Dog brain atlas generated via spatially constrained spectral clustering

Functional parcellations enable reproducible analyses of brain organization, yet the dog fMRI field still lacks a validated, multi-scale functional atlas. We applied spatially constrained spectral clustering to a resting state fMRI dataset (n = 27 dogs) to generate whole-brain parcellations spanning 20-300 parcels. Replicability was evaluated against an independent dataset (n = 20 dogs) using Dice overlap and Adjusted Rand Index (ARI) as measures of spatial overlap and similarity; within-parcel functional coherence was assessed via homogeneity. Functional parcellations were anatomically coherent across scales, with global Dice peaking at 0.625 at N = 140 and 0.623 at N = 100, while decreasing to 0.526 at N = 300. ARI peaked at 0.48 for N = 60 and remained ≥ 0.46 through N = 180. Mean within-parcel homogeneity increased monotonically from 0.08 (N = 20) to 0.19 (N = 300). Concordance with an anatomical atlas (Johnson et al., 2020) increased at the regional level (plateau ~0.16 for N ≥ 140) while diminishing at gyral and lobar levels as resolution increased, consistent with functionally driven sub-regional differentiation. Together, these results indicate functional segmentations that are replicable across datasets and internally coherent across scales, with intermediate resolutions (100-140 parcels) balancing specificity and reproducibility for common analyses. We introduce a comprehensive, multi-scale functional dog brain atlas derived from data-driven clustering, providing an open resource for comparative studies of brain evolution and canine cognition.