Estimating Brain Age and Identifying Superagers Using Bayesian Neural Networks
Poster No:
1145
Submission Type:
Abstract Submission
Authors:
Carlos Silva dos Prazeres1, Alessandra C Goulart2, Claudia K Suemoto3, Paulo A Lotufo4, Isabela M Benseñor5, Claudia da Costa Leite1, Alexandre Chiavegatto Filho2, Maria Concepción García Otaduy1
Institutions:
1LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina, Sao Paulo, Sao Paulo, 2Department of Epidemiology, School of Public Health, Universidade de São Paulo, Sao Paulo, Sao Paulo, 3Division of Geriatrics, faculdade de medicina da USP, Sao Paulo, Sao Paulo, 4Center for Clinical and Epidemiological Research, Hospital Universitario, Universidade de Sao Paulo, Sao Paulo, Sao Paulo, 5Department of Internal Medicine, School of Medicine, Universidade de Sao Paulo, Sao Paulo, Sao Paulo
First Author:
Carlos Silva dos Prazeres
LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina
Sao Paulo, Sao Paulo
LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina
Sao Paulo, Sao Paulo
Co-Author(s):
Alessandra C Goulart
Department of Epidemiology, School of Public Health, Universidade de São Paulo
Sao Paulo, Sao Paulo
Department of Epidemiology, School of Public Health, Universidade de São Paulo
Sao Paulo, Sao Paulo
Paulo A Lotufo
Center for Clinical and Epidemiological Research, Hospital Universitario, Universidade de Sao Paulo
Sao Paulo, Sao Paulo
Center for Clinical and Epidemiological Research, Hospital Universitario, Universidade de Sao Paulo
Sao Paulo, Sao Paulo
Isabela M Benseñor
Department of Internal Medicine, School of Medicine, Universidade de Sao Paulo
Sao Paulo, Sao Paulo
Department of Internal Medicine, School of Medicine, Universidade de Sao Paulo
Sao Paulo, Sao Paulo
Claudia da Costa Leite
LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina
Sao Paulo, Sao Paulo
LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina
Sao Paulo, Sao Paulo
Alexandre Chiavegatto Filho
Department of Epidemiology, School of Public Health, Universidade de São Paulo
Sao Paulo, Sao Paulo
Department of Epidemiology, School of Public Health, Universidade de São Paulo
Sao Paulo, Sao Paulo
Maria Concepción García Otaduy
LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina
Sao Paulo, Sao Paulo
LIM44, Instituto e Departamento de Radiologia, Hospital das Clinicas HCFMUSP, Faculdade de Medicina
Sao Paulo, Sao Paulo
Introduction:
Using machine learning algorithms fed with magnetic resonance imaging (MRI) measures of cortical thickness and subcortical volume, it is possible to predict the brain age of an individual and to compare it to the chronological age with the aim to estimate the brain age gap (BAG) (Niu et al., 2020). The BAG can become an important biomarker to evaluate neurodegeneration and to understand brain aging in different conditions. However, the limited use of data collected in Brazil and the absence of models that estimate uncertainty in predictions restrict the ability to capture demographic particularities and compromise the reliability of the results. This study aims to predict brain age along with its uncertainty in a Brazilian sample and evaluate the usefulness of the model in a population of superagers.
Methods:
The present analysis was based on data from the participants from the Ageing and Brain Working of the ELSA-Brasil, a prospective multicenter cohort (six research centers) (Aquino et al., 2012). Participants were from São Paulo Research Center (age range: 50 to 89 years at the time of MRI). Neuroimaging data were all collected with the same 3T scanner (Achieva, Philips, Best, The Netherlands) using a 32-channel head coil. Parameters of the 3D T1w-TFE sequence are described in Figure 1. The group used to train the algorithm and test it (in proportion 80:20) included 66 participants (51-89 yrs), classified by a total z-score of -1 to 1 based on cognitive battery. The SuperAgers group consisted of 18 elderly subjects (>70 years old), with late memory performance similar to the median of younger individuals (46-54 yrs). The predictors of each individual were obtained after postprocessing of 3D T1-weighted images with FreeSurfer 7.3.2 (Fischl, 2012) and using the Destrieux atlas. All analyses were conducted in Python 3.10.6 (Python Software Foundation, 2019). Features with many missing values or primarily zeros were excluded, and brain volumes were normalized by total intracranial volume. A Pearson correlation threshold of 0.4 and a ξ correlation of 0.15 (Chatterjee, 2019) were set to retain variables with significant correlation with the Telephone Interview for Assessment of Cognitive Status (Fong et al., 2009). Variables with high Pearson correlation with each other (>0.9) were excluded. Finally, standardization of the predictor variables using three standard deviations was made as a criterion for outlier replacement. Analysis was performed using a Bayesian Neural Network (BNN) (Bishop, 1997), which was calibrated with a Gaussian Process. The calibration employed the BNN estimates and predictor variables as inputs to generate residual BNN estimates, which were then subtracted from the original BNN predictions (see Figure 1). The BAG is then determined by the difference between the individual's estimated brain age and the chronological age. In addition to evaluating the test group derived from the same population used for training, we provided an explanation of how each feature impacts the model's predictions using the SHAP method (Lundberg & Lee, 2017). The model was also assessed for a group of superagers. A comparison between these two groups was made using descriptive statistics and a Spearman's correlation coefficient between the predicted and chronological age.
Results:
The average performance of the model, after being tested on 1,000 distribution samples in each of the 10 training cycles, resulted in an RMSE of 7.076 ± 0.344 and PICP of 0.800 ± 0.074. The model was assessed in two groups: a test group with average BAG value of 0.791 ± 6.275 (ρ correlation: 0.76, p value: 1.78e-03) and the superager group with average BAG value of -5.571 ± 6.273 (ρ correlation: 0.11, p: 6.72e-01), described in Figure 2a. Most significant features are depicted in Figure 2b.
Conclusions:
The results indicate that the brain age prediction model performed satisfactorily for our sample, and it was sensitive to depict the high negative BAG characteristic for SuperAgers.
Learning and Memory:
Neural Plasticity and Recovery of Function
Lifespan Development:
Aging 2
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Cortical Anatomy and Brain Mapping
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Cognition
Computational Neuroscience
Machine Learning
MRI
Other - Superagers; Brain Age; Cortical Thickness; Bayesian Neural Network; Predictive Modeling; Subcortical Volume
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.
Chatterjee, S. (2019). A new coefficient of correlation. ArXiv (Cornell University). https://doi.org/10.48550/arxiv.1909.10140
Fischl, B. (2012). FreeSurfer. NeuroImage, 62(2), 774–781. https://doi.org/10.1016/j.neuroimage.2012.01.021
Fong, T. G., Fearing, M. A., Jones, R. N., Shi, P., Marcantonio, E. R., Rudolph, J. L., Yang, F. M., Dan Kiely, K., & Inouye, S. K. (2009). Telephone Interview for Cognitive Status: Creating a crosswalk with the Mini-Mental State Examination. Alzheimer’s & Dementia, 5(6), 492–497. https://doi.org/10.1016/j.jalz.2009.02.007
Lundberg, S., & Lee, S.-I. (2017, November 24). A Unified Approach to Interpreting Model Predictions. ArXiv.org. https://doi.org/10.48550/arXiv.1705.07874
Niu, X., Zhang, F., Kounios, J., & Liang, H. (2020). Improved prediction of brain age using multimodal neuroimaging data. Human Brain Mapping, 41(6), 1626–1643. https://doi.org/10.1002/hbm.24899
Passos, V. M. de A., Caramelli, P., Benseñor, I., Giatti, L., & Barreto, S. M. (2014). Methods of cognitive function investigation in the Longitudinal Study on Adult Health (ELSA-Brasil). Sao Paulo Medical Journal, 132(3), 170–177. https://doi.org/10.1590/1516-3180.2014.1323646
Python Software Foundation. (2019). Welcome to Python.org. Python.org; Python.org. https://www.python.org/doc/