Scholarlink

This study proposes a custom Convolutional Neural Network (CNN) model for multiclass classification of brain tumors using MRI images, with comparative analysis against transfer learning architectures VGG16 and VGG19. The proposed work here, achieved an average accuracy of 99.64%, surpassing the pre-trained counterparts. To enhance model transparency and clinical trust, Explainable AI (XAI) techniques, specifically Local Interpretable Model-agnostic Explanations (LIME), were integrated to visualise the decision-making process. LIME highlighted tumor-relevant regions within MRI scans, offering interpretability and aiding in clinical validation. An interactive Streamlit-based interface was also developed to allow real-time image classification and explanation visualisation. The results demonstrate that the proposed approach provided better classification accuracy and valuable interpretability, making it a reliable tool for computer-aided diagnosis (CAD) in medical imaging.

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

Interpretable Deep Learning for Multiclass Brain Tumor Classification Using MRI and LIME

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ganesh Prasad Sahoo,
  • Chinmayee Dora,
  • Sunil Kumar Mohapatra,
  • Sujata Chakravarty

摘要

This study proposes a custom Convolutional Neural Network (CNN) model for multiclass classification of brain tumors using MRI images, with comparative analysis against transfer learning architectures VGG16 and VGG19. The proposed work here, achieved an average accuracy of 99.64%, surpassing the pre-trained counterparts. To enhance model transparency and clinical trust, Explainable AI (XAI) techniques, specifically Local Interpretable Model-agnostic Explanations (LIME), were integrated to visualise the decision-making process. LIME highlighted tumor-relevant regions within MRI scans, offering interpretability and aiding in clinical validation. An interactive Streamlit-based interface was also developed to allow real-time image classification and explanation visualisation. The results demonstrate that the proposed approach provided better classification accuracy and valuable interpretability, making it a reliable tool for computer-aided diagnosis (CAD) in medical imaging.