Common temporal component when do spatial canICA?

In group ICA, when do spatial ICA, apart from extracted spatial component there are also corresponding temporal component representing some common / average time courses presenting in all subjects. Is there a way to get similar temporal component in canICA in nilearn? (I didn’t find it)

From this paper comparing different ICA versions, probably CanICA can only give spatial maps?

Hello !

after fitting the model, the CanICA.components_ attributes will give you the spatial components while CanICA.transform(data) will project the data in the reduced representation (shape : (n_samples, n_components)), which I guess, is what you’re looking for.


1 Like

Thanks for your response! After trying out, I think this CanICA.transform(data) is related to what I need. However, the returned list has n_subject entries, each with a (n_temporal, n_component) array, which means for each subject, it has a unique temporal component corresponds to one spatial component? If I want a common temporal component among all subjects corresponding to a spatial component when do spatial CanICA, is it done by simply averaging all the temporal components?

Hi @neo

Just to add on: yes, you would expect to see a unique subject-level temporal component (time series) for each group-level spatial component (brain map). The idea is that this shared spatial component differentially contributes to each subject’s overall activity pattern.

If instead you wanted a shared group-level temporal component, you would then expect to see unique subject-level spatial components. This is temporal ICA.

Neither temporal nor spatial ICA will give you shared spatial components with a shared temporal component. This would suggest that all of your subjects had the same activity patterns !



1 Like