Normative Latent Representations of Early Brain Development
Poster No:
1521
Submission Type:
Abstract Submission
Authors:
Mariam Zabihi1, Francesca Biondo1, Jonathan O’Muircheartaigh2, Muriel Bruchhage3, James Cole1
Institutions:
1University College London (UCL), London, United Kingdom, 2Kings' College London (KCL), London, United Kingdom, 3Stavanger Medical Imaging Laboratory (SMIL), Department of Radiology, Stavanger University Hospital, Stavanger, Norway
First Author:
Co-Author(s):
Muriel Bruchhage
Stavanger Medical Imaging Laboratory (SMIL), Department of Radiology, Stavanger University Hospital
Stavanger, Norway
Stavanger Medical Imaging Laboratory (SMIL), Department of Radiology, Stavanger University Hospital
Stavanger, Norway
Introduction:
The human brain undergoes remarkable changes throughout life, with the most dramatic transformations occurring during early childhood, forming the basis for cognitive abilities and behavior (Vasung 2019, Gilmore 2018 ).
MRI studies have provided valuable insights into typical and atypical neurodevelopment, revealing rapid changes in brain metrics e.g., brain volume, cortical thickness, and white matter organization during early life (Geng 2017,Mills 2016). Normative modeling is a powerful tool to study these trajectories, mapping individual brain data against population references to identify deviations. However, traditional models relying on predefined features may overlook the complexity of brain development, while voxel-wise approaches are computationally intensive and less interpretable (Marquand 2016, Rutherford 2022). To address this, we propose a semi-supervised autoencoder (Zabihi, 2024) to learn latent representations directly from high-dimensional neuroimaging data.
MRI studies have provided valuable insights into typical and atypical neurodevelopment, revealing rapid changes in brain metrics e.g., brain volume, cortical thickness, and white matter organization during early life (Geng 2017,Mills 2016). Normative modeling is a powerful tool to study these trajectories, mapping individual brain data against population references to identify deviations. However, traditional models relying on predefined features may overlook the complexity of brain development, while voxel-wise approaches are computationally intensive and less interpretable (Marquand 2016, Rutherford 2022). To address this, we propose a semi-supervised autoencoder (Zabihi, 2024) to learn latent representations directly from high-dimensional neuroimaging data.
Methods:
Data: We used T1-weighted MRI scans from the BAMBAM (Brown University Assessment of Myelination and Behavioral Development Across Maturation) study (Bruchhange, 2020), focusing on neurotypical brain and cognitive development in children (N=564;1,027 scans; Age = 4.90 ± 3.67 years). Preprocessing included standardizing voxel intensities, and a five-fold cross-validation scheme was applied to enhance model generalizability.
We integrated 251 cognitive and behavioral measures from established assessments, including CBCL, Mullen Scales, WPPSI-IV, NEPSY-II, BRIEF, CTOPP-2, and Beery-VMI.
Method: We developed a 3D semi-supervised autoencoder to analyze structural MRI data, compressing high-dimensional data into latent representations while predicting age and sex to capture individual differences in neurodevelopment (Zabihi, 2024). The architecture includes three convolutional layers in both encoder and decoder, with a latent space of size 50. Training used the Adam optimizer, early stopping, and an exponentially decaying learning rate.
To visualize the latent space, we applied Uniform Manifold Approximation and Projection (UMAP), preserving both local and global data structures (McInnes, 2018). Hierarchical Bayesian regression (HBR) was used on the UMAP-reduced space to define normative brain states by removing age and sex effects, enabling calculation of a subject-specific Z-score ('latent index') to quantify deviations from typical development (de Boer, 2024).
We linked latent indices cognitive and behavioral measures and projected the normative latent space back into MRI input space to generate individualized normative brain scans.This allowed visualization of age-specific structural changes and provided biologically interpretable insights into neurodevelopmental trajectories.
We integrated 251 cognitive and behavioral measures from established assessments, including CBCL, Mullen Scales, WPPSI-IV, NEPSY-II, BRIEF, CTOPP-2, and Beery-VMI.
Method: We developed a 3D semi-supervised autoencoder to analyze structural MRI data, compressing high-dimensional data into latent representations while predicting age and sex to capture individual differences in neurodevelopment (Zabihi, 2024). The architecture includes three convolutional layers in both encoder and decoder, with a latent space of size 50. Training used the Adam optimizer, early stopping, and an exponentially decaying learning rate.
To visualize the latent space, we applied Uniform Manifold Approximation and Projection (UMAP), preserving both local and global data structures (McInnes, 2018). Hierarchical Bayesian regression (HBR) was used on the UMAP-reduced space to define normative brain states by removing age and sex effects, enabling calculation of a subject-specific Z-score ('latent index') to quantify deviations from typical development (de Boer, 2024).
We linked latent indices cognitive and behavioral measures and projected the normative latent space back into MRI input space to generate individualized normative brain scans.This allowed visualization of age-specific structural changes and provided biologically interpretable insights into neurodevelopmental trajectories.
Results:
The model achieved a reconstruction error of 0.08±0.01 and 79%±3% accuracy in sex classification and a mean absolute error (MAE) of 0.74±0.09 years with R^2=0.90 for age prediction.
The UMAP visualization of the latent space (Figure 1a) revealed components linked to sex and age, showing that the autoencoder effectively captured these factors. Figure 1b shows the correlations between cognitive/behavioral scores and devatiaon of the normative UMAPs components of the latenets (latent indices).
Back-projection of the normative UMAP into MRI space generated individualized brain scans (Figure 2), illustrating typical brain structures and age-related changes. From 4 to 54 months, white matter intensity increased, while cortical intensity decreased. Between 54 and 162 months, widespread intensity changes highlighted key developmental shifts. These findings connect brain development to cognitive and behavioral outcomes.
The UMAP visualization of the latent space (Figure 1a) revealed components linked to sex and age, showing that the autoencoder effectively captured these factors. Figure 1b shows the correlations between cognitive/behavioral scores and devatiaon of the normative UMAPs components of the latenets (latent indices).
Back-projection of the normative UMAP into MRI space generated individualized brain scans (Figure 2), illustrating typical brain structures and age-related changes. From 4 to 54 months, white matter intensity increased, while cortical intensity decreased. Between 54 and 162 months, widespread intensity changes highlighted key developmental shifts. These findings connect brain development to cognitive and behavioral outcomes.
Conclusions:
Our model provides a multivariate normative model of brain structure. Deviations from the latent correlated with cognitive and behavioral measures, highlighting links to individual differences and potential early indicators of neurodevelopmental trajectories.
Lifespan Development:
Normal Brain Development: Fetus to Adolescence 2
Modeling and Analysis Methods:
Methods Development 1
Keywords:
Modeling
Structures
Other - Normative model;Early age developmental trajectory; Brain trajectory ; Autoencoder; Cognitive behavioral scores; Childern; MRI
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
Gilmore JH.(2018). Imaging structural and functional brain development in early childhood. Nature Review Neuroscience, 19(3):123–137.
Geng X. (2017). Structural and Maturational Covariance in Early Childhood Brain Development. Cereberal Cortex, 27(3):1795–807.
Mills KL. (2016). Structural brain development between childhood and adulthood: Convergence across four longitudinal samples. Neuroimage, 141:273–281.
Marquand AF. (2016). Understanding Heterogeneity in Clinical Cohorts Using Normative Models: Beyond Case-Control Studies. Biological Psychiatry, 80(7):552–261.
Rutherford S. (2022). Charting brain growth and aging at high spatial precision. elife, 11:e72904.
Zabihi M. (2024). Nonlinear latent representations of high-dimensional task-fMRI data: Unveiling cognitive and behavioral insights in heterogeneous spatial maps. PLoS One, 19(8):e0308329
Bruchhage MMK. (2020). Functional connectivity correlates of infant and early childhood cognitive development. Brain structure & function, 225(2): 669–681.
McInnes L.(2018). UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv preprint arXiv:1802.03426.
de Boer AAA. (2024). Non-Gaussian Normative Modelling With Hierarchical Bayesian Regression. Imaging Neuroscience, 2, 1-36.