Related Experiment Video
Updated: Jan 7, 2026
05:17
Hybrid PET/MRI Imaging of Alzheimer's Disease Based on 18F-AV-1451
Published on: April 18, 2025
751
Investigating the Amyloid-Tau-Neurodegeneration Framework in Alzheimer's Disease Using Semi-Supervised Multimodal
You Cheng1,2, Adrián Medina1,3, Cole Korponay1,2
1McLean Imaging Center, McLean Hospital, Belmont, MA, USA.
Medrxiv : the Preprint Server for Health Sciences
|December 25, 2025
Summary
This study used a novel fusion method to combine brain imaging data, improving the prediction of Alzheimer's disease (AD) diagnosis and APOE4 status. The findings reveal key patterns linking amyloid and tau abnormalities to cognitive decline.
Area of Science:
- Neuroimaging
- Biomarkers
- Computational Neuroscience
Background:
- Alzheimer's disease (AD) diagnosis and prognosis are complicated by its heterogeneity.
- The amyloid-tau-neurodegeneration (A-T-N) framework offers a potential pathway to improve diagnostic prediction.
- Identifying patterns linking A-T-N abnormalities is crucial for better clinical outcomes.
Purpose of the Study:
- To apply a semi-supervised multimodal data fusion method (SuperBigFLICA) to predict cognitive decline and clinical diagnosis in Alzheimer's disease.
- To compare the predictive performance of the fusion method against traditional approaches.
- To uncover interpretable imaging patterns associated with A-T-N burden and clinical status.
Main Methods:
- Applied SuperBigFLICA to multimodal data (gray matter density, cortical thickness, surface area, amyloid PET, tau PET) from 274 ADNI-3 participants.
- Trained the model to derive latent components predictive of cognitive decline.
- Used LASSO logistic regression on subject loadings to predict diagnosis and APOE4 status, comparing against baseline models.
Main Results:
- SuperBigFLICA (SBF) loadings-based models significantly outperformed baseline models in predicting diagnosis (AUROC = 0.80) and APOE4 status (AUROC = 0.83).
- Amyloid alterations in sensory areas were key in differentiating dementia, while a multimodal A-T-N pattern related to early cognitive decline.
- Subject loadings on these patterns correlated with cerebrospinal fluid (CSF) biomarkers, linking spatial A-T-N burden to CSF AD biomarkers.
Conclusions:
- Semi-supervised multimodal fusion using SBF enhances the prediction of clinical diagnoses and APOE4 status in Alzheimer's disease.
- The method reveals interpretable imaging patterns that better predict outcomes than traditional methods.
- These findings underscore the utility of advanced data fusion techniques in understanding AD heterogeneity and improving diagnostic accuracy.
