Scholarlink

Classifying brain tumors accurately is important for making precise diagnoses and planning effective treatments. Using the Kaggle brain tumor MRI dataset, this study focuses on how well eight deep learning (DL) models—VGG16, Xception, ResNet152V2, Inception-ResNetV2, AlexNet, VGG19, MobileNet, and DenseNet121, do at classifying different types of brain tumors such as non-tumor, meningioma tumor, glioma tumor, and pituitary tumor. The study utilizes the advantages of some preprocessing techniques including non-local mean (NLM) filtering, histogram equalization, masking, and segmentation, to enhance image quality and prepare the data for analysis. The goal of this study is to determine which models are most effective for this task by evaluating them on metrics such as accuracy, misclassification error, log loss, and Brier score. The results show that VGG16 does the best, with the highest accuracy and lowest mistake rates. VGG19 and AlexNet also do well. Meanwhile, ResNet152V2, InceptionResNetV2, MobileNet, and DenseNet121 show mixed results, highlighting their varied effectiveness depending on the diagnostic context. These results can help to choose the best DL models to improve brain tumor classification to support better clinical decisions and 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

Exploring Pre-trained Deep Learning Models for Brain Tumor Classification: A Comparative Performance Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • T. Akhila,
  • M. Anbazhagan

摘要

Classifying brain tumors accurately is important for making precise diagnoses and planning effective treatments. Using the Kaggle brain tumor MRI dataset, this study focuses on how well eight deep learning (DL) models—VGG16, Xception, ResNet152V2, Inception-ResNetV2, AlexNet, VGG19, MobileNet, and DenseNet121, do at classifying different types of brain tumors such as non-tumor, meningioma tumor, glioma tumor, and pituitary tumor. The study utilizes the advantages of some preprocessing techniques including non-local mean (NLM) filtering, histogram equalization, masking, and segmentation, to enhance image quality and prepare the data for analysis. The goal of this study is to determine which models are most effective for this task by evaluating them on metrics such as accuracy, misclassification error, log loss, and Brier score. The results show that VGG16 does the best, with the highest accuracy and lowest mistake rates. VGG19 and AlexNet also do well. Meanwhile, ResNet152V2, InceptionResNetV2, MobileNet, and DenseNet121 show mixed results, highlighting their varied effectiveness depending on the diagnostic context. These results can help to choose the best DL models to improve brain tumor classification to support better clinical decisions and patient care.