Most of the BCI papers that used Canonical Correlation Analysis (CCA) technique, applied it only on SSVEP data. Why is it so? if not are there any papers that used CCA for MI or other oscillations?
Andrew Li has used Canonical analysis on justice data: Li, A., & Bagger, J. (2007). The Balanced Inventory of Desirable Responding (BIDR) A Reliability Generalization Study. Educational and Psychological Measurement, 67(3), 525-544.
To the best of my knowledge, CCA is currently applied to SSVEP BCI. One main reason for this is that when CCA is used, some templates(e.g., 12 Hz signals and 13 Hz signals for SSVEP) first need to be built offline. Then, the collected EEG data online were matched with the templates to classify them into the corresponding class. For MI, imaging moving left or right hands cannot produce brain signals with different range of frequency. That is, such templates cannot be built. To better understand this issue, you need to figure out CCA and MI.
Yes, the CCA approach cannot be transfered directly from SSVEP to MI,
However, we could successfully use CCA for spatial filtering of EEG to improve classification of P300, error-related potentials, and other evoked potentials in BCI.
Further, we have also applied CCA for regression problems in decoding wirst movement trajectory from ECoG data.
Article Spatial Filtering Based on Canonical Correlation Analysis fo...
Conference Paper Predicting Wrist Movement Trajectory from Ipsilesional ECoG ...