The study of functional MRI (fMRI) data is increasingly performed after mapping from volumetric voxels to surface vertices. Processing pipelines commonly used to achieve this mapping produce meshes with uneven vertex spacing, with closer neighbours in sulci compared to gyri. Consequently, correlations between the fMRI time series of neighbouring sulcal vertices are stronger than expected. However, the causes, extent, and impacts of this “gyral bias” are not completely understood or widely appreciated. We explain the origins of this bias, and using in-silico models of fMRI data, illustrate how it leads to spurious results and leakage of anatomical cortical folding information into fMRI time series. We show that many common analyses can be affected by this bias, including test-retest reliability, fingerprinting, functional parcellations, and regional homogeneity. The recently developed onavg template partly reduces the bias but has relatively high residual variability in vertex spacing when projected to participant-specific surfaces. Finally, we outline recommendations to avoid or remedy the gyral bias.

Aligning brain maps using functional features rather than anatomical landmarks potentially improves individual identifiability and increases power in group neuroimaging studies. However, alignment based purely on functional magnetic resonance imaging (fMRI) risks omitting useful anatomical constraints. An optimized combination of anatomical and functional feature alignment could balance the advantages of each approach. We used 3T fMRI data from 80 Human Connectome Project participants during seven tasks. The effectiveness of functional and anatomical alignment methods was evaluated using interparticipant decoding accuracy. Functional alignment mapped vertices from participants to a template, aligning their fMRI responses to shared responses during movie viewing. The template was derived from the combined fMRI responses of a set of participants. We benchmarked the results against existing functional alignment methods, including the Procrustes method and ridge regression. A common practice in the field is to use the same participants for the alignment cohort and for template generation. We found that this inflates decoding accuracies by mixing anatomical and functional alignment. Based on this, we recommend that a template’s generalizability should be evaluated against held-out participants. Building on these findings, we investigated whether inter-subject alignment could be improved by integrating anatomical and functional information. We studied a modified alignment method where a single parameter interpolates between pure functional alignment and anatomical alignment. Optimizing the parameter with nested cross-validation, we found that integrating anatomical and functional information robustly reduced noise and improved alignment across a variety of alignment methods. Combining anatomical and functional information accounts for individual heterogeneity in functional topographies while incorporating anatomical constraints. The integrated alignment described here improves inter-subject decoding using functional brain maps. These findings also demonstrate that brain anatomy provides a lens into the inherent variability of individual neural landscapes.