Brain age pre-training for prediction of Alzheimer’s disease diagnosis and progression
Poster No:
1396
Submission Type:
Abstract Submission
Authors:
Trevor Wei Kiat Tan1,2,3,4,5, Kim-Ngan Nguyen1,2, Chen Zhang1,2,4,5, Ru Kong1,2,4,5, Susan Cheng1,2,3,4, Fang Ji1,2,4, Joanna Su Xian Chong1,2,4, Eddie Jun Yi Chong6,7, Narayanaswamy Venketasubramanian8, Christopher Chen6,7,9, Juan Helen Zhou1,2,3,4, B. T. Thomas Yeo1,2,3,4,5
Institutions:
1Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore, 2Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore, 3Integrative Sciences and Engineering Programme (ISEP), National University of Singapore, Singapore, Singapore, 4Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore, 5N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore, Singapore, Singapore, 6Memory, Aging and Cognition Centre, National University Health System, Singapore, Singapore, 7Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore, 8Raffles Neuroscience Centre, Raffles Hospital, Singapore, Singapore, 9Department of Pharmacology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore
First Author:
Trevor Wei Kiat Tan
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Co-Author(s):
Kim-Ngan Nguyen
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore
Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore
Singapore, Singapore|Singapore, Singapore
Chen Zhang
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Ruby Kong
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Susan Cheng
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Fang Ji
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Joanna Su Xian Chong
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Eddie Jun Yi Chong
Memory, Aging and Cognition Centre, National University Health System|Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore
Singapore, Singapore|Singapore, Singapore
Memory, Aging and Cognition Centre, National University Health System|Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore
Singapore, Singapore|Singapore, Singapore
Christopher Chen
Memory, Aging and Cognition Centre, National University Health System|Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore|Department of Pharmacology, Yong Loo Lin School of Medicine, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Memory, Aging and Cognition Centre, National University Health System|Department of Psychological Medicine, Yong Loo Lin School of Medicine, National University of Singapore|Department of Pharmacology, Yong Loo Lin School of Medicine, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Juan Helen Zhou
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
B. T. Thomas Yeo
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore|Centre for Translational MR Research, Yong Loo Lin School of Medicine, National University of Singapore|Integrative Sciences and Engineering Programme (ISEP), National University of Singapore|Department of Electrical and Computer Engineering, National University of Singapore|N.1 Institute for Health & Institute for Digital Medicine (WisDM), National University of Singapore
Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore|Singapore, Singapore
Introduction:
An individual's brain age, predicted by a machine learning algorithm using structural MRI, holds vital clinical importance. Brain age higher than chronological age is linked to cognitive decline [1], mortality [2], and brain disorders [3]. Age data is widely available across MRI datasets. Therein lies a theoretical advantage in training brain age models on larger and diverse datasets and applying these pre-trained brain age models for downstream prediction on smaller clinical samples via transfer learning [4]. We aim to investigate if pre-trained brain age models outperform models trained-from-scratch to diagnose Alzheimer's disease (AD) and predict mild cognitively impaired (MCI) progression to AD.
Methods:
The study employed three datasets – Alzheimer's Disease Neuroimaging Initiative (ADNI), Australian Imaging, Biomarkers and Lifestyle (AIBL) study, and Singapore Memory Aging and Cognition Centre (MACC) Harmonization cohort – for AD diagnosis and MCI progression tasks.
Both tasks involved binary classification and used the same nested cross-validation method. Models were compared using Area Under the Curve (AUC) on test set and resampled t-tests with FDR correction. No participant overlap occurred between the two tasks.
Models shared a common network architecture with a pre-trained brain age model [4]. Feature-extracted pre-trained models processed structural MRI and generated 64-dimensional features per participant, which were used for logistic regression label prediction [4]. Finetuned models retrained all layers and replaced the last age prediction layer with a binary prediction layer.
AD diagnosis distinguished between AD and non-cognitively impaired (NCI) individuals, using 856 ADNI, 156 AIBL, and 260 MACC participants. Three models were compared – trained-from-scratch model (Direct-scratch), feature-extracted brain age pre-trained model (Indirect-brainage), and finetuned brain age pre-trained model (Indirect-brainage-finetune), with training + validation set sizes ranging from 50 to 997. Indirect-brainage and Indirect-brainage-finetune were initialized with a state-of-the-art pre-trained brain age model [4].
MCI progression distinguished stable from progressive MCI, using 478 ADNI, 20 AIBL, and 78 MACC participants. Two models were compared – feature-extracted Indirect-brainage-finetune (Indirect-brainage-finetune-AD), and feature-extracted Direct-scratch (Direct-AD), with training + validation set sizes ranging from 50 to 448.
Both tasks involved binary classification and used the same nested cross-validation method. Models were compared using Area Under the Curve (AUC) on test set and resampled t-tests with FDR correction. No participant overlap occurred between the two tasks.
Models shared a common network architecture with a pre-trained brain age model [4]. Feature-extracted pre-trained models processed structural MRI and generated 64-dimensional features per participant, which were used for logistic regression label prediction [4]. Finetuned models retrained all layers and replaced the last age prediction layer with a binary prediction layer.
AD diagnosis distinguished between AD and non-cognitively impaired (NCI) individuals, using 856 ADNI, 156 AIBL, and 260 MACC participants. Three models were compared – trained-from-scratch model (Direct-scratch), feature-extracted brain age pre-trained model (Indirect-brainage), and finetuned brain age pre-trained model (Indirect-brainage-finetune), with training + validation set sizes ranging from 50 to 997. Indirect-brainage and Indirect-brainage-finetune were initialized with a state-of-the-art pre-trained brain age model [4].
MCI progression distinguished stable from progressive MCI, using 478 ADNI, 20 AIBL, and 78 MACC participants. Two models were compared – feature-extracted Indirect-brainage-finetune (Indirect-brainage-finetune-AD), and feature-extracted Direct-scratch (Direct-AD), with training + validation set sizes ranging from 50 to 448.
Results:
Figure 1 shows the relationship between the model's AD diagnosis test AUC and the training + validation set sizes for three models – Indirect-brainage, Indirect-brainage-finetune, and Direct-scratch. Direct-scratch outperformed Indirect-brainage significantly from size 400 or greater up to the largest size of 997. However, there were no significant differences between Direct-scratch and Indirect-brainage-finetune across all training and validation set sizes.
Figure 2 shows the relationship between the model's MCI progression test AUC and the training + validation set sizes for two models – Indirect-brainage-finetune-AD and Direct-AD. Notably, there were no significant differences in test AUCs between the two models across all training and validation set sizes.
Figure 2 shows the relationship between the model's MCI progression test AUC and the training + validation set sizes for two models – Indirect-brainage-finetune-AD and Direct-AD. Notably, there were no significant differences in test AUCs between the two models across all training and validation set sizes.
Conclusions:
For AD diagnosis, finetuning the pre-trained brain age model significantly improves prediction versus feature extracting the pre-trained brain age model. Interestingly, a model trained-from-scratch significantly outperforms a feature extracted pre-trained brain age model at a training and validation sample size equal to 400 or greater. Hence, for smaller sample sizes (less than 400), using the less computationally intensive feature extraction from the pre-trained model is more beneficial than training a model from scratch.
For MCI progression, there is no significant difference in performance between brain age pre-training versus random initialization, when both models feature-extracted AD diagnosis weights, from 50 to 448 sample sizes.
For MCI progression, there is no significant difference in performance between brain age pre-training versus random initialization, when both models feature-extracted AD diagnosis weights, from 50 to 448 sample sizes.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Keywords:
Machine Learning
MRI
Neurological
Other - brain age; dementia; mild cognitive impairment; Alzheimer's disease; transfer learning; deep learning; diagnosis; prognosis; disease progression
1|2Indicates the priority used for review
Provide references using author date format
[2] Elliott, M. L., D. W. Belsky, A. R. Knodt, D. Ireland, T. R. Melzer, R. Poulton, S. Ramrakha, A. Caspi, T. E. Moffitt and A. R. Hariri (2021). "Brain-age in midlife is associated with accelerated biological aging and cognitive decline in a longitudinal birth cohort." Mol Psychiatry 26(8): 3829-3838.
[3] Kaufmann, T., D. van der Meer, N. T. Doan, E. Schwarz, M. J. Lund, I. Agartz, D. Alnaes, D. M. Barch, R. Baur-Streubel, A. Bertolino, F. Bettella, M. K. Beyer, E. Boen, S. Borgwardt, C. L. Brandt, J. Buitelaar, E. G. Celius, S. Cervenka, A. Conzelmann, A. Cordova-Palomera, A. M. Dale, D. J. F. de Quervain, P. Di Carlo, S. Djurovic, E. S. Dorum, S. Eisenacher, T. Elvsashagen, T. Espeseth, H. Fatouros-Bergman, L. Flyckt, B. Franke, O. Frei, B. Haatveit, A. K. Haberg, H. F. Harbo, C. A. Hartman, D. Heslenfeld, P. J. Hoekstra, E. A. Hogestol, T. L. Jernigan, R. Jonassen, E. G. Jonsson, P. Karolinska Schizophrenia, P. Kirsch, I. Kloszewska, K. K. Kolskar, N. I. Landro, S. Le Hellard, K. P. Lesch, S. Lovestone, A. Lundervold, A. J. Lundervold, L. A. Maglanoc, U. F. Malt, P. Mecocci, I. Melle, A. Meyer-Lindenberg, T. Moberget, L. B. Norbom, J. E. Nordvik, L. Nyberg, J. Oosterlaan, M. Papalino, A. Papassotiropoulos, P. Pauli, G. Pergola, K. Persson, G. Richard, J. Rokicki, A. M. Sanders, G. Selbaek, A. A. Shadrin, O. B. Smeland, H. Soininen, P. Sowa, V. M. Steen, M. Tsolaki, K. M. Ulrichsen, B. Vellas, L. Wang, E. Westman, G. C. Ziegler, M. Zink, O. A. Andreassen and L. T. Westlye (2019). "Common brain disorders are associated with heritable patterns of apparent aging of the brain." Nat Neurosci 22(10): 1617-1623.
[4] Leonardsen, E. H., H. Peng, T. Kaufmann, I. Agartz, O. A. Andreassen, E. G. Celius, T. Espeseth, H. F. Harbo, E. A. Hogestol, A. M. Lange, A. F. Marquand, D. Vidal-Pineiro, J. M. Roe, G. Selbaek, O. Sorensen, S. M. Smith, L. T. Westlye, T. Wolfers and Y. Wang (2022). "Deep neural networks learn general and clinically relevant representations of the ageing brain." Neuroimage 256: 119210.