Characterising spatio-spectral subject differences in fMRI using KME of distributions

CECILIA JARNE; DIEGO VIDAURRE

Montreal

Conferencia; 29th Annual Meeting of the Organization for Human Brain Mapping; 2023

Organization for Human Brain Mapping

Building on the machinery of kernel embeddings, we considered the Maximum Mean Discrepancy or MMD, a distance metric defined on the space of probability measures, based on the frequency content of signal recorded in fMRI of the different brain regions. MMD can take advantage of the so-called kernel trick, which allows us to compute and use principled metrics across many data domains. Specifically, we used the frequency content of the signals to build probability distributions considering the complete time series; this way, we were able to predict subject differences in various phenotypes and demographic traits. Here, for simplicity, we focused on age.