Autoreject computation time

Support & Discussions
1

I’m currently analysing MEG data from over 200 subjects. My goal is to identify and reject bad channels using autoreject, perform ICA to remove blink-related components, and then reject bad epochs based on peak-to-peak amplitude criteria for both gradiometers and magnetometers.

However, when testing this autoreject → ICA pipeline on two subjects, the computation time was extremely long. Could you suggest a more efficient or scalable approach — for example, by adjusting autoreject parameters or using RANSAC instead? I’m particularly interested in whether RANSAC would be less computationally demanding while providing comparable results.

Thanks a lot for your help!

image
image1268×954 122 KB

image
image956×336 37.3 KB

2

Hi,

Sorry for the slow response. You can try to use autoreject (global), that get_rejection_threshold function that returns a rejection dictionary and can be provided to the ICA functions in MNE.

Hope that helps,
Mainak