Scholarlink

Brain tumors, particularly those of the highest grade, can substantially lower life expectancy. A precise diagnosis is required for individuals with brain tumors to receive the best possible care. The conventional way of diagnosing brain tumors is Magnetic Resonance Imaging (MRI) while recent tumor detection models leverage Convolution Neural Networks (CNN) and pre-trained models, many prioritize accuracy over providing explanation and interpretability. This paper addresses these shortcomings by proposing an innovative approach to the categorization of brain tumors by exploring transfer learning and Explainable Artificial Intelligence (XAI). The study uses the top five pre-trained models to classify three different types of tumors: VGG16, ResNet50, Xception, InceptionV3 and DenseNet201. Notably, a customized VGG16 model obtains a promising test accuracy of 99.04%, which is comparable to top literature scores. A significant contribution of this work is the integration of the top three explainable AI tools: Gradient-weighted Class Activation Mapping (Grad-CAM), Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). These tools enhance visualization and provide valuable explanations for model predictions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MRI Image-Based Brain Tumor Classification Using Transfer Learning and XAI

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Masum Rayhan,
  • Saykat Mondal,
  • Farhana Tazmim Pinki

摘要

Brain tumors, particularly those of the highest grade, can substantially lower life expectancy. A precise diagnosis is required for individuals with brain tumors to receive the best possible care. The conventional way of diagnosing brain tumors is Magnetic Resonance Imaging (MRI) while recent tumor detection models leverage Convolution Neural Networks (CNN) and pre-trained models, many prioritize accuracy over providing explanation and interpretability. This paper addresses these shortcomings by proposing an innovative approach to the categorization of brain tumors by exploring transfer learning and Explainable Artificial Intelligence (XAI). The study uses the top five pre-trained models to classify three different types of tumors: VGG16, ResNet50, Xception, InceptionV3 and DenseNet201. Notably, a customized VGG16 model obtains a promising test accuracy of 99.04%, which is comparable to top literature scores. A significant contribution of this work is the integration of the top three explainable AI tools: Gradient-weighted Class Activation Mapping (Grad-CAM), Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). These tools enhance visualization and provide valuable explanations for model predictions.