Related Experiment Video
Updated: Aug 23, 2026
Probing the Brain in Autism Using fMRI and Diffusion Tensor Imaging
Published on: September 12, 2011
Autism Spectrum Disorder (ASD) Through the Lens of Hidden Markov Models (HMMs) Applied to Resting-State fMRI
Mohamed Eltalkhawy1, Amro K Barakat1, Omar Ahmed Abdelaal1
1Mansoura University Hospitals, Mansoura, Egypt.
Purpose:
Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by atypical interactions between brain networks. These interactions and the resulting dynamic shifts are effectively captured by Hidden Markov Models (HMMs) in comparison to static functional connectivity (FC)-based approaches. This review aims to meta-analyze the application of HMM to resting-state fMRI (rs-fMRI) in individuals with ASD.
Methods:
A systematic search of PubMed, Scopus, and Web of Science was conducted in May 2025. Screening followed PRISMA 2020 guidelines, with independent review and consensus resolution. Eligible studies were peer-reviewed, English-language publications that have applied HMMs to rs-fMRI in ASD, with either classification performance or state-metric outcomes reported.
Results:
Seven studies met eligibility criteria. HMM-derived metrics demonstrated diagnostic utility, with a pooled log odds ratio (log OR) of 2.86 (95% CI: 1.74-3.98; z = 5.01, p < 0.001; I² = 92%) and a pooled AUC of 0.85 (95% CI: 0.72-0.98; I² = 96.9%). Substantial heterogeneity exists across both accuracy outcomes. Relative to typically developing controls, individuals with ASD showed markedly reduced mean lifetime (MLT) in default mode network (DMN)-associated states (pooled Hedges' g = -4.19; 95% CI: -5.52 to -2.85; I² = 98%) and prolonged MLT in sensory/attention hyperactivation states (g = 3.80; 95% CI: 3.43-4.16; I² = 65%), the latter rated as high certainty evidence. Fractional occupancy (FO) in DMN states was also substantially reduced (g = -6.22; 95% CI: -9.87 to -2.58; I² = 99.6%), though this outcome was rated low certainty. Narrative synthesis across consistently identified reduced FO and MLT in DMN-hypersynchrony states alongside increased occupancy in sensory-motor and attention states, replicated across two studies despite variation in atlas choice, number of HMM states, and participant samples. HMM-derived metrics were significantly negatively correlated with ADOS scores (pooled r = -0.20; 95% CI: -0.27 to -0.13; I² = 0%; p < 0.001), rated as high certainty evidence. Transition probability analyses, reported narratively due to incompatible state taxonomies, indicated reduced transitions from sensory-related to DMN-related states and increased self-transitions in sensory-motor states in ASD. Overall risk of bias was low to moderate, with incomplete confounder adjustment being the most common limitation.
Conclusions:
ASD involves rigid, imbalanced temporal dynamics, with reduced engagement of integrative networks and dominance of sensory states. Leveraging its high diagnostic accuracy, HMMs capture these alterations and hold promise for mechanistic insight and personalized diagnostics, though heterogeneity remains a challenge.
Review Registration:
PROSPERO CRD420251057196.
