Epileptic EEG Detection Using a Multi-View Fuzzy Clustering Algorithm with Multi-Medoid

Qianyi Zhan,Wei Hu,Yuan Liu,Yizhang Jiang,Kaijian Xia,Huangxing Lin,Jing Xue

doi:10.1109/access.2019.2947689

Abstract

Using clustering algorithms to automatically analyze EEGs of patients and to identify the characteristic waves of epilepsy is of high clinical value. Traditional clustering algorithms mostly use a calculated virtual single representative medoid point to describe the cluster structure, but this single representative medoid point has insufficient information. To accurately capture more accurate intracluster structural information, a representative multi-medoid points strategy is adopted, which describes the cluster structure by assigning representative weights to each sample in the cluster. Considering that the multi-view learning mechanism combines information from each view to improve the algorithm's clustering performance, a multi-view fuzzy clustering algorithm with multi-medoid (MvFMMdd) is proposed. This algorithm discards the approach of the traditional fuzzy clustering algorithm, which uses a single virtual representative point to characterize the cluster structure, and uses several real representative points to describe the cluster structure. Experiments verify the medical significance of the proposed algorithm.

Highlights

Epilepsy, a disease of the brain, causes dysfunction in consciousness, sensation, movement and mentality, causing substantial pain and serious physical and mental damage to the patient [1]
(3) Experiments verify the effectiveness of the clustering performance of the multi-view fuzzy clustering algorithm with multi-medoid (MvFMMdd) algorithm; the strategy of expressing the cluster structure with real representative multi-medoid points is more relevant for medical data
MULTI-VIEW FUZZY CLUSTERING ALGORITHM WITH MULTI-MEDOID Considering the multi-view learning mechanism mentioned above and the multi-medoid representative point expression strategy in related work, a multi-view fuzzy clustering algorithm with multi-medoid (MvFMMdd) is given

Summary

INTRODUCTION

A disease of the brain, causes dysfunction in consciousness, sensation, movement and mentality, causing substantial pain and serious physical and mental damage to the patient [1]. The fuzzy clustering algorithm automatically diagnoses epilepsy by classifying normal and seizure EEG signals. To accurately capture more accurate intracluster structural information, a cluster structure expression strategy based on multiple representative medoid points is introduced. (3) Experiments verify the effectiveness of the clustering performance of the MvFMMdd algorithm; the strategy of expressing the cluster structure with real representative multi-medoid points is more relevant for medical data. The schematic diagram of the strategy is as follows: FIGURE 1 shows a sample graph of the virtual singlemedoid representative point expression cluster structure. MULTI-VIEW FUZZY CLUSTERING ALGORITHM WITH MULTI-MEDOID Considering the multi-view learning mechanism mentioned above and the multi-medoid representative point expression strategy in related work, a multi-view fuzzy clustering algorithm with multi-medoid (MvFMMdd) is given. Substituting Eq (8) and (9) into Eq (5), the expression of uci,r is as follows

The integration strategy for the global division result is as follows

Output the global fuzzy partition matrix U

CONCLUSION