V3I1P18

Lightweight Explainable Deep CNN Framework for Multi-Class Brain Tumor Detection in MRI Imaging

Dr. Aditi Tulchhia^1*,  Ankit Porwal², Dr. Abhilasha Dangi³

Abstract

This research presents an explainable, lightweight deep learning framework based on Convolutional Neural Networks (CNNs) for robust multi-class brain tumor detection using Magnetic Resonance Imaging (MRI). The proposed model accurately classifies glioma, meningioma, pituitary tumors, and non-tumor brain scans with enhanced interpretability through Grad-CAM visualizations. Beyond achieving superior accuracy (98.2%), the framework emphasizes computational efficiency, transparency, and clinical readiness. Comparative analysis with benchmark architectures demonstrates improved performance across accuracy, precision, recall, F1-score, and AUC. Additional experiments integrating hybrid approaches (CNN + XGBoost, CNN + Transformers) highlight further performance gains. The system’s explainable heatmaps reinforce diagnostic trust, making it viable for real-time clinical deployment, particularly in resource-constrained settings. This work provides an extended foundation for AI-driven, interpretable diagnostic tools in neuro-oncology.

Keywords:

Brain Tumor Classification, Convolutional Neural Networks, MRI Imaging, Deep Learning, Explainable AI, Grad-CAM, Hybrid Models, Medical Imaging, Clinical Decision Support