Related Experiment Video
Updated: Feb 17, 2026
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.9K
Explainable bidirectional encoder representations from image transformers for Alzheimer's disease prediction.
Sheikh Muhammad Saqib1, Mona A Alkhattabi2, Muhammad Amir Khan3
1Department of Computing and Information Technology, Gomal University, Dera Ismail Khan, Pakistan.
Digital Health
|February 16, 2026
Summary
This study introduces an AI framework using Bidirectional-Encoder representations from Image Transformers (BEiT) for accurate Alzheimer's disease (AD) classification from MRI scans. The model achieved high accuracy, aiding early diagnosis and intervention strategies.
Area of Science:
- Artificial Intelligence in Medicine
- Neuroimaging Analysis
- Machine Learning for Diagnostics
Background:
- Alzheimer's disease (AD) causes progressive neurological decline, impacting cognition, behavior, and quality of life for patients and caregivers.
- Early and precise diagnosis of AD is crucial for implementing effective intervention strategies.
- Artificial intelligence (AI) shows significant promise in medical imaging for AD detection and classification.
Purpose of the Study:
- To develop and evaluate an explainable transformer-based AI framework for automated AD stage classification.
- To leverage Bidirectional-Encoder representations from Image Transformers (BEiT) for analyzing magnetic resonance imaging (MRI) brain scans.
- To enhance the precision of AD diagnosis through advanced machine learning techniques.
Main Methods:
- Utilized a dataset of 8511 MRI brain images across three diagnostic groups: mild, moderate, and no impairment.
- Employed BEiT as a feature extractor within the proposed AI framework.
- Addressed class imbalance using a Wasserstein generative adversarial network with gradient penalty for synthetic MRI image generation and data augmentation.
Main Results:
- Achieved outstanding classification accuracy of 96%.
- Reported high F1-scores: 0.94 (mild AD), 1.00 (moderate AD), and 0.95 (no AD).
- Demonstrated strong performance with a mean absolute error of 0.0727, Cohen's kappa of 0.9451, Matthews correlation coefficient of 0.9455, and Hamming loss of 0.0365.
Conclusions:
- The developed explainable transformer-based framework demonstrates high efficacy in classifying AD stages from MRI scans.
- The AI model's performance indicates its potential as a valuable tool for early and accurate Alzheimer's disease diagnosis.
- The study highlights the significant role of advanced AI techniques, like BEiT, in neuroimaging for neurodegenerative disease assessment.
