Scholarlink

This paper presents a framework for automated brain tumor diagnosis using deep-learning approach, providing an end-to-end solution for detection, classification, and segmentation of tumors from MRI scans. Brain tumors can be accurately detected and classified into glioma, meningioma, and pituitary tumors using custom Convolutional Neural Networks (CNNs) models. Additionally, state-of-the-art deep learning models like VGG16, ResNet50, and MobileNetV2 are compared with the custom CNN to identify which architecture performs best for classification tasks. Furthermore, Grad-CAM, Grad-CAM++, and Score-CAM techniques are employed to visualize and interpret the predictions of these CNN models, providing deeper insights into their decision-making processes. In order to ensure accurate segmentation of the tumor regions, a U-Net model which captures small features of the tumor boundaries while maintaining spatial information is utilized. Furthermore, a Streamlit-based application incorporates these tasks, providing a user-friendly platform for practical execution. This comprehensive solution provides significant insights and a streamlined methodology for brain tumor analysis, which can help to improve clinical decision-making.

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

An Interpretable Deep Learning Framework For Brain Tumor Diagnosis Using UNET and CAM Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Apeksha Shah,
  • Vidhi Kotak,
  • Vaidik Pandya,
  • Rajeev Kumar Gupta

摘要

This paper presents a framework for automated brain tumor diagnosis using deep-learning approach, providing an end-to-end solution for detection, classification, and segmentation of tumors from MRI scans. Brain tumors can be accurately detected and classified into glioma, meningioma, and pituitary tumors using custom Convolutional Neural Networks (CNNs) models. Additionally, state-of-the-art deep learning models like VGG16, ResNet50, and MobileNetV2 are compared with the custom CNN to identify which architecture performs best for classification tasks. Furthermore, Grad-CAM, Grad-CAM++, and Score-CAM techniques are employed to visualize and interpret the predictions of these CNN models, providing deeper insights into their decision-making processes. In order to ensure accurate segmentation of the tumor regions, a U-Net model which captures small features of the tumor boundaries while maintaining spatial information is utilized. Furthermore, a Streamlit-based application incorporates these tasks, providing a user-friendly platform for practical execution. This comprehensive solution provides significant insights and a streamlined methodology for brain tumor analysis, which can help to improve clinical decision-making.