Scholarlink

Background <p>Classifying brain tumors using magnetic resonance imaging (MRI) is still a big problem in neuroradiology. For clinical use, it needs to be both very accurate and easy to understand. Conventional single-model methodologies frequently demonstrate inadequacies in representing the varied morphological attributes of distinct tumor types.</p> Objective <p>This research creates and tests a new ensemble learning framework that combines several convolutional neural networks (CNNs) with explainable artificial intelligence (XAI) methods to automatically classify brain tumors from MRI scans.</p> Methods <p>We used an ensemble architecture that combined the VGG16, ResNet50, and InceptionV3 networks. We trained it on a large dataset of 7,523 MRI images from four groups: glioma, meningioma, non-tumor, and pituitary tumor. The framework combines predictions using weighted averaging and shows how the predictions work using Gradient-weighted Class Activation Mapping (Grad-CAM). We used strict cross-validation and the metrics accuracy, precision, recall, F1-score, and area under the curve (AUC) to check how well it worked.</p> Results <p>The ensemble framework achieved superior performance with 94% accuracy, 93% precision, 93% recall, and 93% F1-score, significantly outperforming individual models (VGG16: 90%, ResNet50: 92%, InceptionV3: 91%). With an AUC score of 0.99, it was very good at telling the difference between things. Grad-CAM visualizations highlighted clinically significant anatomical regions, aligning with recognized radiological markers.</p> Conclusions <p>The proposed ensemble framework ameliorates significant shortcomings in medical AI by integrating enhanced classification accuracy with comprehensible decision-making. Combining different CNN architectures with XAI techniques is a good way to help doctors make decisions about neuroradiology.</p>

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 ensemble learning framework for brain tumour classification with explainable AI for medical diagnosis

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Parthasarathy Sundararajan,
  • Sivakumar Dharmaraj,
  • Prasad Gadiraju

摘要

Background

Classifying brain tumors using magnetic resonance imaging (MRI) is still a big problem in neuroradiology. For clinical use, it needs to be both very accurate and easy to understand. Conventional single-model methodologies frequently demonstrate inadequacies in representing the varied morphological attributes of distinct tumor types.

Objective

This research creates and tests a new ensemble learning framework that combines several convolutional neural networks (CNNs) with explainable artificial intelligence (XAI) methods to automatically classify brain tumors from MRI scans.

Methods

We used an ensemble architecture that combined the VGG16, ResNet50, and InceptionV3 networks. We trained it on a large dataset of 7,523 MRI images from four groups: glioma, meningioma, non-tumor, and pituitary tumor. The framework combines predictions using weighted averaging and shows how the predictions work using Gradient-weighted Class Activation Mapping (Grad-CAM). We used strict cross-validation and the metrics accuracy, precision, recall, F1-score, and area under the curve (AUC) to check how well it worked.

Results

The ensemble framework achieved superior performance with 94% accuracy, 93% precision, 93% recall, and 93% F1-score, significantly outperforming individual models (VGG16: 90%, ResNet50: 92%, InceptionV3: 91%). With an AUC score of 0.99, it was very good at telling the difference between things. Grad-CAM visualizations highlighted clinically significant anatomical regions, aligning with recognized radiological markers.

Conclusions

The proposed ensemble framework ameliorates significant shortcomings in medical AI by integrating enhanced classification accuracy with comprehensible decision-making. Combining different CNN architectures with XAI techniques is a good way to help doctors make decisions about neuroradiology.