Brain Tumor Classification on MRI Images with Big Transfer and Vision Transformer: Comparative Study

• Main Authors: Jahangir, Raiyan, Sakib, Tanjim, Juboraj, Md Fahmid-Ul-Alam, Feroz, Shejuti Binte, Sharar, Md Munkaser Islam
• Format: Conference Proceeding
• Language: English
• Subjects: Big Transfer; Brain modeling; Brain Tumor; Convolutional neural networks; Machine learning; Magnetic resonance imaging; Medical services; MRI; Training; Transformers; Vision Transformer
• ISSN: 2837-8245
• DOI: 10.1109/WIECON-ECE60392.2023.10456372

A brain tumor is a severe neurological condition that happens because of the uncontrolled growth of cells inside the brain or skull. The number of deaths because of this condition is increasing at an abrupt rate. That is why early diagnosis and treatment of brain tumors are important. If not treated timely, the brain tumors can worsen the situation and can result in death. MRI images of the brain are inspected to detect and classify brain tumors. However, such tasks are carried out by physicians manually, which takes time. Therefore, automated approaches are necessary to make this task easier. Machine Learning (ML) and Convolutional Neural Network (CNN) models have been used to identify and classify brain tumors on MRI images. But with time, new technologies are developed which are expected to replace the existing ones. Two such state-of-the-art technologies are Big Transfer (BiT) and Vision Transformer (ViT). The use of these technologies in brain tumor classification is still scant. Therefore, this research aims to classify brain tumors with the help of these two technologies and evaluate their performances based on precision, recall, and f1-score. Finally, the results are compared, which shows that Big Transfer (BiT) performs better than Vision Transformer (ViT) for both training (100% precision, 100% recall, 100% f1-score) and test data (95.928% precision, 95.922% recall, 95.924 % f1-score) in classifying brain tumors.