Vision Transformer-based Glioma Classification using Multi-Modal MRI and Wavelet Fusion

Gliomas are a common type of brain tumors, and they present very challenging issues in medical imaging and categorization. Magnetic Resonance Imaging (MRI) is an essential noninvasive diagnostic method for the detection and evaluation of gliomas. Nevertheless, manual segmentation of tumors is time-consuming and prone to human errors, calling for the development of automated methods. This study implements a deep learning model for automatic glioma detection utilizing the BraTS 2024 dataset, which comprises multi-modal MRI scans (T1, T1c, T2, FLAIR) and their corresponding segmentation masks, using a Vision Transformer (ViT). The methodology focuses on conducting tasks sequentially, which includes: Data Pre-processing, where images are sliced into 2D parts, normalized and resized, Feature Extraction, where imaging is fused using Wavelet Transformation and texture is analyzed using a Gray-Level Co-occurrence Matrix (GLCM), and Glioma Classification, using vision transformer model is used to incorporate the spatial information of the MRI scans. The experimental results showed that accuracy, were all superior, proving the Vision transformer model was more effective than traditional CNN models. This study confirms the promise of transformer models in medical imaging by providing a robust and effective solution for glioma detection, thereby improving the accuracy of diagnosis in neuron-oncology.