Scholarlink

<p>Brain tumors pose a major global health challenge, necessitating early and precise diagnosis for successful treatment. This research tackles the problem of identifying various brain tumor types using MRI images. Existing approaches, which depend on traditional and basic deep learning techniques, struggle with issues such as lower accuracy, poor generalization, and difficulties in managing MRI image variations. These methods also fail to provide real-time solutions and are impractical for transfer learning. Our proposed model leverages advanced deep learning with the DenseNet121 architecture, enhanced by adaptive learning rates and callbacks for improved generalization and efficiency. Adaptive learning rates dynamically adjust the training pace in response to changes in tumor patterns, overcoming the drawbacks of fixed learning rates. Callbacks, including model checkpointing and early stopping, optimize training by saving the best configurations and preventing overfitting, addressing problems like lower accuracy and poor generalization. The model achieves a training accuracy of 98.81% and a test accuracy of 98.09%, outperforming current methods. It demonstrates balanced precision, recall, and F1-score across tumor types, validating its effectiveness in clinical neuro-oncology applications.</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

Brain tumor diagnosis using modified DenseNet121 architecture with adaptive learning rate and callback mechanism

  • 期刊论文

  • 发表日期：

    2025年3月19日

查看全文
  • Chandrasekar Venkatachalam,
  • Priyanka Shah,
  • P. Renukadevi,
  • Sincy John,
  • Shanmugavalli Venkatachalam

摘要

Brain tumors pose a major global health challenge, necessitating early and precise diagnosis for successful treatment. This research tackles the problem of identifying various brain tumor types using MRI images. Existing approaches, which depend on traditional and basic deep learning techniques, struggle with issues such as lower accuracy, poor generalization, and difficulties in managing MRI image variations. These methods also fail to provide real-time solutions and are impractical for transfer learning. Our proposed model leverages advanced deep learning with the DenseNet121 architecture, enhanced by adaptive learning rates and callbacks for improved generalization and efficiency. Adaptive learning rates dynamically adjust the training pace in response to changes in tumor patterns, overcoming the drawbacks of fixed learning rates. Callbacks, including model checkpointing and early stopping, optimize training by saving the best configurations and preventing overfitting, addressing problems like lower accuracy and poor generalization. The model achieves a training accuracy of 98.81% and a test accuracy of 98.09%, outperforming current methods. It demonstrates balanced precision, recall, and F1-score across tumor types, validating its effectiveness in clinical neuro-oncology applications.