Parsing altered brain connectivity in neurodevelopmental disorders by integrating graph-based normative modeling and deep generative networks

TL;DR

The Brain Representation via Individualized Deep Generative Embedding (BRIDGE) framework is presented, which integrates normative modeling with a bio-inspired deep generative model to create a reference trajectory of connectivity transformation as part of neurotypical development and forms powerful tools for quantifying developmental divergence in connectivity patterns.

Abstract

Divergent brain connectivity is thought to underlie the behavioral and cognitive symptoms observed in many neurodevelopmental disorders. Quantifying divergence from neurotypical connectivity patterns offers a promising pathway to inform diagnosis and therapeutic interventions. While advanced neuroimaging techniques, such as diffusion MRI (dMRI), have facilitated the mapping of brain's structural connectome, the challenge lies in accurately modeling developmental trajectories within these complex networked structures to create robust neurodivergence markers. In this work, we present the Brain Representation via Individualized Deep Generative Embedding (BRIDGE) framework, which integrates normative modeling with a bio-inspired deep generative model to create a reference trajectory of connectivity transformation as part of neurotypical development. This will enable the assessment of neurodivergence by comparing individuals to the established neurotypical trajectory. BRIDGE provides a global neurodivergence score based on the difference between connectivity-based brain age and chronological age, along with region-wise neurodivergence maps that highlight localized connectivity differences. Application of BRIDGE to a large cohort of children with autism spectrum disorder demonstrates that the global neurodivergence score correlates with clinical assessments in autism, and the regional map offers insights into the heterogeneity at the individual level in neurodevelopmental disorders. Together, the neurodivergence score and map form powerful tools for quantifying developmental divergence in connectivity patterns, advancing the development of imaging markers for personalized diagnosis and intervention in various clinical contexts.

Figures (7)

• Figure 1: Overview of the BRIDGE framework. A) Architecture of the BRIDGE model for learning neurotypical developmental trajectories of structural connectivity using connectomes. Graph-based VAE was used to encode the structural connectivity information in connectome via lower-dimensional latent features. The BRIDGE decoder mapped these latent features to each brain region and inferred the regional features to generate structural connectomes. This generative process was guided by two biological wiring constraints, Hebbian learning and cost minimization. B) Illustration showing how two properties of brain connectivity, homophilic facilitation and rich-club organization, could be replicated by the BRIDGE model via specific regional features. C) Derivation of the region-wise neurodivergence map and the global neurodivergence score.
• Figure 2: Performance of the BRIDGE model were compared with two other deep generative models (FC-VAE, GCN-VAE) and 13 classic generative network models based on the dissimilarity between generated and actual structural connectomes. For the CAR dataset, dissimilarity was computed separately for the neurotypical group and the autism group. The BRIDGE model achieved the lowest dissimilarity, positioning it as the optimal choice for both datasets irrespective of diagnostic status.
• Figure 3: Visualization of latent space using t-SNE. Latent features derived from BRIDGE showed a notable age-related trend for neurotypical participants in both the PNC (left) and CAR (right) cohorts. Feature dimensions were reduced to 2D using t-SNE for visualization. Each dot was colored according to the participant's chronological age. The average trajectories of neurotypical participants across different ages were represented using as the black dashed lines.
• Figure 4: Ablation study by removing each building component in BRIDGE. The constituent elements of BRIDGE and three competing models were illustrated in the table. Comparative assessment of percentage changes upon removal of each building components was evaluated through MAE, RMSE, and R metrics.
• Figure 5: Illustration of neurodivergence in one dimension of latent features captured by the BRIDGE model. Neurotypical participants are represented by blue dots, while autistic children are marked with red "X"s. The neurotypical trajectory across different chronological ages is shown as a blue line, enveloped by a shaded area representing the 95% confidence interval, reflecting the variability among neurotypicals. Autistic children display divergent developmental patterns in their latent features compared to neurotypicals.