Supervised Network-Based Fuzzy Learning of EEG Signals for Alzheimer's Disease Identification

Accurate identification of Alzheimer's disease (AD) with electroencephalograph (EEG) is crucial in the clinical diagnosis of neurological disorders. However, the effectiveness and accuracy of manually labeling EEG signals are barely satisfactory, due to lacking effective biomarkers. In this paper, we propose a novel machine learning method network-based Takagi-Sugeno-Kang (N-TSK) for AD identification which employs the complex network theory and TSK fuzzy system. With the construction of functional network of AD subjects, the topological features of weighted and unweighted networks are extracted. Taken the network parameters as independent inputs, a fuzzy-system-based TSK model is established and further trained to identify AD EEG signals. Experimental results demonstrate the effectiveness of the proposed scheme in AD identification and ability of N-TSK fuzzy classifiers. The highest accuracy can achieve 97.3% for patients with closed eyes and 94.78% with open eyes. In addition, the performance of weighted N-TSK largely exceeds unweighted N-TSK. By further optimizing the network features utilized in the N-TSK fuzzy classifiers, it is found that local efficiency and clustering coefficient are the most effective factors in AD identification. This work provides a potential tool for identifying neurological disorders from the perspective of functional networks with EEG signal, especially contributing to the diagnosis and identification of AD.