Longitudinal Normative Modeling in Dementia with Generative Models
Poster No:
1517
Submission Type:
Abstract Submission
Authors:
Vasiliki Tassopoulou1, Sai Spandana Chintapalli1, 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:
Normative modeling is a powerful approach for understanding deviations from typical trajectories in clinical populations. Cross-sectional normative modeling focuses on removing disease effects from baseline imaging data, thereby aligning them closer to a normative distribution. Longitudinal normative modeling extends this by predicting how individual brain regions change over time, utilizing evidence only from healthy controls.
In this work, we propose a composite framework for longitudinal normative modeling composed of two essential steps. First, a cross-sectional normative model corrects disease effects from baseline MRI, enabling a baseline that closely resembles healthy controls. Second, this corrected baseline is used as input to a longitudinal normative model, which predicts normative ROI trajectories over time. Both steps are critical; using only the longitudinal model trained on healthy controls is insufficient when baseline images contain disease effects, as expected in a demented population.
In this work, we propose a composite framework for longitudinal normative modeling composed of two essential steps. First, a cross-sectional normative model corrects disease effects from baseline MRI, enabling a baseline that closely resembles healthy controls. Second, this corrected baseline is used as input to a longitudinal normative model, which predicts normative ROI trajectories over time. Both steps are critical; using only the longitudinal model trained on healthy controls is insufficient when baseline images contain disease effects, as expected in a demented population.
Methods:
We employed two complementary normative modeling approaches: a cross-sectional normative model and a longitudinal normative model. The cross-sectional model, based on generative adversarial networks (GANs), was used to correct for disease-specific effects in baseline imaging data. The corrected baseline served as input to the longitudinal normative model, a deep kernel Gaussian process regression model (Tassopoulou, 2022). This model was trained exclusively on healthy control trajectories to predict region-of-interest (ROI) trajectories over time. Data from the iSTAGING consortium (Habes, 2021) were utilized for training and evaluation. We excluded the Healthy Controls from OASIS dataset for evaluation of the longitudinal normative model and the OASIS MCI/Demented subjects for showcasing our method in reconstructing normative trajectories.
Results:
The results demonstrate the effectiveness of combining cross-sectional and longitudinal normative models. The cross-sectional GAN successfully aligned the distributions of ROIs closer to the normative range, as evidenced by Figure 1A, where the distribution of the hippocampus with the reconstructed baselines is shifted closer to zero and thus aligns more with the distribution of healthy controls. On the contrary, the distribution from the original demented baseline is more negative, indicating greater atrophy in the hippocampus and a higher distance from the healthy control distribution.
Additionally, as shown in Figure 1B, the predicted trajectory using the reconstructed baseline deviates more from the uncorrected longitudinal samples, indicating that our composite normative modeling framework shifts the whole trajectory closer to the healthy distribution. This is also qualitatively depicted in Figure 2, where we demonstrate this trajectory shift in 10 OASIS demented subjects. These findings quantitatively and qualitatively confirm that the cross-sectional correction facilitates more accurate longitudinal normative modeling.
Additionally, as shown in Figure 1B, the predicted trajectory using the reconstructed baseline deviates more from the uncorrected longitudinal samples, indicating that our composite normative modeling framework shifts the whole trajectory closer to the healthy distribution. This is also qualitatively depicted in Figure 2, where we demonstrate this trajectory shift in 10 OASIS demented subjects. These findings quantitatively and qualitatively confirm that the cross-sectional correction facilitates more accurate longitudinal normative modeling.
Conclusions:
This study demonstrates the utility of integrating cross-sectional and longitudinal normative models to study patterns of neurodegeneration. The cross-sectional GAN effectively corrected disease-related baseline deviations, while the longitudinal Gaussian process model generated reliable ROI trajectories. Together, these models offer a robust framework for disentangling normal aging from disease progression. This approach holds promise for reconstructing healthy trajectories from MCI and demented subjects and simulating longitudinal brain trajectories of subjects as if they were healthy controls.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Bayesian Modeling
Methods Development 1
Keywords:
Computational Neuroscience
Data analysis
Degenerative Disease
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.
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, in Proceedings of Machine Learning Research, 193:539-551. Available from https://proceedings.mlr.press/v193/tassopoulou22a.html.