SsvEEGc: An efficient EEG clustering method using balance-constrained voting mechanism. (March 2023)

• Record Type: Journal Article
• Authors: Li, Guanghui; Li, Dong; Dai, Chenglong
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: Partially labeled EEG clustering -- Voting mechanism -- Balance constraint -- Theoretical clustering property; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104539
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: With the increasing amount of unlabeled electroencephalography (EEG) trials in the brain–computer interface (BCI) domain, neurocognitive disorder diagnoses, and rehabilitation, etc., supervised EEG analysis that requires completely prior information of EEG labels has become unpractical. Although a few methods for unlabeled EEG clustering have recently emerged, their models or objective functions are pretty complex and time-consuming, which degrades the practicability in BCI or disease diagnosis that demands high efficiency of methods. To improve the performance of analyzing and clustering EEG trials for BCI applications, we simultaneously consider (1) the efficiency and efficacy, (2) the clustering balance and (3) the theoretical property satisfiability for partially labeled EEG trials in this paper. We propose an easily implemented and efficient EEG clustering method called ssvEEGc for partially labeled EEG trials by designing a high-efficiency and effective voting mechanism with clustering balance constraints. ssvEEGc satisfies 2 of 3 clustering theoretical properties and yields high-quality EEG clusters with low time consumption. Further, comprehensive experiments on 15 real-world EEG datasets demonstrate that ssvEEGc is superior to 11 state-of-the-art unsupervised clustering methods in terms of efficiency, effectiveness and theoretical property satisfiability. The results indicate that ssvEEGc can be applied to larger EEG datasets in biomedical signal processing and control application scenarios. Highlights: We design an efficiently effective voting mechanism for partially labeled EEG. We propose an easily implemented algorithm to efficiently yield good EEG clusters. We evaluate our method by comparing it with 11 SOTA methods on 15 EEG datasets. We analyze the satisfiability of our method on 3 theoretical clustering properties.