An Explainable Hybrid CNN–Transformer Framework for Multiclass Brain Tumor Classification Using MRI Imagery

Brain tumor analysis is a tough challenge, because there is a lot of variation in Magnetic Resonance Imaging (MRI) and the requirement for global context reasoning for accurate multiclass classification. This paper introduces a hybrid model for multiclass tumor classification named IntegIVRV-Net, which combines InceptionV3, ResNet50, and a Vision Transformer (ViT-L16) to address the challenges of extracting multi-scale features, conducting deep residual structural learning, and modeling global context. IntegIVRV-Net, a hybrid model combines InceptionV3 for multi-scale feature extraction, ResNet50 for deep residual structural learning and ViT-L16 for global context modeling in a single network was proposed in this paper. This work emphasizes that the current literature on CNN, Vision Transformer and hybrid segmentation/classification methods on the BraTS, Kaggle, and Figshare benchmarks sheds light on a series of limitations that hinder the reliability and deployability of these methods in clinical practice, such as the heavy dependency on computational power and data availability in transformer-based pipelines, single-metric evaluation failing to capture clinically relevant failure modes, and a lack of studies incorporating rotational invariance with global context modeling, along with a long-standing disconnect between interpretability and deployment readiness. To overcome these deficiencies, the proposed methodology includes a complete pipeline: intensity normalization and skull-striping preprocessing, three-branch feature extraction (InceptionV3, ResNet50, ViT-L16), feature fusion with the help of refinement head, explainability methods (Grad-CAM, attention), and edge–cloud deployment with encrypted transmission (ONNX/TFLite). The framework is designed to be evaluated on benchmark datasets, and the paper highlights the knowledge gaps and presents the entire proposed methodology, experimental validation, and results will be presented in a later stage of this study.