AI-Based Hybrid Deep Learning Framework for Accurate Brain Tumor Classification from MRI Images Using CNN and Vision Transformer

Brain tumors remain among the most life-threatening neurological disorders, and their outcome depends heavily on how early and how accurately they are diagnosed. Magnetic Resonance Imaging (MRI) is the modality of choice for tumor evaluation, yet manual reading of MRI scans is slow and subject to inter-observer disagreement. This paper proposes NeuroFusionNet, a lightweight hybrid deep learning framework that fuses a MobileNetV2-based Convolutional Neural Network (CNN) branch with a Vision Transformer (ViT) branch through a Squeeze-and-Excitation (SE) gated fusion mechanism, allowing the model to combine fine-grained local texture cues with long-range global context in a single forward pass. The framework further incorporates a U-Net-based segmentation head to delineate tumor boundaries, a five-method Explainable AI (XAI) suite for transparent decision support, and a Large Language Model (LLM) module that converts model outputs into structured, human-readable clinical reports. Evaluated on the BRISC 2025 benchmark, which contains expert-annotated MRI scans spanning glioma, meningioma, pituitary tumor, and healthy classes, the proposed model achieves 96.2% test accuracy while using only 7.8 million parameters, indicating that competitive diagnostic accuracy can be obtained without the computational burden typical of large-scale transformer or ensemble models. The results suggest that NeuroFusionNet is a practical candidate for deployment in resource-constrained clinical settings that still require interpretable, end-to-end decision support.