Resting-state brain signals reveal developmental differences within autism spectrum disorder using frequency-specific measures

A new Frontiers in Neuroscience study uncovers how resting-state brain signals, particularly slow-4 ReHo, can differentiate children from adolescents with autism, highlighting the potential for developmental insights through frequency-specific measures.

A new Frontiers in Neuroscience study suggests that resting-state brain signals may help distinguish children from adolescents within autism spectrum disorder, using frequency-specific measures that appear more informative in some bands than others. Analysing data from 251 people with autism across 10 sites in the Autism Brain Imaging Data Exchange, the researchers found that features from the slow-4 and conventional frequency ranges separated the two age groups more effectively than slow-5 features, with the strongest single-metric result coming from slow-4 regional homogeneity, or ReHo, in a logistic regression model.

The study used a careful machine-learning pipeline designed to limit leakage between training and testing. The sample was split into a training group of 200 and a held-out test group of 51, and site effects were corrected with CovBat harmonisation using only the training data. The team then extracted regional measures from 246 brain areas in the Brainnetome Atlas, applied LASSO feature selection and compared logistic regression, support vector machines and random forests. On the independent test set, the best single-model result reached an AUC of 0.811, while an exploratory model combining slow-4 ALFF and ReHo produced the top overall AUC of 0.819.

Model interpretation pointed to a distributed network rather than a single brain region. SHAP analysis linked the strongest contributions to the inferior parietal lobule, lateral occipital cortex, middle and inferior frontal gyri, basal ganglia and thalamus. The authors argue that this pattern fits a broader view of brain maturation as a system-wide process involving frontoparietal, visual and subcortical circuits. They also note that the combined ALFF and ReHo model drew on partly overlapping but not identical regions, suggesting the two measures capture complementary aspects of intrinsic brain activity.

As a contextual comparison, the researchers also examined a separate sample of 325 typically developing children and adolescents. After harmonisation and correction for sex and head motion, they found no significant child-adolescent differences across any of the three frequency bands in either ALFF or ReHo. The authors say this exploratory result does not establish a diagnosis-by-development interaction, but it does sharpen the contrast with the autism findings.

The study’s main limitation is that it is cross-sectional, so it cannot show how these brain features change within individuals over time. The models were also not meant to diagnose autism or predict clinical outcomes. Even so, the authors say frequency-specific resting-state measures, especially slow-4 ReHo, may offer a useful way to characterise developmental heterogeneity in autism and deserve testing in longitudinal and independent cohorts.