Scholarlink

<p>Accurate and early classification of gliomas based on their histopathological grade is critical for determining appropriate treatment strategies and improving patient outcomes. Researchers have examined and studied various deep learning techniques that can help with potential medical assistance. This model leverages multi-scale feature extraction to capture global and localized pathological characteristics, whereas contrastive learning enhances the ability of the model to learn discriminative representations without extensive labeled data. This study proposes a novel deep-learning framework for glioma grade classification using a ResNet-50 backbone combined with multi-scale feature fusion and self-supervised contrastive learning. The proposed method was tested on a large dataset of glioma pictures, yielding an accuracy (97.89%), precision (98.3%), recall (97.8%), specificity (94.9%), F1 score (96%), and Matthews Correlation Coefficient (MCC) (96.2%). Our findings show that merging contrastive learning with a multi-scale feature fusion technique greatly improves the model’s performance in classifying glioma grades. The high precision and recall rates indicated the strong potential of this approach for application in real-world diagnostic workflows.</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

Glioma Grading Unveiled: Optimizing Classification with Multi-Scale Fusion and Self-Supervised Contrastive Learning

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • J. Pearline Sheba Grace,
  • P. Ezhilarasi,
  • S. Rajesh Kannan

摘要

Accurate and early classification of gliomas based on their histopathological grade is critical for determining appropriate treatment strategies and improving patient outcomes. Researchers have examined and studied various deep learning techniques that can help with potential medical assistance. This model leverages multi-scale feature extraction to capture global and localized pathological characteristics, whereas contrastive learning enhances the ability of the model to learn discriminative representations without extensive labeled data. This study proposes a novel deep-learning framework for glioma grade classification using a ResNet-50 backbone combined with multi-scale feature fusion and self-supervised contrastive learning. The proposed method was tested on a large dataset of glioma pictures, yielding an accuracy (97.89%), precision (98.3%), recall (97.8%), specificity (94.9%), F1 score (96%), and Matthews Correlation Coefficient (MCC) (96.2%). Our findings show that merging contrastive learning with a multi-scale feature fusion technique greatly improves the model’s performance in classifying glioma grades. The high precision and recall rates indicated the strong potential of this approach for application in real-world diagnostic workflows.