Do similar network patterns emerge in autistic and typical participants?

This small pilot compares resting-state fMRI from 72 autistic participants with the existing 96 NYU control participants in ABIDE I. Each cohort is fitted independently using the same cortical atlas and rank-12 settings. The question is whether similar spatial components emerge despite learning a different predictive basis in each group.

Main findings

Several early components are closely shared. Control Networks 2–5 have their strongest matrix-entry correlation with the same-numbered ASD network: N2: r = 0.98, N3: r = 0.92, N4: r = 0.90, N5: r = 0.92. This supports substantial resemblance in the spatial organization recovered from the two cohorts.
Reversed orientation also preserves spatial resemblance. Networks 6, 7, 8, 11 have strong negative best-match correlations. Their coefficient patterns resemble an oppositely oriented counterpart; signs alone are not a diagnostic difference.
Ordering is not exact across independent fits. Control Network 9 best matches ASD Network 10, and control Network 10 best matches ASD Network 9. The comparison identifies correspondences without changing either model.

Inspect the cohort comparison · Explore the ASD functional systems

Functional anatomy within the ASD fit

Network 4 has a visual association. Its negative pole resembles this system (similarity 0.84; 70% of that pole’s weight). Its opposite pole resembles salience / ventral attention (similarity 0.38; 18% of that pole’s weight).
Network 3 has a visual association. Its positive pole resembles this system (similarity 0.76; 33% of that pole’s weight). Its opposite pole resembles default mode (similarity 0.61; 74% of that pole’s weight).
Network 2 has a default mode association. Its positive pole resembles this system (similarity 0.65; 47% of that pole’s weight). Its opposite pole resembles salience / ventral attention (similarity 0.43; 26% of that pole’s weight).
These are the strongest qualifying similarities from this ASD fit, with at most one entry per network. Network numbers are specific to this fitted basis; they do not identify the corresponding control networks or establish an ASD–control difference.

Inspect the ASD matching table and atlas

What this pilot establishes

Independently fitted cohort summaries recover several similar spatial patterns. This is a starting point for studying shared organization and possible differences, rather than evidence that autism changes any particular network. The groups are not matched for age, sex or motion; estimation stability, held-out replication and participant-level group comparisons remain to be assessed. Temporal dynamics have not been compared here.