Page 74 - AIH-2-1

Artificial Intelligence in Health

ORIGINAL RESEARCH ARTICLE
Vision transformers for glioma classification

using T1 magnetic resonance imaging

W. M. S. P. B. Wickramasinghe ,and Maheshi B. Dissanayake*

Department of Electrical and Electronics Engineering, Faculty of Engineering, University of
Peradeniya, Peradeniya, Sri Lanka
(This article belongs to the Special Issue: Artificial intelligence for diagnosing brain diseases)

Abstract

Automated image analysis and classification have increasingly advanced in recent
decades owing to machine learning and computer vision. In particular, deep learning
(DL) architectures have become popular in resource-limited and labor-restricted
environments such as the health-care sector. Transformer architecture, a DL method
with self-attention mechanism, excels in natural language processing; however, its
application in image-based diagnosis in health-care sector remains limited. Herein,
the feasibility, bottlenecks, and performance of transformers in magnetic resonance
imaging (MRI)-based brain tumor classification were investigated. To this end, a
vision transformer (ViT) model was trained and tested using the popular Brain Tumor
Segmentation (BraTS) 2015 dataset for glioma classification. Owing to limited data
*Corresponding author: availability, domain adaptation techniques were used to pretrain the ViT model and
Maheshi B. Dissanayake
(maheshid@eng.pdn.ac.lk) the BraTS 2015 dataset was used for its fine-tuning. With the model only trained for
100 epochs, the confusion matrix for the two-class problem of tumor and nontumor
Citation: Wickramasinghe,
WMSPB and Dissanayake MB. classification showed an overall classification accuracy of 81.8%. In conclusion,
Vision transformers for glioma although convolutional neural networks are traditionally used for DL-based
classification using T1 magnetic medical image classification owing to their attention mechanism and long-range
resonance imaging. Artif Intell
Health. 2025;2(1):68-80. dependency-capturing capability, ViTs can outperform them in MRI-based brain
doi: 10.36922/aih.4155 tumor classification.
Received: July 5, 2024
1st revised: August 28, 2024 Keywords: Vision transformers; Medical image analysis; Deep neural networks; Magnetic
resonance imaging; Convolutional neural network; Glioma detection
2nd revised: September 9, 2024
Accepted: September 19, 2024
Published Online: November 6,
2024 1. Introduction
Copyright: © 2024 Author(s). Medical imaging is crucial in the health-care sector for noninvasive diagnostic procedures
This is an Open-Access article because it can provide functional and visual representations of internal organs. X-ray
distributed under the terms of the
Creative Commons Attribution imaging, nuclear imaging, magnetic resonance imaging (MRI), mammography,
License, permitting distribution, computed tomography (CT), and ultrasound imaging are some popular imaging
and reproduction in any medium,
1
provided the original work is techniques. The four primary phases of medical image analysis include image formation,
properly cited. reconstruction, processing, and analysis. These phases help to create two-dimensional
Publisher’s Note: AccScience and three-dimensional (3D) images and enhance them; quantitative data are used for
Publishing remains neutral with segmentation, classification, and object identification. Modern advancements in
2,3
regard to jurisdictional claims in
published maps and institutional artificial intelligence (AI), computer vision (CV), machine learning, and deep learning
affiliations. (DL) techniques can qualitatively and quantitatively improve medical image analysis.

Volume 2 Issue 1 (2025) 68 doi: 10.36922/aih.4155