Multiway Canonical Correlation Analysis of Brain Signals

[jinglescode]

Paper

Link: https://www.biorxiv.org/content/10.1101/344960v1.full
Year: 2019

Summary

• CCA does not address the issue of comparing or merging responses across more than two subjects
• Multiway CCA can be applied effectively to multi-subject datasets of EEG, to denoise the data prior to further analyses, and to summarize the data and reveal traits common across the population of subjects
• MCCA-based denoising yields significantly better scores in an auditory stimulus-response classification task, and MCCA-based joint analysis of fMRI data reveals detailed subject-specific activation topographies

Methods

• interested in finding these “shared sources” and suppressing the noise
  
• MCCA finds a linear transform applicable to each data matrix within a data set to align them to common coordinates and reveal shared patterns. It can be used in several ways: as a denoising tool applicable to an individual data matrix, as a tool for dimensionality reduction, as a tool to align data matrices within a common space to allow comparisons, or as a tool to summarize data and reveal patterns that are general across data matrices.

Results

• used both to design spatial filters to denoise data of each individual subject, and to summarize data across subjects

Comments