NeuroFusionNet: A Hybrid Deep Learning Framework for Multimodal Brain Tumor Classification

The rising cases of brain tumors like glioma, meningioma and pituitary adenoma require the establishment of automated, accurate and interpretable diagnostic systems. This paper presents a new hybrid deep learning architecture, NeuroFusionNet, a combination of Convolutional Neural Networks (CNNs) and Transformers to classify brain tumours of multimodal (using MRI (T1, T2, and FLAIR)) data. CNN backbone is effective to extract the spatial and morphological features and Transformer encoder extracts long-range dependencies and inter-modality correlations. A Hybrid Particle Swarm-Grey Wolf Optimization (HPS-GWO) algorithm is used as an adaptive layer of optimization where hyperparameters are fine-tuned and the convergence is faster and the generalization is better. Moreover, explainable AI methods like Grad-CAM offer visual interpretability, which illustrates the location of tumors that affect predictions. The experiments on the BRATS 2021 and Figshare Brain MRI data show that NeuroFusionNet has 98.57 % accuracy, which is higher than traditional CNNs and Transformer-only models. NeuroFusionNet is also a quality and transparent solution to clinical decision support and real-time diagnosis of brain tumors through the combination of hybrid fusion, adaptive tuning, and explainability.