FViT: Finetuned Vision Transformer for Improved Brain Tumor Classification

Timely detection and treatment of brain tumors (BT) remain a critical challenge due to their rapid growth and complex morphologies. Recent deep learning (DL) approaches and large vision models (LVMs) have promising potential in early and accurate diagnosis of BTs. However, the optimal tuning of hyperparameters is crucial for increasing the model’s efficacy and robustness. Large vision models have significant potential for improving the accuracy of diagnosing BT using MRI images. Additionally, we evaluated and compared the performance of recent DL models ResNet50, VGG16, and Long Short-Term Memory (LSTM) using BT MRI dataset. Through rigorous experimentation, our results highlight the superior diagnostic capabilities of the FViT model with an accuracy (ACC) of 99.68% and an F1-score (FS) of 99.68% for BT classification (using training-validation set). This result is also verified with high ACC of 99.31% and FS of 99.31%using testing set. The study demonstrates how well FViT model performs in classifying BT accurately, showcasing its potential in clinical settings. These findings emphasize the critical role of fine-tuning LVM models in advancing BT diagnosis through brain MRI images.