Multimodal MRI-based neuromarkers trace longitudinal changes in cognitive functioning in ADHD

Abstract

The National Institute of Mental Healths Research Domain Criteria (RDoC) framework conceptualises cognition as a core functional domain for psychopathology that should be studied across multiple units of analysis and developmental timescales. In ADHD, however, it remains unclear whether neuroimaging-derived markers can not only predict inter-individual differences in cognition but also track longitudinal cognitive development and capture cognition-psychopathology relationships. Using the longitudinal Oregon ADHD-1000 study (n = 594 participants; 1,053 observations), we developed multimodal machine-learning markers of general cognitive functioning (g) from structural and resting-state functional MRI. The multimodal marker achieved an out-of-sample correlation of r = .46 and generalised similarly across children with and without ADHD. The marker explained 25.01% of interindividual variance (r = .48) and 18.82% of intraindividual variance (r = .52) in cognitive functioning. Commonality analyses showed that it captured 60.87% of intraindividual age-related cognitive variation (5.18% of the total variance in g). The marker also accounted for substantial portions of the cognition-hyperactivity association (58.79%; 6.39% of total variance in g) and the cognition-inattention association (25.99%; 4.13% of total variance in g). These findings provide evidence that multimodal structural and functional MRI can generate RDoC-informed markers that predict cognitive functioning, track cognitive development over time, and capture meaningful links between cognition and ADHD symptoms. Although predictive performance remains below levels required for clinical application, the results establish a foundation for longitudinal, RDoC-inspired investigations of cognitive development in ADHD.