Explainable AI-driven MRI-based brain tumor classification: a novel deep learning approach

Introduction: Brain tumors are among the most aggressive forms of cancer, requiring precise diagnosis and treatment planning to improve patient outcomes. This study aims to develop an efficient deep learning-based framework for the classification of brain tumors using MRI data. Methods: The methodology employs Convolutional Neural Networks (CNNs) to accurately classify tumors into four categories: normal, glioma, pituitary, and meningioma. Key preprocessing techniques, including noise reduction,resizing, and data augmentation, were applied to enhance the robustness of the model. Advanced architectures such as DenseNet50, VGG19, and other transfer learning models, along with CNN variants, were trained and evaluated for their performance. Explainable AI (XAI) techniques, including Grad-CAM, LIME, and feature map visualizations, played a crucial role in providing better visualizations of the model's decision-making process and identifying areas of improvement during model training and to establish a better model. Results: The best-performing model, a 4-conv-1-dense-1-dropout CNN, achieved a classification accuracy of 95.86%, outperforming deeper architectures and transfer learning approaches. The findings underscore the potential of deep learning models for reliable and efficient brain tumor classification. This work concludes with recommendations for real-time deployment in clinical settings and explores future integration with Large Language Models (LLMs) to generate detailed diagnostic reports.