Effectiveness of Deep Convolutional Autoencoder Clustering in Generating EEG Microstates
1 Technological University Dublin (TU Dublin)
EEG microstates are quasi-stable states with a few distinct scalp potential topographies. These microstate maps capture the spatiotemporal aspects of EEG brain signals and are formed by clustering. However, the choice of the clustering algorithm to generate the template maps is arbitrary. Available microstate tools such as Cartool and EEGLAB limit to a few shallow clustering methods, thereby making a large number of remaining clustering methods unexplored. Shallow clustering algorithms are proven inefficient when the data does not conform to the standard template. For example, K-means are suboptimal for clustering if the data is not in a convex shape. Moreover, research into the potential use of deep learning techniques to generate EEG microstates needs to be initiated. Recent research has shown that deep clustering can improve cluster quality over shallow clustering in domains such as computer vision and natural language processing involving images and text data. We assess the effectiveness of autoencoder-based deep clustering on EEG microstates. The results are evaluated in two distinct phases: assessing clustering effectiveness and analysing the stability of the generated microstate sequences. Clustering effectiveness is quantified using internal clustering measures, such as the Silhouette score, Davies-Bouldin Index, and Calinski-Harabasz Index. The stability of the microstates is assessed through the duration parameter. Stability indicates how consistently each microstate is maintained over time, providing insights into the temporal stability of the microstate sequences. Our work contributes to the body of knowledge by introducing deep clustering as a potential alternative to shallow clustering in the formation of microstates.