Brain age prediction using fixel-based measures from advanced diffusion-weighted imaging
Poster No:
2359
Submission Type:
Abstract Submission
Authors:
Remika Mito1,2, Jocelyn Halim1, Heath Pardoe1, James Cole3, Maria Di Biase2, Andrew Zalesky2,4
Institutions:
1Florey Institute of Neuroscience and Mental Health, Melbourne, Victoria, Australia, 2Department of Psychiatry, The University of Melbourne, Melbourne, Victoria, Australia, 3Centre for Medical Image Computing, University College London, London, London, UK, 4Department of Biomedical Engineering, The University of Melbourne, Melbourne, Victoria, Australia
First Author:
Remika Mito
Florey Institute of Neuroscience and Mental Health|Department of Psychiatry, The University of Melbourne
Melbourne, Victoria, Australia|Melbourne, Victoria, Australia
Florey Institute of Neuroscience and Mental Health|Department of Psychiatry, The University of Melbourne
Melbourne, Victoria, Australia|Melbourne, Victoria, Australia
Co-Author(s):
Andrew ZALESKY, PhD
Department of Psychiatry, The University of Melbourne|Department of Biomedical Engineering, The University of Melbourne
Melbourne, Victoria, Australia|Melbourne, Victoria, Australia
Department of Psychiatry, The University of Melbourne|Department of Biomedical Engineering, The University of Melbourne
Melbourne, Victoria, Australia|Melbourne, Victoria, Australia
Introduction:
Machine learning applied to structural neuroimaging data can accurately predict chronological age in neurologically healthy individuals. Diffusion-weighted imaging (DWI) can be used to probe the in vivo white matter architecture of the brain, which is also known to exhibit substantial age-associated change. However, the ability of DWI to quantify changes in the brain's white matter fibre pathways depends on the way in which data are acquired, modelled and analysed. Previous work has demonstrated the ability to estimate brain age using both diffusion tensor imaging (DTI)-derived and advanced diffusion MRI measures(1,2). Here, we implement an advanced DWI modelling and analysis (fixel-based) approach to predict chronological age in a large cohort of older adults from the Human Connectome Project Lifespan study.
Methods:
Participants from the Lifespan Human Connectome Project Aging (HCP-A; n=680 (382 females); age range: 36.0 – 89.8) were included.
Multi-shell DWI data were collected(3) and preprocessed(4) as previously described. Following this, fibre orientation distribution (FOD) functions were extracted using multi-shell multi-tissue constrained spherical deconvolution (MSMT-CSD[5]). Spatial correspondence was achieved with a study-specific template through FOD-guided registration. The fixel-based analysis (FBA(6)) pipeline was used to extract a measure of fibre density and cross-section (FDC) at each white matter 'fixel' (fibre population within a voxel). Mean FDC values were extracted for each participant in 72 white matter fibre structures, defined by performing TractSeg(7) on the study-specific template FOD image.
FreeSurfer-derived measures of cortical thickness and volume were extracted for the same participants from preprocessed structural imaging data (Lifespan 2.0 Release). 116 thickness and volume features were used for age prediction as per (8) (using Desikan-Killiany atlas and aseg Freesurfer atlas(9)).
Brain-predicted age was calculated using Gaussian Process Regression, trained on the following imaging data:
(i) Model 1: FDC values only
(ii) Model 2: Cortical thickness and subcortical volumes
(iii) Model 3: Combined FDC and cortical thickness and subcortical volumes
For all models, predicted age was estimated using the same 80% training set and validated on a 20% test set. The predicted brain-age accuracy of each these models was assessed on the validation set using mean absolute error (MAE), root mean square error (RMSE) Pearson's r, and total variance explained (R2).
Figure 1 provides an overview of the pipeline for this study.
Multi-shell DWI data were collected(3) and preprocessed(4) as previously described. Following this, fibre orientation distribution (FOD) functions were extracted using multi-shell multi-tissue constrained spherical deconvolution (MSMT-CSD[5]). Spatial correspondence was achieved with a study-specific template through FOD-guided registration. The fixel-based analysis (FBA(6)) pipeline was used to extract a measure of fibre density and cross-section (FDC) at each white matter 'fixel' (fibre population within a voxel). Mean FDC values were extracted for each participant in 72 white matter fibre structures, defined by performing TractSeg(7) on the study-specific template FOD image.
FreeSurfer-derived measures of cortical thickness and volume were extracted for the same participants from preprocessed structural imaging data (Lifespan 2.0 Release). 116 thickness and volume features were used for age prediction as per (8) (using Desikan-Killiany atlas and aseg Freesurfer atlas(9)).
Brain-predicted age was calculated using Gaussian Process Regression, trained on the following imaging data:
(i) Model 1: FDC values only
(ii) Model 2: Cortical thickness and subcortical volumes
(iii) Model 3: Combined FDC and cortical thickness and subcortical volumes
For all models, predicted age was estimated using the same 80% training set and validated on a 20% test set. The predicted brain-age accuracy of each these models was assessed on the validation set using mean absolute error (MAE), root mean square error (RMSE) Pearson's r, and total variance explained (R2).
Figure 1 provides an overview of the pipeline for this study.
·Figure 1
Results:
Predictive model using only FDC inputs alone predicted chronological age with a mean average error (MAE) of 7.55 years, and a strong correlation with true age (r = 0.78; Figure 2). Using cortical thickness and subcortical volume values predicted chronological age with an MAE of 6.40 years, as well as a strong correlation with true age (r = 0.85). Combining the input values further improved model predictions (r = 0.87, MAE = 5.96 years).
·Figure 2
Conclusions:
Brain age prediction using an advanced DWI measure (FDC) demonstrated comparable prediction accuracy to previously published models using conventional diffusion tensor-based metrics(1,2), and improved the accuracy when combined with standard structural MRI measures. Combining DWI and structural brain measures yielded the most accurate predictions. Examining deviations in these brain age predictions could be valuable biomarkers in clinical cohorts where fibre-specific white matter abnormalities have been reported(10). Future studies may benefit from the use of fixel-level (rather than tract-summarised) measures of FDC in brain age prediction.
Lifespan Development:
Aging
Early life, Adolescence, Aging 2
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity
Novel Imaging Acquisition Methods:
Diffusion MRI 1
Keywords:
ADULTS
Aging
Machine Learning
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Provide references using author date format
2. Beck, D. et al. White matter microstructure across the adult lifespan: A mixed longitudinal and cross-sectional study using advanced diffusion models and brain-age prediction. Neuroimage 224, 117441 (2021).
3. Harms, M. P. et al. Extending the Human Connectome Project across ages: Imaging protocols for the Lifespan Development and Aging projects. Neuroimage 183, 972–984 (2018).
4. Glasser, M. F. et al. The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80, 105–124 (2013).
5. Jeurissen, B., Tournier, J. D., Dhollander, T., Connelly, A. & Sijbers, J. Multi-tissue constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data. Neuroimage 103, 411–426 (2014).
6. Raffelt, D. A. et al. Investigating white matter fibre density and morphology using fixel-based analysis. Neuroimage 144, 58–73 (2017).
7. Wasserthal, J., Neher, P. & Maier-Hein, K. H. TractSeg-Fast and accurate white matter tract segmentation. Neuroimage 183, 239–253 (2018).
8. Dafflon, J. et al. An automated machine learning approach to predict brain age from cortical anatomical measures. Hum Brain Mapp 41, 3555–3566 (2020).
9. Desikan, R. S. et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 31, 968–980 (2006).
10. Mito, R. et al. Fibre-specific white matter reductions in Alzheimer’s disease and mild cognitive impairment. Brain 141, 888–902 (2018).