Zenghe Yue
Over the past decade, numerous studies have used data-driven approaches (e.g., clustering and normative modelling) to discover biological subtypes of autism spectrum disorder (ASD) from MRI data. However, the reported number, definitions, and clinical correlates of these subtypes are inconsistent across studies, external validation is rare, and the field faces a credibility crisis. This project evaluates the reproducibility of neuroimaging-based ASD subtypes in three parts. (1) A systematic review and meta-analysis of all peer-reviewed original studies applying data-driven subtyping to structural or functional MRI of individuals with diagnosed ASD, pooling between-subtype effect sizes for clinical symptom severity (ADOS, SRS), IQ, age, and sex ratio. (2) A harmonised re-analysis of the public ABIDE I/II datasets using a single unified pipeline (FreeSurfer -> ComBat harmonisation -> PCNtoolkit normative modelling -> clustering) to assess the stability of the number of subtypes and individual subtype assignments. (3) A replication test examining whether subtype-clinical associations reported in the literature reproduce under the unified framework. Expected outcomes include a quantitative summary of subtype characteristics and between-subtype differences, estimates of subtype-assignment stability (bootstrap adjusted Rand index), and an empirical assessment of the reproducibility of published subtype-clinical associations.