Three autism subtypes based on single-subject gray matter network revealed by semi-supervised machine learning.

Area of Science:

• Neuroscience
• Developmental Neuroscience
• Computational Neuroscience

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with significant heterogeneity.
• Understanding the neural basis of ASD heterogeneity is crucial for targeted interventions.
• Previous research often treats ASD as a monolithic condition, overlooking potential subtypes.

Purpose of the Study:

• To delineate autism spectrum disorder (ASD) subtypes based on individual gray matter brain networks.
• To provide novel insights into ASD heterogeneity using a graph theory approach.
• To explore differences in brain network properties and clinical measures among identified ASD subtypes.

Main Methods:

• Extraction and normalization of single-subject gray matter brain networks.
• Calculation of topological properties for each brain network.
• Application of the Heterogeneity through Discriminative Analysis (HYDRA) method for subtyping patients.
• Exploration of differences in network properties (global and nodal) and clinical measures (VIQ, PIQ, ADOS) across subtypes.

Main Results:

• Identification of three distinct ASD subtypes based on gray matter network properties.
• Significant differences in brain network topology were observed between subtypes, particularly involving the precentral gyrus, lingual gyrus, and middle frontal gyrus.
• Clinical differences were noted, with subtype 1 showing lower VIQ/PIQ than subtype 3, and higher ADOS scores than subtype 2.

Conclusions:

• Distinct ASD subtypes exist, characterized by unique gray matter network configurations.
• These subtypes exhibit differential brain structural properties and clinical presentations in male individuals with ASD.
• The findings offer valuable insights into the neural mechanisms underlying ASD heterogeneity and support personalized approaches.