MRI Based High-Grade and Low-Grade Brain Gliomas Classification by Using Video Vision Transformers

Gliomas, which constitute the majority of brain tumors, are among the most common and aggressive brain tumors of the central nervous system. Since gliomas are often difficult to treat and significantly affect the quality of life and survival of patients, the successful classification of glioma grade is crucial. In this study, we used the Video Vision Transformer (ViViT) approach, known for its success in object detection problems, to classify gliomas as high-grade gliomas (HGG) or low-grade gliomas (LGG) using MR images obtained from a well-known brain tumor database, BraTS-2019. Instead of utilizing all MR images, the model was trained with fewer images to efficiently represent the dataset. The proposed method performs end-to-end feature extraction and classification by eliminating the need for preprocessing steps such as merging different sequences or segmenting the tumor, which typically yields to extra processing load. In the study, we also utilized a novel loss function to address false positive (FP) and false negative (FN) situations in binary classification problems, and the obtained results were evaluated with the AUC metric, which is widely used for imbalanced data sets. To the best of our knowledge, this is the first study to apply the ViViT architecture for the classification of glioma grades. We obtained promising results (AUC: 0.9410) which suggest significant potential for future research in this area.