Spatial-Temporal Characteristics of Brain Activity in Autism Spectrum Disorder Based on Hidden Markov Model and Dynamic Graph Theory: A Resting-State fMRI Study.

Qian and colleagues scanned 60 people with ASD and 60 matched controls during quiet rest. They used a Hidden Markov Model to spot repeating whole-brain patterns that last only seconds. Then they tracked how long each group stayed in each pattern and how the networks talked to one another.

The team focused on three key nets: the default mode (day-dream), ventral attention (surprise), and visual (sight) networks. They asked: do ASD brains switch states the same way as typical brains?

ASD brains lingered far longer in one ultra-connected state. In this state the three target networks were cut off from the rest of the brain like islands. Controls moved in and out of this state quickly; ASD participants stayed there about twice as long.

The longer the brain stayed in this isolated state, the higher the child’s ADOS score. The pattern gave 82 % accuracy when used alone to sort ASD from typical.

Kovačič et al. (2020) saw similar disconnections but used older static scans. Qian’s 2024 work shows the same problem is dynamic — it is the stubborn停留 time, not just the wiring, that marks ASD.

Li et al. (2024) used the same math and found girls with ASD drive the global efficiency changes. Qian did not split by sex, so the new long-state signature could be girl-heavy; clinicians should watch for sex effects before using the marker.

Guo et al. (2023) also found two ASD sub-types with different salience-network rhythms. Qian’s single long-state might hide these sub-groups; combining both methods could give cleaner patient clusters.

You now have a five-minute resting scan that flags ASD with four-fifths accuracy. While you can’t run fMRI in clinic, you can ask referring physicians for the report: “long isolated state time” predicts rigid behavior and social scores. Pair this biomarker with sex-split data and salience-subtyping to sharpen early referral and track change during ABA. Push for pre- and post-treatment scans in your next collaboration; if state dwell time drops as skills grow, you have neural proof your program works.