Joint Blind Source Separation for Neurophysiological Data Analysis: Multiset and multimodal methods

Conventional blind source separation (BSS) methods have become widely adopted tools for neurophysiological data analysis. However, the increasing availability of multiset and multimodal neurophysiological data has posed new challenges for BSS methods originally designed to analyze one data set at a time. Concomitantly, there is growing recognition that joint analysis of neurophysiological data has the potential to substantially enhance our understanding of brain function by extracting information from complementary modalities and synergistically combining the results. Therefore, joint data analysis methods represent both a challenge and an opportunity for the neurophysiological signal processing community that attempts to enhance understanding of normal brain function and the pathophysiology of many brain diseases. Over the past decade, various joint blind source separation (JBSS) methods have been proposed to simultaneously accommodate multiple data sets. In this article, we provide an overview and taxonomy of representative JBSS methods. We show, through illustrative numerical simulations, that different statistical assumptions and tradeoffs underlie different JBSS methods, affecting which method should be ideally chosen for a given application. We then discuss several real-world neurophysiological applications from both multiset and multimodal perspectives, highlighting the benefits of the JBSS methods as effective and promising tools for neurophysiological data analysis. Finally, we discuss remaining challenges for future JBSS development.