Bids-compliant connectivity matrix dimensionality reduction transforms?

,
1

I have a brain connectivity analysis tool that leverages dimensionality reducing transformations based on HCP-YA and other HCP-related studies. For example a PCA transformation to reduce a set of 500x500 FC matrices to 256 components. I only use the upper triangular portion, so the transform contains 124750x256 values, which is around 128MB (float32 stored in .npy currently). I have a few hundred such transforms using different parcellations and different connectivity estimation techniques. I am interested in putting these on openneuro as a derivatives dataset.

Is there any documentation or discussion about how such a dataset might be structured?

2

Hey @kjamison,

there is quite some work going concerning such data in the BIDS Connectivity Project, including atlases, Relationship matrices (including connectivity) and dimensionality reduction-based networks. There are also discussions regarding file types, e.g. here and here.

You could have a look at those, maybe briefly evaluate if the proposals would work for/cover your use case and share your feedback if possible. The respective folks and discussions are definitely open to it.

HTH, Cheers, Peer

1 Like