BrainFusion: Deep Learning for Early Detection and Grading of Brain Tumors Using Medical Imaging and Clinical Data

Abstract

Brain tumors are still one of the most dangerous and deadly types of cancer. Early detection and correct grading are very important for effective treatment. This study presents a dual-stage artificial intelligence framework that utilizes advanced transformer-based deep learning architectures, specifically the Swin Transformer and Vision Transformer (ViT), for the early detection and grading of brain tumors through MRI images and clinical metadata. Our approach takes advantage of these transformer models' better feature extraction and contextual representation, which is not the case with traditional CNN-based methods. The framework consists of two core components: (1) a transformer-based classifier for detecting tumor presence in 2D MRI slices, and (2) a multimodal fusion model combining imaging features and clinical data (age, gender, tumor type, grade) to predict tumor grade (HGG vs LGG). We also made a web interface powered by Gradio that lets users upload images and get a diagnosis in real time. This makes the model available to both researchers and doctors. We use publicly available brain MRI datasets like BraTS and TCGA to create and test our model. show that the Swin Transformer is better than current CNN-based methods at finding tumors and classifying grades, with an accuracy of up to 98.3% and 95.2%, respectively. These findings validate the effectiveness and promise of transformer architectures and web-based AI tools in enhancing neuro-oncological diagnosis.