Interpretable Deep Learning Model for Multiclass Brain Tumor Classification

While deep learning techniques like convolutional neural networks (CNNs) have shown promise in automating brain tumor detection from magnetic resonance imaging (MRI) images, a critical gap remains in model accuracy and efficiency for multi-class model classification and model explainability. This lack of accuracy and interpretability hinders trust and adoption in clinical settings. This study introduces an innovative approach by enhancing the traditional ResNet50 architecture with advanced regularization techniques and additional convolutional layers to address these challenges. Our model improves training efficiency and robustness, achieving an impressive 98% accuracy on multi-class MRI image classification. Furthermore, integrating Gradient-weighted Class Activation Mapping(Grad-CAM) with modified ResNet50 architecture can improve patient outcomes by explaining automated brain tumor detection.