Adaptive Structural MRI Preprocessing for Explainable Brain Tumor Classification

Abstract

Brain tumor diagnosis from magnetic resonance imaging (MRI) is a challenging problem due to low contrast, intensity variations, and the need for clinically interpretable decisions. Although deep convolutional neural networks have improved classification accuracy, many existing systems still rely on generic preprocessing and provide limited explanation for their predictions. This paper proposes a unified framework called Self-Supervised MRI Preprocessing (SSMP) for explainable brain tumor classification. The framework combines adaptive noise suppression, entropy-guided enhancement, tumor-sensitive preprocessing, lightweight deep classification, and heatmap-based visual explanation in a single pipeline.