Related Experiment Video
Updated: Aug 12, 2026
DeepOmicsAE: Representing Signaling Modules in Alzheimer's Disease with Deep Learning Analysis of Proteomics, Metabolomics, and Clinical Data
Published on: December 15, 2023
Multi-modal deep learning and explainable AI for predicting multiple dementia-related neuropathologies from brain
Tamoghna Chattopadhyay1, Rudransh Kush1, Rahul H Ankarath1
1Imaging Genetics Center, Mark and Mary Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of Southern California, Marina del Rey, CA, United States.
This study uses artificial intelligence and MRI scans to predict multiple Alzheimer's disease pathologies non-invasively. The AI models accurately identified various brain changes, offering a promising alternative to invasive biomarker tests for dementia diagnosis.
Area of Science:
- Neuroimaging and Artificial Intelligence
- Neuropathology and Dementia Research
Background:
- Alzheimer's disease and related dementias (ADRD) involve complex, overlapping pathologies (e.g., amyloid-β, tau, alpha-synuclein) complicating diagnosis and treatment.
- Current biomarkers (PET, CSF) for Aβ and tau are invasive, costly, and not widely accessible.
- Structural MRI offers a non-invasive, scalable alternative for neuropathological prediction, especially when combined with AI.
Purpose of the Study:
- To develop and evaluate a hybrid deep learning framework for jointly predicting multiple ADRD pathologies from 3D T1-weighted brain MRI and other covariates.
- To assess the performance of a 3D convolutional neural network and AutoGluon for predicting six ADRD pathologies in living individuals.
- To enhance model transparency using explainable AI (XAI) methods and compare feature importance maps with traditional VBM analyses.
Main Methods:
- Developed a hybrid deep learning framework integrating 3D T1-weighted MRI with demographic, clinical, and genetic data.
- Trained and tested models using autopsy-confirmed neuropathology data from individuals scanned antemortem.
- Evaluated two machine learning models: a 3D convolutional neural network and AutoGluon, incorporating XAI techniques (OSA, Grad-CAM, IG).
Main Results:
- The AI models demonstrated strong performance in predicting multiple ADRD pathologies non-invasively using MRI and covariate data.
- Explainable AI methods provided insights into the spatial contribution of brain regions to pathology predictions.
- Feature importance maps generated by AI showed biological plausibility when compared to VBM analyses.
Conclusions:
- Multimodal, interpretable AI approaches show significant promise for comprehensive, non-invasive profiling of dementia-related pathologies.
- AI-driven MRI analysis can potentially complement or replace invasive biomarker tests for ADRD.
- This approach facilitates earlier and more accurate diagnosis and personalized treatment strategies for dementia.
