An Enhanced Conditional Variational Autoencoder-Based Normative Model for Neuroimaging Analysis
Poster No:
1119
Submission Type:
Abstract Submission
Authors:
Mai Ho1, Yang Song2, Lei Fan1, Nikita Husein1, Perminder Sachdev1,3, Jiyang Jiang1, Wei Wen1,3
Institutions:
1Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW), Sydney, NSW, Australia, 2School of Computer Science and Engineering, University of New South Wales (UNSW), Sydney, NSW, Australia, 3Neuropsychiatric Institute (NPI), Euroa Centre, Prince of Wales Hospital, Randwick, NSW, Australia
First Author:
Mai Ho
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Co-Author(s):
Yang Song, PhD
School of Computer Science and Engineering, University of New South Wales (UNSW)
Sydney, NSW, Australia
School of Computer Science and Engineering, University of New South Wales (UNSW)
Sydney, NSW, Australia
Lei Fan, PhD
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Nikita Husein
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Perminder Sachdev
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)|Neuropsychiatric Institute (NPI), Euroa Centre, Prince of Wales Hospital
Sydney, NSW, Australia|Randwick, NSW, Australia
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)|Neuropsychiatric Institute (NPI), Euroa Centre, Prince of Wales Hospital
Sydney, NSW, Australia|Randwick, NSW, Australia
Jiyang Jiang, PhD
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)
Sydney, NSW, Australia
Wei Wen
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)|Neuropsychiatric Institute (NPI), Euroa Centre, Prince of Wales Hospital
Sydney, NSW, Australia|Randwick, NSW, Australia
Centre for Healthy Brain Aging (CHeBA), School of Psychiatry, University of New South Wales (UNSW)|Neuropsychiatric Institute (NPI), Euroa Centre, Prince of Wales Hospital
Sydney, NSW, Australia|Randwick, NSW, Australia
Introduction:
Normative modelling in neuroimaging provides a valuable framework for quantifying individual brain variations against population norms (Marquand et al., 2016). While traditional statistical approaches can analyse multiple brain measures, they face practical limitations in scaling to high-dimensional neuroimaging data (Habeck, 2010). Deep learning approaches like conditional variational autoencoders (cVAEs) show promise for multivariate analysis (Lawry Aguila et al., 2022; Pinaya et al., 2019). However, existing cVAE frameworks typically employ a dual-input structure, requiring both observed values and covariates for predictions, which presents challenges for generating reliable probabilistic estimates. We propose an enhanced cVAE framework that addresses these limitations through improved probabilistic inference while enabling simultaneous modelling of multiple brain measures.
Methods:
Our model was developed and validated using 30 white matter hyperintensity (WMH) features from MRI scans of 26,731 UK Biobank participants (Sudlow et al., 2015), extracted via the UBO pipeline (Jiang et al., 2018). The cVAE architecture (Figure 1A) integrates an encoding-decoding process that maps multiple WMH features to a lower-dimensional latent space and reconstructs probabilistic predictions, conditioned on relevant covariates (age, sex, total intracranial volume, and vascular risk factors including diabetes, hypercholesterolemia, obesity, smoking). To enhance prediction robustness, we developed a novel inference method that samples latent vectors from a standard normal distribution and uses the trained decoder to generate predictions conditioned on covariates (Figure 1B). This approach produces probabilistic predictions by repeatedly sampling 1,000 times from the latent space and synthesising outcomes, followed by 1,000 bootstrap iterations to estimate summary statistics. We then calculated standardised deviation scores (z-scores) by comparing observed WMH volumes against model predictions. Model performance was benchmarked against three established approaches: Generalised Additive Models for Location, Scale, and Shape (GAMLSS) (Rigby & Stasinopoulos, 2005), Multivariate Fractional Polynomial Regression (MFPR) (Royston & Altman, 1994), and Hierarchical Bayesian Regression (HBR) (Lindley & Smith, 1972). For clinical validation, we examined relationships between model-derived deviations and hypertension severity in 20,746 hypertensive and 2,566 normotensive participants. The framework is publicly available at: github.com/maiho24/BrainNormativeCVAE.
Results:
Our cVAE framework effectively captured complex relationships between WMH features and covariates, achieving comparable performance to traditional statistical methods across various metrics. Analysis of extreme deviations (z-score > 2.58) in the hold-out dataset revealed distinct spatial patterns across models (Figure 2). Notably, the cVAE showed better generalisation, often showing 0% extreme deviations in most regions, while MFPR exhibited consistently higher rates. However, predictions were less reliable in regions with sparse data, such as the cerebellum and posterior artery callosal areas. For clinical validation, Spearman correlation analysis revealed that all models detected systematic increases in z-scores with increasing hypertension severity (e.g., p < 0.001 for whole-brain WMH volumes), confirming their ability to detect clinically meaningful variations.
Conclusions:
Our enhanced cVAE framework advances normative modelling by integrating deep learning capabilities and clinical requirements. Despite the challenges encountered in modelling regions with sparse data, the cVAE framework excels in advancing current normative models through a multivariate approach and robust probabilistic predictions. Future development will focus on improving the model architecture and optimising region-specific predictions while maintaining computational efficiency.
Modeling and Analysis Methods:
Bayesian Modeling
Classification and Predictive Modeling 1
Methods Development
Multivariate Approaches 2
Keywords:
Computational Neuroscience
Modeling
Multivariate
White Matter
Other - Normative Modeling; White Matter Hyperintensity; Conditional Variational Autoencoder; Hypertension; Probabilistic Inference
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:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Jiang, J., Liu, T., Zhu, W., Koncz, R., Liu, H., Lee, T., Sachdev, P. S., & Wen, W. (2018). UBO Detector – A cluster-based, fully automated pipeline for extracting white matter hyperintensities. NeuroImage, 174, 539–549. https://doi.org/10.1016/j.neuroimage.2018.03.050
Lawry Aguila, A., Chapman, J., Janahi, M., & Altmann, A. (2022). Conditional VAEs for Confound Removal and Normative Modelling of Neurodegenerative Diseases. In L. Wang, Q. Dou, P. T. Fletcher, S. Speidel, & S. Li (Eds.), Medical Image Computing and Computer Assisted Intervention – MICCAI 2022 (pp. 430–440). Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-16431-6_41
Lindley, D. V., & Smith, A. F. M. (1972). Bayes Estimates for the Linear Model. Journal of the Royal Statistical Society. Series B (Methodological), 34(1), 1–41.
Marquand, A. F., Rezek, I., Buitelaar, J., & Beckmann, C. F. (2016). Understanding Heterogeneity in Clinical Cohorts Using Normative Models: Beyond Case-Control Studies. Biological Psychiatry, 80(7), 552–561. https://doi.org/10.1016/j.biopsych.2015.12.023
Pinaya, W. H. L., Mechelli, A., & Sato, J. R. (2019). Using deep autoencoders to identify abnormal brain structural patterns in neuropsychiatric disorders: A large-scale multi-sample study. Human Brain Mapping, 40(3), 944–954. https://doi.org/10.1002/hbm.24423
Rigby, R. A., & Stasinopoulos, D. M. (2005). Generalized Additive Models for Location, Scale and Shape. Journal of the Royal Statistical Society Series C: Applied Statistics, 54(3), 507–554. https://doi.org/10.1111/j.1467-9876.2005.00510.x
Royston, P., & Altman, D. G. (1994). Regression Using Fractional Polynomials of Continuous Covariates: Parsimonious Parametric Modelling. Journal of the Royal Statistical Society. Series C (Applied Statistics), 43(3), 429–467. https://doi.org/10.2307/2986270
Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., Downey, P., Elliott, P., Green, J., Landray, M., Liu, B., Matthews, P., Ong, G., Pell, J., Silman, A., Young, A., Sprosen, T., Peakman, T., & Collins, R. (2015). UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLOS Medicine, 12(3), e1001779. https://doi.org/10.1371/journal.pmed.1001779