Forecasting Brain Trajectories: Adaptive and Interpretable Long-Term Prediction in Aging and AD
Poster No:
1514
Submission Type:
Abstract Submission
Authors:
Vasiliki Tassopoulou1, Haochang Shou1, Christos Davatzikos2
Institutions:
1University of Pennsylvania, Philadelphia, PA, 2Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA
First Author:
Co-Author(s):
Introduction:
Longitudinal neuroimaging studies offer critical insights into the evolving structure of the human brain over time, capturing patterns of development, aging, and disease progression. Modeling these trajectories is crucial for understanding disorders such as Alzheimer's disease (AD), evaluating treatment effects, and informing clinical decisions and trial design. Yet, this task is challenging due to inherent biological variability, limited follow-ups, irregular time intervals between measurements, and potential scanner inconsistencies. Traditional modeling approaches often lack the flexibility and scalability needed to capture complex, high-dimensional neuroimaging data, while conventional deep learning frameworks may struggle with interpretability and adaptability over extended time horizons.
Here, we present a novel, interpretable Deep Kernel Regression framework tailored for forecasting long-term trajectories of brain biomarkers from baseline imaging and clinical data. Our approach integrates population-level patterns with subject-specific data to predict future brain changes, including atrophy rates and composite neuroimaging biomarkers. Central to our framework is the Adaptive Posterior Shrinkage Estimation technique that seamlessly combines a population-level DKGP model with a subject-specific component, dynamically adjusting the weighting between these two predictive distributions as new follow-up acquisitions become available.
Here, we present a novel, interpretable Deep Kernel Regression framework tailored for forecasting long-term trajectories of brain biomarkers from baseline imaging and clinical data. Our approach integrates population-level patterns with subject-specific data to predict future brain changes, including atrophy rates and composite neuroimaging biomarkers. Central to our framework is the Adaptive Posterior Shrinkage Estimation technique that seamlessly combines a population-level DKGP model with a subject-specific component, dynamically adjusting the weighting between these two predictive distributions as new follow-up acquisitions become available.
Methods:
Our framework leverages Deep Kernel Gaussian Processes (DKGPs) (Wilson, 2016) (Tassopoulou, 2022) to integrate both population-level and subject-specific modeling. First, we train a population-level DKGP (p-DKGP) on a large, heterogeneous dataset to learn a latent transformation that maps high-dimensional imaging and clinical features into a lower-dimensional space predictive of biomarker progression. As follow-ups for a new subject become available, we train a subject-specific DKGP (ss-DKGP) using the fixed transformation from the population model. While the p-DKGP captures global trends, the ss-DKGP adapts predictions to each individual's trajectory. We then introduce the Adaptive Posterior Shrinkage Estimation technique to determine a parameter α, which governs how the final prediction-referred to as pers-DKGP-balances population-level and subject-specific information. A conceptual overview of this approach is presented in Figure 1B.
Results:
We use T1-weighted MRI and clinical data from the iSTAGING consortium (Habes, 2021). On ADNI and BLSA test data, our method outperforms Linear Mixed Effects, Generalized Additive Models, and deep learning baselines in predicting ROI trajectories across different progression statuses. Applying it to composite neuroimaging biomarkers, SPARE-AD and SPARE-BA, further demonstrates versatility (Figure 2A). We also test on OASIS, AIBL, and PreventAD studies without retraining. Our method maintains strong performance, showing robust generalization across demographics and follow-up intervals. (Figure 2B). Interpretability analyses confirm that α behaves sensibly. As follow-ups accumulate, α declines, increasing reliance on the subject-specific trajectory. This data-driven personalization is transparent and clinically meaningful. (Figure 2C)
Conclusions:
We introduced an interpretable, adaptive framework that integrates population-level and subject-specific Gaussian Processes to forecast long-term brain trajectories. By adjusting its predictions as individual follow-up data emerge, our method achieves personalized, biologically meaningful predictions of neuroimaging biomarkers. The approach outperforms traditional statistical and deep learning baselines, generalizes to external cohorts, and provides a transparent mechanism for understanding model decisions. As a tool for clinical research and patient management, this framework can enhance disease modelling, trial design, and therapeutic decision-making in aging and Alzheimer's disease.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Bayesian Modeling
Classification and Predictive Modeling
Methods Development 1
Keywords:
Computational Neuroscience
Data analysis
Informatics
Machine Learning
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?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
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?
Provide references using APA citation style.
Habes, M. (2016). Advanced brain aging: Relationship with epidemiologic and genetic risk factors, and overlap with Alzheimer disease atrophy patterns. Translational Psychiatry, 6(4), e775. https://doi.org/10.1038/tp.2016.39
Habes, M. (2021). The brain chart of aging: Machine-learning analytics reveals links between brain aging, white matter disease, amyloid burden, and cognition in the iSTAGING consortium of 10,216 harmonized MR scans. Alzheimer’s & Dementia, 17(1), 89–102. https://doi.org/10.1002/alz.12172
Tassopoulou, V. (2022). Deep kernel learning with temporal Gaussian processes for clinical variable prediction in Alzheimer’s disease. Proceedings of the 2nd Machine Learning for Health Symposium, Proceedings of Machine Learning Research, 193, 539–551. Available from https://proceedings.mlr.press/v193/tassopoulou22a.html
Wilson, A. G. (2016). Deep kernel learning. Proceedings of the 19th International Conference on Artificial Intelligence and Statistics, Proceedings of Machine Learning Research, 51, 370–378. Available from https://proceedings.mlr.press/v51/wilson16.html