Advances in Brain Tumor Diagnosis Through Multimodal Imaging and Deep Learning Techniques

Brain tumors represent abnormal growths of cells—either benign or malignant—within the confined space of the human skull. These growths can lead to severe health complications and long-term psychological and physical consequences, greatly diminishing a patient's quality of life. The survival rate for individuals diagnosed with malignant brain tumors decreases significantly over time if early interventions are not implemented. This study investigates advanced hybrid architectures, namely CNN-UNet, ResNet-UNet, EfficientNet-UNet, and EfficientNet integrated with Vision Transformers. A comprehensive comparative analysis of these hybrid architectures is conducted to recognize the most efficient model. The results indicate that EfficientNet-UNet architecture achieved superior performance on the BraTS dataset with an accuracy of 99.47%, while EfficientNet-Vision Transformer on the Nickparvar dataset from Kaggle reached an accuracy of 97%. These findings highlight the efficiency of advanced hybrid deep learning models in detecting brain tumors from multimodal MRI images.