Brain Tumor Detection and Classification Using Vision Transformers, Ensemble Learning, and Transfer Learning with Explainable AI Insights

Authors- Mrs. L.Yamuna, D.Surya Naga Laxmi Sahithi, K.S.S. Durga Hari Prasad3, V.Subhash, M.Hari Abhilash, Shaik Farha

Abstract-The abnormal development of either malignant or benign tissues in the brain results in long-term damage to its function. Magnetic Resonance Imaging (MRI) is a widely used technique for detecting brain tumors. Upon receiving MRI images, specialists physically examine the filters to assess whether a patient has a brain tumor. However, the interpretations of MRI images may vary between different experts, potentially leading to inconsistent results as professionals may have differing evaluation methods. Moreover, simply identifying the presence of a tumor is insufficient; it is also critical to determine the tumor type to initiate treatment promptly. This paper focuses on the multiclass classification of brain tumors, as much research has been conducted on binary classification. To enhance the speed, accuracy, and reliability of tumor detection, we explored the effectiveness of several deep learning (DL) architectures, including VGG16, InceptionV3, VGG19, ResNet50, InceptionResNetV2, and Exception. Based on this analysis, we propose a transfer learning (TL)-based multiclass classification model, IVX16, which leverages the top three performing TL models. The dataset used consists of 3,264 images, and after conducting thorough experiments, we obtained the following peak accuracies: 95.11% for VGG16, 93.88% for InceptionV3, 94.19% for VGG19, 93.88% for ResNet50, 93.58% for InceptionResNetV2, 94.5% for Exception, and 96.94% for IVX16. Additionally, we applied Explainable AI techniques to assess the performance and validity of each DL model and incorporated the recently developed Vision Transformer (ViT) models. We compared the results obtained from these ViT models with those from the TL and ensemble models.

DOI: /10.61463/ijset.vol.13.issue2.257