# Mapping ADHD Heterogeneity and Biotypes by Topological Deviations in Morphometric Similarity Networks

## TL;DR

This study uses brain network analysis to identify three distinct ADHD subtypes in children, each with unique brain patterns and clinical features.

## Contribution

A novel method combining normative modeling and clustering identifies ADHD biotypes with distinct neurobiological and clinical profiles.

## Key findings

- Three ADHD biotypes were identified with distinct brain network deviations and clinical profiles.
- Each biotype showed unique neurochemical and functional correlates.
- Findings were validated in an independent cohort, showing robust generalizability.

## Abstract

This case-control study investigates if normative modeling of topological properties derived from brain morphometric similarity networks yields robust stratification biomarkers for pediatric populations with attention-deficit/hyperactivity disorder (ADHD).

Can normative modeling of topological properties derived from brain morphometric similarity networks yield robust stratification biomarkers for pediatric populations with attention-deficit/hyperactivity disorder (ADHD)?

This multisite case-control study included 1154 participants, characterizing ADHD heterogeneity through hub-centric topological deviations derived from morphometric similarity networks. Three distinct biotypes emerged, each exhibiting unique clinical-neural profiles with characteristic neurochemical and functional correlates, validated in an independent transdiagnostic cohort of 554 ADHD cases.

The integration of normative modeling with heterogeneity through discriminative analysis (HYDRA) clustering yielded both dimensional and categorical insights into ADHD heterogeneity, thereby enhancing our understanding of the ADHD’s neurobiological complexity.

Attention-deficit/hyperactivity disorder (ADHD) is characterized by considerable clinical heterogeneity, and existing classification frameworks constrain the development of neurobiologically informed subtyping approaches.

To investigate whether normative modeling of topological properties derived from brain morphometry similarity networks can provide robust stratification markers for children with ADHD.

This case-control study leveraged multisite cross-sectional neurodevelopmental datasets with a longitudinal follow-up cognitive assessment for a subset. Morphometric similarity networks were constructed and normative models were developed for 3 topological metrics: degree centrality, nodal efficiency, and participation coefficient. Through semisupervised clustering, putative biotypes were delineated and their clinical profiles were examined. Brain profiles of these biotypes were further contextualized in terms of their neurochemical and functional correlates using large-scale databases, and model generalizability was assessed with external validation performed in an independent transdiagnostic cohort. Study data were analyzed from November 2023 to January 2025.

Normative modeling of topological properties derived from brain morphometry.

Topological deviations in morphometric similarity networks derived from brain structural image.

The discovery cohort comprised 446 children with ADHD (mean [SD] age, 11.5 [2.6] years; 339 male [76.0%]) and 708 controls (mean [SD] age, 11.0 [2.3] years; 429 male [60.6%]), whereas the validation cohort included 554 children with ADHD (mean [SD] age, 10.1 [2.8]; 372 male [67.1%]) and 123 controls (mean [SD] age, 10.1 [3.0]; 70 male [56.9%]). ADHD exhibited atypical hub organization across all 3 topological metrics, with significant case-control differences primarily localized to a covarying multimetric component in the orbitofrontal cortex. Three biotypes emerged: severe-combined with emotional dysregulation (widespread medial prefrontal cortex-pallidum alterations, n = 142), predominantly hyperactive/impulsive (anterior cingulate cortex-pallidum circuit alterations, n = 177), and predominantly inattentive (superior frontal gyrus alterations, n = 127), each characterized by distinct clinical profiles and longitudinal trajectories. These neural profiles of each biotype showed distinct neurochemical and functional correlates. Critically, the core findings were replicated in the validation cohort, demonstrating robust generalizability.

Results of this case-control study reveal that the integration of normative modeling with semisupervised clustering provided both dimensional and categorical insights into ADHD heterogeneity, identifying 3 distinct ADHD biotypes with unique clinical-neural profiles that advance the understanding of ADHD’s neurobiological complexity and lay the groundwork for personalized management.