Leakage-Aware Cross-Atlas Analysis of ASD Connectome: Frontal-Temporal-DMN Hypoconnectivity and Thalamic Hyperconnectivity in ABIDE I with Validated Prediction

Abstract

We present a leakage-aware, cross-atlas analysis of functional connectivity in autism spectrum disorder (ASD) using the ABIDE I dataset. Exploratory analyses indicate a network-level phenotype of frontal-temporal-default mode network (DMN) hypoconnectivity and thalamic hyperconnectivity. To avoid inflated performance, we used strict leakage-safe validation (AAL atlas, leave-one-site-out), which gives a conservative estimate of predictive generalization (accuracy 62.2%). Symptom-connectivity regressions (ADOS/SRS/SCQ/ADI-R) were null and are reported transparently. We distinguish exploratory findings from validated ones. This is a preprint that has not been peer reviewed.