Scholarlink

The presence of a brain tumour is a severe life-threatening condition that contributes significantly to global mortality rates. However, early identification of brain tumours may substantially reduce the mortality associated with this disease. Hence, the proposed research introduces an effective multi-level learning methodology designed for the segmentation and classification of brain tumours in Magnetic Resonance Images (MRI). The proposed pipeline consists of three-level architectures, viz. (I) Residual-Attention-UNet for tumour segmentation, (II) Binary classification module formed by concatenating the features extracted from the 4th(last) encoder block down the line from the encoder, bridge and decoder blocks for the efficient classification of tumours. (III) Classification into subtypes of low-grade gliomas using Kerenel attention transformer and histopathologic data. Experimental results demonstrate that the method improves overall accuracy compared to existing state-of-the-art procedures. The Level-1 segmentation results are verified through the Dice coefficients(Dice), Sensitivity, and Specificity. Level-2 classification results are verified using overall Accuracy, and Level-3 results are verified by F1-score. It is observed that the proposed Residual-Attention-UNet achieves around Dice score of 89%, 95% Sensitivity, and a specificity of 98%. The binary classification results in terms of Accuracy(ACC) are 96.57%. Level-3 sub-classification results achieve more than 94% F1-score.

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

A Novel Segmentation and Classification Approach for Brain Tumours by Multilevel Deep Learning Architecture

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Syed Sajid Hussain,
  • Ashiq Nazir Bhat,
  • Syed Mujtaba Hussaine

摘要

The presence of a brain tumour is a severe life-threatening condition that contributes significantly to global mortality rates. However, early identification of brain tumours may substantially reduce the mortality associated with this disease. Hence, the proposed research introduces an effective multi-level learning methodology designed for the segmentation and classification of brain tumours in Magnetic Resonance Images (MRI). The proposed pipeline consists of three-level architectures, viz. (I) Residual-Attention-UNet for tumour segmentation, (II) Binary classification module formed by concatenating the features extracted from the 4th(last) encoder block down the line from the encoder, bridge and decoder blocks for the efficient classification of tumours. (III) Classification into subtypes of low-grade gliomas using Kerenel attention transformer and histopathologic data. Experimental results demonstrate that the method improves overall accuracy compared to existing state-of-the-art procedures. The Level-1 segmentation results are verified through the Dice coefficients(Dice), Sensitivity, and Specificity. Level-2 classification results are verified using overall Accuracy, and Level-3 results are verified by F1-score. It is observed that the proposed Residual-Attention-UNet achieves around Dice score of 89%, 95% Sensitivity, and a specificity of 98%. The binary classification results in terms of Accuracy(ACC) are 96.57%. Level-3 sub-classification results achieve more than 94% F1-score.