Functional connectivity recovers canine-human network homology along the evolutionary cortical hierarchy

Mapping functional brain organization between species is a prerequisite for translating connectivity biomarkers from animal models to patients, but connectivity-based correspondence has been demonstrated almost exclusively in primates. We tested whether canine-human network homology can be recovered from functional connectivity alone, using only five network-level anatomical priors, and benchmarked a spectral manifold alignment framework (Laplacian-eigenmap embedding with Procrustes alignment on network centroids) against a simple connectivity-signature matching baseline. Data comprised resting-state fMRI from 15 canines with endovascular permanent middle cerebral artery occlusion (pre-occlusion baseline and hyperacute post-occlusion sessions) and two public OpenNeuro datasets: 29 chronic-phase post-stroke aphasia patients scanned before and after rehabilitation (ds003999) and 10 healthy controls (ds000224). Held-out canine-human network homologies were recovered well above chance (top-5 network correspondence up to 54.5% vs. permutation null $\approx$ 18%, $p=0.002$ ), and correspondence followed the evolutionary cortical hierarchy: visual and somatomotor homologies were recovered several-fold above chance, default-mode only weakly-extending a gradient previously described in primates to a carnivore. Notably, simple signature matching (54.5%) outperformed manifold alignment (35.5%), indicating that the homology signal is strong enough to be captured without embedding machinery at translational cohort sizes. Within species, canine network metrics showed no treatment-arm differences (all $p>0.3$ ), and chronic-phase patients did not differ from controls in modularity ( $p=0.93$ ); a sensitivity analysis bounds these nulls. All results are fully reproducible from shared code and public data.