基于视角-规则的深度TSK模糊分类器及其在多元癫痫脑电信号识别中的应用

在癫痫脑电信号分类检测中,传统机器学习方法分类效果不理想,深度学习模型虽然具有较好的特征学习优势,但其“黑盒”学习方式不具备可解释性,不能很好地应用于临床辅助诊断;并且,现有的多视角深度TSK模糊系统难以有效表征各视角特征之间的相关性.针对以上问题,提出一种基于视角-规则的深度Takagi-SugenoKang (TSK)模糊分类器(view-to-rule Takagi-Sugeno-Kang fuzzy classifier, VR-TSK-FC),并将其应用于多元癫痫脑电信号检测中.该算法在原始数据上构建前件规则以保证模型的可解释性,利用一维卷积神经网络(1-dimensional convolutional neural network, 1D-CNN)从多角度抓取多元脑电信号深度特征.每个模糊规则的后件部分分别采用一个视角的脑电信号深度特征作为其后件变量,视角-规则的学习方式提高了VR-TSK-FC表征能力.在Bonn和CHB-MIT数据集上, VR-TSK-FC算法模糊逻辑推理过程保证可解释的基础上达到了较好分类效果.

Recognition of multivariate epilepsy EEG signals based on view-to-rule deep TSK fuzzy classifier

Traditional machine learning methods perform poorly in classification and detection epilepsy electroencephalogram(EEG) signals, while the state-of-the-art deep learning models show excellent predictive performance due to their powerful feature abstraction capabilities, but their behavior is black-box, leading to uninterpretable and not well suited for clinical diagnosis. Moreover, the existing multi-view deep TSK fuzzy system is difficult to effectively represent the correlation between the features of each view. To address the problems above, in this paper, we propose a view-to-rule deep TSK fuzzy classifier, i.e., VR-TSK-FC, and apply it to multivariate epilepsy EEG signal detection. The proposed classifier constructs antecedent-part of fuzzy rules on the original data to ensure interpretability, and the one-dimensional convolutional neural network(1D-CNN) learns deep features of multivariate EEG signals from multi-view. The consequent-part of each fuzzy rule adopts the EEG signal deep feature of each view as its consequent-part variable, and the fuzzy-deep view-to-rule learning method improves the representation ability of the proposed VR-TSK-FC. Experiments on the Bonn and CHB-MIT datasets demonstrate that, the fuzzy logic inference process of the proposed VR-TSK-FC achieves better classification results as well as concise interpretability.