Document Type : Full Research Paper
Authors
1
Department of Biomedical Engineering, Ma.C., Islamic Azad University, Mashhad, Iran.
2
Department of Biomedical Engineering, Ma.C., Islamic Azad University, Birjand, Iran.
3
Department of Neurology, School of Medicine, Neurology Research Center, Birjand University of Medical Sciences, Birjand, Iran.
4
Department of Biomedical Engineering, Bi.C., Islamic Azad University, Birjand, Iran.
10.22041/ijbme.2026.2089964.2037
Abstract
Early diagnosis of brain Arteriovenous Malformation (AVM) plays a vital role in preventing irreversible
complications; however, standard diagnostic methods such as angiography are often invasive and costly,
and accurate detection of this condition in conventional MRI images is difficult even for specialists due to
subtle tissue differences. The present research was conducted with the aim of developing an intelligent
system for automatic AVM and other Arteriovenous Malformation diagnosis based on standard Brain MRI
images (T2-weighted and FLAIR) without the need for supplementary MRA data. In this study, a dataset
comprising 538 cases (274 patients and 264 healthy individuals) was collected from Valiasr Hospital in
Birjand, and after applying preprocessing and selecting key slices, the performance of several advanced
Vision Transformer (ViT) architectures, including Google ViT, Microsoft BEiT, and Facebook ViT, was
evaluated using a voting mechanism. The results demonstrated that the Google ViT model, achieving 100%
sensitivity (Recall) and an F1-score of 83.9%, delivered the superior performance and succeeded in
identifying all positive cases with zero false negatives, surpassing the performance of human experts. By
identifying hidden patterns and secondary changes in brain tissue, this model provides an accurate, rapid,
and non-invasive tool for AVM screening that can be utilized as a complement or alternative to complex
diagnostic methods in clinical settings.
Keywords
Subjects
