Sensors, Journal Year: 2024, Volume and Issue: 24(15), P. 4920 - 4920

Published: July 29, 2024

Electroencephalography (EEG)-based applications in brain-computer interfaces (BCIs), neurological disease diagnosis, rehabilitation, etc., rely on supervised approaches such as classification that requires given labels. However, with the ever-increasing amount of EEG data, incomplete or incorrectly labeled unlabeled data are increasing. It likely degrades performance approaches. In this work, we put forward a novel unsupervised exploratory analysis solution by clustering based low-dimensional prototypes latent space associated respective clusters. Having prototype baseline each cluster, compositive similarity is defined to act critic function clustering, which incorporates similarities three levels. The approach implemented Generative Adversarial Network (GAN), termed W-SLOGAN, extending Stein Latent Optimization for GANs (SLOGAN). Gaussian Mixture Model (GMM) utilized distribution adapt diversity signal patterns. W-SLOGAN ensures images generated from component belong cluster. adaptively learned mixing coefficients make model remain effective dealing an imbalanced dataset. By applying proposed two public intracranial (iEEG) epilepsy datasets, our experiments demonstrate results close data. Moreover, present several findings were discovered intra-class and cross-analysis classification. They show attractive practice diagnosis epileptic subtype, multiple labelling etc.

Language: Английский