An Explainable and Computationally Efficient Deep Learning Framework for Brain Tumor Classification from MRI Images

Abstract

Brain tumor classification has become an important research area in modern medical diagnostics because it helps neuro-oncologists monitor pathological growth and patient brain health efficiently. Traditional diagnostic methods rely heavily on manual observation, where radiologists visually inspect multi-sequence MRI scans to detect brain tumors, growth patterns, and tissue abnormalities. This manual method requires significant time, effort, and expert labor. It also increases the possibility of human error because small changes in tissue texture or subtle tumor margins may not be noticed immediately. As medical imaging data volume increases, manual monitoring becomes more difficult and inefficient. Recent advancements in computer vision and artificial intelligence have introduced promising approaches for automated brain tumor classification. These technologies enable deep learning models to analyze MRI images and learn visual patterns associated with different tumor categories. The proposed framework is designed to integrate image preprocessing techniques, including noise reduction, contrast enhancement, and skull stripping, with lightweight convolutional feature extraction. It also incorporates Grad-CAM to provide visual explanations of the classification process. The proposed framework aims to support medical image analysis by combining computationally efficient feature extraction with visual interpretability. It considers four classification categories: Glioma, Meningioma, Pituitary Tumor, and No Tumor. The methodology establishes a foundation for future implementation and experimental evaluation in computer-assisted medical image analysis.