Early Alzheimer's risk detection via diffusion tensor imaging using a few-shot multichannel attention residual
Arpit Shet1, Saad Sabahuddin1, Priyadarshini B1
1School of Electronics Engineering, Vellore Institute of Technology, Chennai, India.
Frontiers in Artificial Intelligence
|June 30, 2026
Summary
Early Alzheimer's disease detection is improved with a new deep learning model using diffusion tensor imaging (DTI). This Multichannel Attention Residual Learning-DTI (MARL-DTI) framework accurately identifies Mild Cognitive Impairment (MCI), a precursor to AD.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Diagnostics
Background:
- Alzheimer's disease (AD) is a leading cause of dementia, with Mild Cognitive Impairment (MCI) as its earliest sign.
- Early MCI detection is crucial for timely intervention and improved patient quality of life.
- Diffusion Tensor Imaging (DTI) offers high sensitivity to white matter microstructural changes, surpassing traditional structural imaging for MCI detection.
Purpose of the Study:
- To develop and evaluate a novel deep learning architecture, Multichannel Attention Residual Learning-DTI (MARL-DTI), for the early detection of MCI using DTI data.
- To leverage attention mechanisms, residual learning, and few-shot learning to enhance model performance with limited neuroimaging data.
- To ensure model transparency and clinical relevance through explainable AI techniques.
Main Methods:
- A Multichannel Attention Residual Learning-DTI (MARL-DTI) model was designed, incorporating attention mechanisms and residual learning.
- The model utilized multiple DTI-derived metrics: fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AxD).
- A few-shot learning framework was employed to address data scarcity, and Local Interpretable Model-Agnostic Explanations (LIME) were used for model interpretability.
Main Results:
- The MARL-DTI model achieved a classification accuracy of 92.76% on the primary dataset.
- Consistent high accuracy (91.67%) was observed on an independent, unseen dataset, confirming model reliability.
- LIME analysis provided instance-level explanations, detailing feature contributions and enabling qualitative assessment of model decisions.
Conclusions:
- The MARL-DTI framework demonstrates superior performance compared to existing deep learning models for early MCI detection using DTI.
- The model's consistent accuracy and transparency, validated by LIME, support its potential for reliable clinical application.
- Combining multi-channel DTI data with explainable AI offers a promising avenue for early and accurate Alzheimer's disease diagnosis.
