A Performance Analysis of Brain Tumor Classification from MRI Images using Vision Transformers and CNN-based Classifiers

Abnormal growth of cells inside the brain or skull may result in brain tumors, a deadly neurological condition. Every year, the number of people dying of this condition is increasing. Early identification and treatment of brain tumors are necessary because, if not treated timely, the condition of the tumors deteriorates, which can indeed lead to death. MRI images are inspected by physicians to identify and classify brain tumors. However, automated approaches can make the identification and classification process easier. Recently, Machine Learning (ML) and Convolutional Neural Networks (CNNs) have been used to identify and classify brain tumors from MRI images. The CNN-based classifiers have already shown promising results in classifying brain tumors. Recently, Vision Transformers (ViT) evolved as the new alternative to CNN-based models on image classification tasks. The use of ViTs in brain tumor classification is still uncommon. Therefore, the objectives of this research are to train some ViT and CNN models to classify brain tumors and evaluate their performances based on precision, recall, and f1-score. The results showed that both ViT and CNN-based classifiers have satisfactory performances, with ViT showing better performance on train data (100% precision, 100% recall, 100% f1-score) and CNN-based classifiers showing better performance on test data (98.233% precision, 98.271% recall, 98.234% f1-score).