Page 39 - AIH-2-4

Artificial Intelligence in Health

ORIGINAL RESEARCH ARTICLE
Deep vision transformers in neurodegenerative

disease diagnosis using F-fluorodeoxyglucose
18
positron emission tomography scans and

anatomical brain atlas

Pooriya Khorramyar* , Amira Soliman , Farzaneh Etminani , and Stefan Byttner
Center for Applied Intelligent Systems Research in Health (CAISR Health), The School of Information
Technology, Halmstad University, Halmstad, Halland, Sweden
(This article belongs to the Special Issue: Artificial intelligence for diagnosing brain diseases)

Abstract

This research explores adapting vision transformers (ViTs) to classify
neurodegenerative diseases while ensuring their decision-making process is
interpretable. We developed a model to classify F-fluorodeoxyglucose ( F-FDG)
18
18
positron emission tomography (PET) brain scans into three categories: cognitively
normal (CN), mild cognitive impairment (MCI), and Alzheimer’s disease (AD). The
*Corresponding author:
18
Pooriya Khorramyar dataset utilized in this research contains 580 samples of F-FDG PET scans obtained
(pookho20@student.hh.se) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The proposed model
Citation: Khorramyar P, Soliman A, obtained an F1 score of 81% (macro-average of all classes) on the test dataset, a
Etminani F, Byttner S. Deep vision significant performance improvement compared to the literature. Furthermore,
transformers in neurodegenerative we combined the model’s attention maps with the Automated Anatomical Atlas 3
disease diagnosis using
18F-fluorodeoxyglucose positron (AAL3), which represents a digital brain map, to identify the most influential areas on
emission tomography scans and the model’s predictions and to conduct a regions’ importance study as a step toward
anatomical brain atlas. Artif Intell explainability. We demonstrated that ViTs can achieve competitive performance
Health. 2025;2(4):33-46.
doi: 10.36922/AIH025140026 compared to convolutional neural networks while enabling the development of
explainable models without extra computations due to the attention mechanism.
Received: March 31, 2025
1st revised: April 12, 2025
Keywords: Vision transformer; Neurodegenerative disease; F-FDG PET; Medical image
18
2nd revised: May 22, 2025 analysis; Brain scan; Deep neural network
Accepted: May 26, 2025
Published online: June 19, 2025
Copyright: © 2025 Author(s). 1. Introduction
This is an Open-Access article
distributed under the terms of the Neurodegenerative diseases (NDDs) lead to progressive deterioration and death
Creative Commons Attribution of neurons, damaging the nervous system and brain. Affecting more than 55
License, permitting distribution, and
reproduction in any medium, which million patients with a yearly increase rate of 10 million new cases worldwide,
provided that the original work is NDDs are a prominent cause of disability and death. In addition, Alzheimer’s
1
properly cited. disease (AD), as the most widespread form, accounts for 70% of NDD cases and
1
Publisher’s Note: AccScience plays a significant role in these statistics. Although NDDs have a heavy impact
Publishing remains neutral with on healthcare systems and patients’ lives, they remain incurable as of today.
1
regard to jurisdictional claims in
published maps and institutional However, timely diagnosis is pivotal in disease management and improving the
affiliations. patient’s quality of life. 2

Volume 2 Issue 4 (2025) 33 doi: 10.36922/AIH025140026