Voxel-Based Normative Modelling of Brain Microstructure with GAMLSS
Poster No:
1680
Submission Type:
Abstract Submission
Authors:
Tommaso Pavan1, Ludwig Tiston2, Yasser Alemán-Gómez3, Raoul Jenni1, Pascal Steullet1, Martin Cleusix1, Luis Alameda1, Kim Do4, Philippe Conus5, Patric Hagmann3, Jonas richiardi1, Paul Klauser4, Ileana Jelescu6
Institutions:
1Lausanne University Hospital (CHUV), Lausanne, Vaud, 2Faculty of Science and Technology, Uppsala University, Uppsala, Sweden, 3Lausanne University Hospital and University of Lausanne (CHUV-UNIL), Lausanne, Vaud, 4Center for Psychiatric Neuroscience, Department of Psychiatry, Lausanne University Hospital and UNIL, Lausanne, Vaud, 5Service of General Psychiatry, Lausanne University Hospital (CHUV), Lausanne, Switzerland, Lausanne, Vaud, 6Department of Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Vaud
First Author:
Co-Author(s):
Yasser Alemán-Gómez
Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Vaud
Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Vaud
Kim Do
Center for Psychiatric Neuroscience, Department of Psychiatry, Lausanne University Hospital and UNIL
Lausanne, Vaud
Center for Psychiatric Neuroscience, Department of Psychiatry, Lausanne University Hospital and UNIL
Lausanne, Vaud
Philippe Conus
Service of General Psychiatry, Lausanne University Hospital (CHUV), Lausanne, Switzerland
Lausanne, Vaud
Service of General Psychiatry, Lausanne University Hospital (CHUV), Lausanne, Switzerland
Lausanne, Vaud
Paul Klauser
Center for Psychiatric Neuroscience, Department of Psychiatry, Lausanne University Hospital and UNIL
Lausanne, Vaud
Center for Psychiatric Neuroscience, Department of Psychiatry, Lausanne University Hospital and UNIL
Lausanne, Vaud
Ileana Jelescu
Department of Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL)
Lausanne, Vaud
Department of Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL)
Lausanne, Vaud
Introduction:
In the normative modelling framework, individual differences are mapped to a reference level defined as the "norm" (Rutherford, 2022). The normative reference is estimated on a large cohort of healthy participants to predict a response variable (metric of interest) from a set of explanatory variables (e.g. age, sex), to which the observed empirical value of any individual (e.g. early psychosis (EP) patient) is compared.
Generalized Additive Models for Location Scale and Shape (GAMLSS) is an univariate distributional regression framework that allows modelling any non-Gaussian distribution, accounting for location, scale, skewness and kurtosis (Dinga, 2021). GAMLSS has long been used for normative modelling of brain volumes (Bethlehem, 2022) and by the World Health Organization for growth charts (Onis, 2007). Here, we present an optimized R package: "VBGAMLSS" to flexibly apply GAMLSS to voxel- or vertex-wise neuroimaging data for individual level-analysis, and we demonstrate an application to diffusion MRI data from the Human Connectome Project (HCP, Van Essen, 2013) and our psychosis cohort (LSP).
Generalized Additive Models for Location Scale and Shape (GAMLSS) is an univariate distributional regression framework that allows modelling any non-Gaussian distribution, accounting for location, scale, skewness and kurtosis (Dinga, 2021). GAMLSS has long been used for normative modelling of brain volumes (Bethlehem, 2022) and by the World Health Organization for growth charts (Onis, 2007). Here, we present an optimized R package: "VBGAMLSS" to flexibly apply GAMLSS to voxel- or vertex-wise neuroimaging data for individual level-analysis, and we demonstrate an application to diffusion MRI data from the Human Connectome Project (HCP, Van Essen, 2013) and our psychosis cohort (LSP).
Methods:
VBGAMLSS builds on the gamlss2 R package (Rigby, 2005), parallelizing and optimizing it for voxel/vertex computations, also featuring cross-validation (CV), segmentation-aware modelling, and an integrated high-performance computing cluster submission/monitoring system.
As demonstration of our package, MRI data from 2033 healthy participants from 6 datasets (304 HCP-Development, 512 HCP-Aging, 1032 HCP-YoungAdults, 56 HCP-EarlyPsychosis, 59 LSP-Prisma, and 70 LSP-Trio) aged 15 to 70 years old, were used to generate 55 age-specific multimodal templates using ANTs. Template generation uses T1w, fractional anisotropy (FA), and white and gray matter segmentation maps. Age-specific templates were used to form a final template across ages, where every healthy subject white matter (WM) mean diffusivity and kurtosis maps (MD, MK; Jensen, 2010) were projected to. This approach allows the prediction of expected values in template space and computation of z-scores in native space.
To model MD and MK as a function of age, sex, dataset, and brain volume for each WM voxel, we used SHASH distribution (Jones & Pewsey, 2009). Fourteen models were considered, from which we selected the best across folds and WM voxels by running a 5-fold CV, and ranking models based on best model frequency, minimum global deviance (GD=-2log(L)) and Akaike Weights (Wagenmakers & Farrell, 2004). Finally, we estimated the z-scores for MD and MK maps in EP patients from the HCP-EarlyPsychosis dataset (N=112) in both native and template space.
As demonstration of our package, MRI data from 2033 healthy participants from 6 datasets (304 HCP-Development, 512 HCP-Aging, 1032 HCP-YoungAdults, 56 HCP-EarlyPsychosis, 59 LSP-Prisma, and 70 LSP-Trio) aged 15 to 70 years old, were used to generate 55 age-specific multimodal templates using ANTs. Template generation uses T1w, fractional anisotropy (FA), and white and gray matter segmentation maps. Age-specific templates were used to form a final template across ages, where every healthy subject white matter (WM) mean diffusivity and kurtosis maps (MD, MK; Jensen, 2010) were projected to. This approach allows the prediction of expected values in template space and computation of z-scores in native space.
To model MD and MK as a function of age, sex, dataset, and brain volume for each WM voxel, we used SHASH distribution (Jones & Pewsey, 2009). Fourteen models were considered, from which we selected the best across folds and WM voxels by running a 5-fold CV, and ranking models based on best model frequency, minimum global deviance (GD=-2log(L)) and Akaike Weights (Wagenmakers & Farrell, 2004). Finally, we estimated the z-scores for MD and MK maps in EP patients from the HCP-EarlyPsychosis dataset (N=112) in both native and template space.
Results:
The CV model selection settled on a model including three smooth effects, one for age for each sex (Fig.1A,B & E,F) and one for the brain volume (Fig.1C,G), while dataset/study effects were corrected via a random intercept for dataset (Fig.1D,H). In our application, at individual EP patient level, the genu of the corpus callosum showed elevated MD (z>0) and reduced MK (z<0, Fig.2A,D). At group level, the pattern was more pronounced (Fig.2B,E), though with high variability across patients (Fig.2C,E). Finally, a highly negative z-score rim can be noticed around the ventricles in the average-patient MK z-score map (Fig.2F), possibly indicating ventricular enlargement.
Conclusions:
We developed a framework to train and estimate normative models at voxel or vertex levels. We envision our framework as a powerful resource for researchers and clinicians to gain deeper insights into individual-level deviations. Our use case example shows how z-scores maps can be used to generate individualized alterations maps of the WM which highlight regions of high interest in psychosis (Kelly, 2018). The package is accessible on "github.com/tmspvn/VBGAMLSS" under apache-2.0 license.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 2
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis
Methods Development
Multivariate Approaches
Univariate Modeling 1
Keywords:
Computing
Data analysis
Machine Learning
Modeling
Multivariate
Psychiatric
Schizophrenia
Statistical Methods
White Matter
Other - Normative modelling
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.
Dinga, R., Fraza, C., Bayer Jmm, Seyed Mostafa Kia, Beckmann, C. F., & Marquand, A. F. (2021). Normative modeling of neuroimaging data using generalized additive models of location scale and shape. https://doi.org/10.1101/2021.06.14.448106
Jones, M. C., & Pewsey, A. (2009). Sinh-arcsinh distributions. Biometrika, 96(4), 761–780. https://doi.org/10.1093/biomet/asp053
Kelly, S., Jahanshad, N., Zalesky, A., Kochunov, P., Agartz, I., Alloza, C., Andreassen, O. A., Arango, C., Banaj, N., Bouix, S., Bousman, C. A., Brouwer, R. M., Bruggemann, J., Bustillo, J., Cahn, W., Calhoun, V., Cannon, D., Carr, V., Catts, S., & Chen, J. (2017). Widespread white matter microstructural differences in schizophrenia across 4322 individuals: results from the ENIGMA Schizophrenia DTI Working Group. Molecular Psychiatry, 23(5), 1261–1269. https://doi.org/10.1038/mp.2017.170
Rigby, R. A., & Stasinopoulos, D. M. (2005). Generalized additive models for location, scale and shape (with discussion). 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
Rutherford, S., Kia, S.M., Wolfers, T. et al. The normative modeling framework for computational psychiatry. Nat Protoc 17, 1711–1734 (2022). https://doi.org/10.1038/s41596-022-00696-5
Van Essen, D. C., Smith, S. M., Barch, D. M., Behrens, T. E. J., Yacoub, E., & Ugurbil, K. (2013). The WU-Minn Human Connectome Project: An overview. NeuroImage, 80, 62–79. https://doi.org/10.1016/j.neuroimage.2013.05.041
WHO Child Growth Standards based on length/height, weight and age. Acta Paediatrica, 95(S450), 76–85. https://doi.org/10.1111/j.1651-2227.2006.tb02378.x