Common Spatial Pattern (CSP) Interpretation

💬 Support & Discussions
, , , ,
1

Hello everyone!

I am working with 33 channels of EEG data with 25 epochs belonging to class 1 and 25 epochs belonging to class 2. I want to maximize the EEG signal difference between the two classes.
After running CSP, I get the spatial filter matrix of dimensions (33 x 33), which I transpose and multiply with each trial of my EEG matrix (according to Blankertz et al. 2008).
I am treating CSP as a filter to enhance the EEG signal difference between the two classes and then perform subsequent analysis (ERP, PSD etc.) and interpretation. Is that a viable approach? Will the application of the CSP filter change the interpretation of the signals in sensor space?

I am just starting out with CSP so I apologize if the question is very simple.
Thank you in advance!

2

It depends a bit on what you want to do. Generally CSP is a useful tool to maximize classification performance, but you should not apply it to your data if you want to interpret ERPs or PSDs and compare them to other results from the literature, where typically no CSP filtering is applied.