Scholarlink

The variety of brain tumors and their potential for rapid growth make them a serious health concern. For the best course of therapy and patient outcomes, an early and precise diagnosis is vital. Using deep learning models, this study examines how well ensemble learning methods classify brain tumors. 7023 MRI images of four different tumor types—glioma, meningioma, pituitary, and no tumor—were used as the dataset. The dataset was used to fine-tune eight pre-trained convolutional neural networks (CNNs), and their performance was evaluated. The predictions of the best three performing individual models (DenseNet201, ResNet50, and NASNetMobile) were then combined using three ensemble techniques: Hard Voting, Soft Voting, and Stacking. The outcomes showed that the ensemble approaches outperformed the individual models by a considerable margin. The Stacking Ensemble with Logistic Regression meta-classifier showed remarkable predicting abilities by achieving top accuracy of 99.55% surpassing state-of-the-art methods. This study highlights the potential of ensemble learning techniques in enhancing the accuracy and robustness of brain tumor classification model, paving the way for improved diagnostic tools and improved patient care.

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

Classification of Brain Tumor Using Ensemble of Deep Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • N. Yibenthung Tungoe,
  • Amit Doegar

摘要

The variety of brain tumors and their potential for rapid growth make them a serious health concern. For the best course of therapy and patient outcomes, an early and precise diagnosis is vital. Using deep learning models, this study examines how well ensemble learning methods classify brain tumors. 7023 MRI images of four different tumor types—glioma, meningioma, pituitary, and no tumor—were used as the dataset. The dataset was used to fine-tune eight pre-trained convolutional neural networks (CNNs), and their performance was evaluated. The predictions of the best three performing individual models (DenseNet201, ResNet50, and NASNetMobile) were then combined using three ensemble techniques: Hard Voting, Soft Voting, and Stacking. The outcomes showed that the ensemble approaches outperformed the individual models by a considerable margin. The Stacking Ensemble with Logistic Regression meta-classifier showed remarkable predicting abilities by achieving top accuracy of 99.55% surpassing state-of-the-art methods. This study highlights the potential of ensemble learning techniques in enhancing the accuracy and robustness of brain tumor classification model, paving the way for improved diagnostic tools and improved patient care.