Brain Tumor Detection using UNET and Classification using Transformers

Abstract

Patients can be recovered easily if the any tumor is at its earlier stage. In this study brain tumor is classified using Magnetic Resonance Imaging (MRI) scan of the brain. Previously there is need of huge medical staff as well as more time for brain tumor diagnosis. Recently there is advancement in the medical diagnosis which makes process automatic diagnosis and it is able to classify MRI scan images. Many machine learning techniques has been implemented for brain tumor diagnosis but there are some limitations such as huge time required for training and lower accuracy. So, to overcome the limitations of existing techniques in this project 3D UNET is used for identification of location of tumor. Transformer based deep learning model is used for classification of tumor into 4 different categories. Performance of proposed algorithm with existing algorithm such as VGG16 is compared to show the superiority of proposed transformer-based model. Proposed model based on transformers is more efficient than state of art techniques which is measured using accuracy, precision, recall and F-score. To show the performance comparison between two algorithms bar graph is plotted which shows the superior performance of proposed transformer-based model for brain tumor classification.