Scholarlink

<p>The diagnosis of a brain tumour is a critical challenge in medical imaging, necessitating high-accuracy segmentation and strong classification to provide prompt treatment planning. Traditional deep learning models often encounter issues, including inaccurate border delineation and computational inefficiency, which impede their practical use in clinical operations. This paper offers an advanced end-to-end hybrid framework that incorporates transformer-based topologies, attention-guided feature refinement, and a unique metaheuristic optimisation technique to address these constraints. MRI pictures undergo preprocessing using methods such as N4 bias field correction, Z-score normalisation, and Contrast-Limited Adaptive Histogram Equalisation (CLAHE) to improve image quality. A UNETR (U-Net Transformer) architecture is used for segmentation to gather both local and global contextual information, facilitating accurate tumour border identification. Multi-head self-attention and channel-spatial attention modules enhance the deep features obtained by the UNETR encoder. The Zebra Optimisation Algorithm (ZOA) is used to enhance feature selection and hyperparameter optimisation, guaranteeing the extraction of the most discriminative features and the fine-tuning of model parameters for optimal performance. A Quantum-Classical Variational Quantum Classifier (VQC) is used for classification, using quantum-enhanced feature mapping and variational circuits to attain better classification accuracy while ensuring computational viability. The suggested system attains a Dice Similarity Coefficient (DSC) of 97.1% and an Intersection over Union (IoU) of 94.8% for segmentation, considerably surpassing baseline UNet and CNN-based methodologies. The VQC achieves a classification accuracy of 98.7%, sensitivity of 98.1%, specificity of 99.0%, and F1-score of 98.3%. Furthermore, it exhibits enhanced computing efficiency, confirming its feasibility for actual clinical use. This ZOA-optimized transformer and quantum-classical hybrid approach provides a reliable and effective solution for precise brain tumour identification and classification in MRI data.</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

Enhanced Brain Tumor Segmentation and Classification Based on Quantum-Classical Attention-Guided Hybrid Framework with Zebra Optimization Algorithm

  • 期刊论文

  • 发表日期：

    2026年2月6日

查看全文
  • Amina Salhi,
  • Mohamed M. Hassan

摘要

The diagnosis of a brain tumour is a critical challenge in medical imaging, necessitating high-accuracy segmentation and strong classification to provide prompt treatment planning. Traditional deep learning models often encounter issues, including inaccurate border delineation and computational inefficiency, which impede their practical use in clinical operations. This paper offers an advanced end-to-end hybrid framework that incorporates transformer-based topologies, attention-guided feature refinement, and a unique metaheuristic optimisation technique to address these constraints. MRI pictures undergo preprocessing using methods such as N4 bias field correction, Z-score normalisation, and Contrast-Limited Adaptive Histogram Equalisation (CLAHE) to improve image quality. A UNETR (U-Net Transformer) architecture is used for segmentation to gather both local and global contextual information, facilitating accurate tumour border identification. Multi-head self-attention and channel-spatial attention modules enhance the deep features obtained by the UNETR encoder. The Zebra Optimisation Algorithm (ZOA) is used to enhance feature selection and hyperparameter optimisation, guaranteeing the extraction of the most discriminative features and the fine-tuning of model parameters for optimal performance. A Quantum-Classical Variational Quantum Classifier (VQC) is used for classification, using quantum-enhanced feature mapping and variational circuits to attain better classification accuracy while ensuring computational viability. The suggested system attains a Dice Similarity Coefficient (DSC) of 97.1% and an Intersection over Union (IoU) of 94.8% for segmentation, considerably surpassing baseline UNet and CNN-based methodologies. The VQC achieves a classification accuracy of 98.7%, sensitivity of 98.1%, specificity of 99.0%, and F1-score of 98.3%. Furthermore, it exhibits enhanced computing efficiency, confirming its feasibility for actual clinical use. This ZOA-optimized transformer and quantum-classical hybrid approach provides a reliable and effective solution for precise brain tumour identification and classification in MRI data.