Scholarlink

<p>Magnetic resonance imaging (MRI) plays a vital role in diagnosing brain tumors, which is an important task in neuro-oncology. However, this process can be difficult due to the complex and varying appearance of different types of tumors. New deep learning algorithms have demonstrated their potential to raise diagnostic accuracy. In this work, we suggest a new tripartite deep learning system for MRI-based brain tumor classification. We combine one hand-crafted convolutional neural network (CNN) and support vector machine (SVM) classifier with pretrained deep convolutional features, i.e., VGG-16 and AlexNet. These CNNs are adopted as stationary feature extractors, and the feature vectors are calculated and trained on the SVMs in tumor classification. The 3264 MRIs of 4 classes, glioma, meningioma, pituitary tumor, and non-tumor, are used to analyze the proposed models TDLM1 (VGG-16 + SVM), TDLM2 (AlexNet SVM), and TDLM3 (self-made CNN). Having the best accuracy and recall of 91.6 and 88.1, TDLM2 was the best and meningiomas in gliomas, 86.1 and 86.1, whereas meningiomas in TDLM3 had 95.0 precision and recall, which was the best compared to others during the classifications of pituitary tumors. TDLM1 has the best precision of 95.9% in the non-tumor detection. In general, the tripartite structure has shown a stable and good performance, though TDLM2 has proved to be the best model.</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

Deep Tripartite Architecture for Precise MRI-Based Brain Tumor Classification Using CNN and SVM Fusion

  • 期刊论文

  • 发表日期：

    2025年10月15日

查看全文
  • Yogesh Kumar Rathore,
  • Saroj Kumar Pandey,
  • Rishav Dubey,
  • Anurag Sinha,
  • Pratyush Maheshwari,
  • Nitish Kumar,
  • Pankaj Kumar Mishra

摘要

Magnetic resonance imaging (MRI) plays a vital role in diagnosing brain tumors, which is an important task in neuro-oncology. However, this process can be difficult due to the complex and varying appearance of different types of tumors. New deep learning algorithms have demonstrated their potential to raise diagnostic accuracy. In this work, we suggest a new tripartite deep learning system for MRI-based brain tumor classification. We combine one hand-crafted convolutional neural network (CNN) and support vector machine (SVM) classifier with pretrained deep convolutional features, i.e., VGG-16 and AlexNet. These CNNs are adopted as stationary feature extractors, and the feature vectors are calculated and trained on the SVMs in tumor classification. The 3264 MRIs of 4 classes, glioma, meningioma, pituitary tumor, and non-tumor, are used to analyze the proposed models TDLM1 (VGG-16 + SVM), TDLM2 (AlexNet SVM), and TDLM3 (self-made CNN). Having the best accuracy and recall of 91.6 and 88.1, TDLM2 was the best and meningiomas in gliomas, 86.1 and 86.1, whereas meningiomas in TDLM3 had 95.0 precision and recall, which was the best compared to others during the classifications of pituitary tumors. TDLM1 has the best precision of 95.9% in the non-tumor detection. In general, the tripartite structure has shown a stable and good performance, though TDLM2 has proved to be the best model.