Divide and Conquer: Sub-Grouping of ASD Improves ASD Detection Based on Brain Morphometry.

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Biostatistics

Background:

• Autism spectrum disorder (ASD) classification using brain morphometry has shown limited success (<60%) and inconsistent findings, often attributed to significant heterogeneity within the ASD population.
• Previous studies on brain morphometric abnormalities in ASD have yielded conflicting results, hindering the development of reliable diagnostic biomarkers.

Purpose of the Study:

• To investigate the heterogeneity of brain morphometry in ASD and demonstrate how incorporating clinical variables can improve classification accuracy.
• To explore the potential of sub-grouping ASD individuals based on autism severity (AS), verbal IQ (VIQ), and age for identifying more robust brain-based biomarkers.
• To assess the utility of structural MRI (sMRI) features, optimized for specific subgroups, for early detection and monitoring of ASD.

Main Methods:

• Structural MRI (sMRI) data from 734 male participants (361 with ASD, 373 controls) in the Autism Brain Imaging Data Exchange (ABIDE) dataset were processed using FreeSurfer to extract morphometric features.
• Machine learning classifiers, including Random Forest and Gradient Boosting, were employed to classify ASD versus controls.
• Participants were sub-grouped based on AS, VIQ, and age, and classification performance (Area Under the Curve - AUC) was evaluated with and without the inclusion of these clinical variables.

Main Results:

• Initial classification using only morphometric features achieved an AUC of 0.61. Incorporating VIQ and age improved the AUC to 0.68.
• Sub-grouping significantly enhanced classification, with the highest AUC of 0.92 achieved in the low AS subgroup (AS = 4-5) when subjects were matched on age and/or VIQ.
• Classification performance varied across subgroups, decreasing with higher AS and VIQ, and being lowest in the mid-age group (13-18 years). Important features varied by AS, VIQ, and age, with frontal and temporal features prominent in younger subjects.

Conclusions:

• ASD brain morphometry is highly heterogeneous, and accounting for this heterogeneity using clinical variables like AS, VIQ, and age is crucial for improving classification accuracy.
• Identifying ASD biomarkers within specific subgroups, rather than across the entire spectrum, yields more robust and insightful results.
• Subgroup-specific biomarkers, particularly those identified in younger individuals, hold promise for optimizing early detection and monitoring of ASD, enhancing the clinical utility of sMRI.