Multiclass brain tumor classification using vision transformer

Brain tumours are a dangerous disease that affects humans. The type and location of the tumour will largely determine the course of medical treatment. Developing an automated method is crucial for early detection and diagnosis. Despite the various methods proposed by researchers, prediction accuracy has always remained a challenge. This research presents a brain tumour classification model based on a vision transformer (ViT) and a sequential classifier. The dataset was obtained from Kaggle for the purpose of training and testing the network. The data undergoes preprocessing to align with the ViT model, whereby pictures are segmented into patches and analysed using self-attention layers for feature extractions. Then the sequential classifier is used for classification of tumor images. The efficacy is evaluated by employing accuracy, F1 score, recall, and precision. In comparison to deep transfer learning models, the suggested model has an accuracy of 99.00%.