FastSRM: A fast, memory efficient and identifiable implementation of the shared response model

Abstract

The shared response model (SRM) provides a simple but effective framework to analyze fMRI data of subjects exposed to naturalistic stimuli. However, when the number of subjects or runs is large, fitting the model requires a large amount of memory and computational power, which limits its use in practice. Furthermore, SRM is not identifiable, which makes the shared response difficult to interpret. In this work, we implement an identifiable version of SRM and show on real data that it improves the stability of the recovered shared response. We then introduce FastSRM, which relies on a dimension reduction step and yields the same solution as the original algorithm. We show experimentally using synthetic and real fMRI data that FastSRM is considerably faster and more memory efficient than current implementations. The experiments performed in this article are fully reproducible: our code available at https://github.com/hugorichard/FastSRM allows you to download the data, run the experiments and plot the figures. For a more pleasant reading of the math equations, we advise the reader to read the version of this article formatted with latex:https://github.com/hugorichard/FastSRM/blob/master/paper/main.pdf.

Individual Brain Charting dataset extension, third release for movie watching and retinotopy data

“…Nevertheless, one disadvantage that arises from ISC is the fact it misses correspondences that are not accurate at the voxel level. We used the FastSRM implementation for fMRI data–as described in Richard&Thirion 34 –to analyze the preprocessed imaging data of Clips and Raiders and extract a common subjects' response to these tasks together with their individual spatial components. These individual spatial components thus refer to the subject-specific effects-of-interest elicited by the tasks.…”

“…We also present their technical validation, using Quality-Assurance (QA) metrics that show functional activation to be a direct response to behavior. Additionally and given the aforementioned challenges posed by the analysis of fMRI data relative to naturalistic paradigms, we complement the technical validation of the Clips and Raiders tasks with an application of the Fast Shared Response Model (FastSRM), as described in Richard&Thirion 34 . To this end, we showcase how to extract networks linked to mental functions elicited by these tasks.…”

“…Concretely, we demonstrate that cognitive networks of functional data collected from naturalistic paradigms can be extracted using FastSRM—an unsupervised data-driven method—without explicitly modelling features of the stimuli. Due to various shortcuts described in Richard&Thirion (2023) 34 , the fast and memory-efficient implementation of FastSRM becomes very useful under high-dimensional regimes, such as modelling naturalistic stimuli. In addition, we demonstrate that results obtained for every task–which are adapted from previous studies–are in agreement with the ones originally reported.…”