Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Multi-View Temporal Alignment for Cross-Species fMRI

Kurt Braunlich¹, Marianne M. Duyck², Bevil Conway³, Chris I. Baker¹; ¹National Institutes of Health, ²Académie d'Aix-Marseille, ³National Institute of Mental Health

Presenter: Kurt Braunlich

Cross-species neuroimaging is critical for linking cellular-resolution findings in non-human primates to human brain function. However, comparisons are complicated by differences in brain organization, behavior, and imaging signals. In particular, macaques are typically scanned using monocrystalline iron oxide nanoparticles (MION), which improves signal-to-noise but produces a hemodynamic response that differs from the blood-oxygen-level-dependent (BOLD) response in humans. This confounds data-driven analyses that rely on temporal alignment between species. Here, we introduce Multiview Temporal Alignment (MVTA), a multivariate method that directly aligns macaque MION signals to human BOLD. By learning the temporal transform from variance that is shared across species, MVTA avoids contamination from species-specific and idiosyncratic signals and outperforms approaches based on canonical hemodynamic response functions (HRFs) or voxel-space objectives. We first validate MVTA using simulations by comparing four pipelines for handling cross-species temporal correspondence: 1) comparisons of representational structure without temporal alignment (baseline), 2) an existing “double convolution” approach that involves convolving the observed data in each species with the canonical HRF of the other, 3) a temporal transform learned from voxel space, and 4) MVTA. MVTA outperforms all other methods across several metrics. We demonstrate how MVTA can be combined with hyperalignment to identify shared cross-species representational structure using a novel fMRI dataset in which macaques (3T, n=3) and humans (7T, n=9) viewed identical movies and functional localizers. Using human ROIs as seeds, we identified corresponding macaque regions: V1 to V1, face-selective regions to face-selective regions, and V4 overlapping ventral portion of its standard atlas definition (Jung et al., 2021). This pipeline provides a quantitative measure of cross-species representational correspondence under naturalistic stimulation. By improving temporal alignment, MVTA enables exploratory cross-species analyses that were previously infeasible.

Topic Area: Methods, Tools, Theory & Neural Coding