using multiscale bubble entropy analysis from decomposed EEG wave. Then, second feature set is formed by using simple and efficient power procedure to obtain absolute power index and relative power index. Using two machine learning approaches, classification performance of proposed approach is explored to correctly identify the epileptic seizures. To show the robustness of multiscale bubble entropy, the stability analysis is performed with normal EEG dataset. Experimental results demonstrate that our proposed technique can effectively detect the epileptic seizures and achieve a superior classification performance with the ANN classifier compared to KNN classifier. This method provides higher discriminating capability with greater stability, so that they could detect wider range of seizure and thus help advance the current diagnosis system.

在本文中，我们探索使用多尺度气泡熵和功率度量进行特征提取过程，并通过 MLA 和稳定性分析对其进行扩展，以设计可靠的多通道癫痫检测技术。首先，我们以 2D 矩阵形式表示多通道 EEG 信号，然后应用 AM FM 模型来利用 EEG 的分解形式。此后，我们使用分解脑电波的多尺度气泡熵分析来构造复杂性系数。然后，通过使用简单且高效的功率过程来形成第二特征集，以获得绝对功率指数和相对功率指数。使用两种机器学习方法，探索所提出方法的分类性能，以正确识别癫痫发作。为了显示多尺度气泡熵的鲁棒性，使用正常脑电图数据集进行稳定性分析。实验结果表明，与 KNN 分类器相比，我们提出的技术可以有效地检测癫痫发作，并使用 ANN 分类器实现更好的分类性能。该方法提供了更高的辨别能力和更高的稳定性，从而可以检测到更广泛的癫痫发作，从而有助于推进当前的诊断系统。