Scholarlink

Brain tumors represent a significant neurological burden, characterized by their heterogeneity and diagnostic complexity. Traditional diagnosis, which relies on expert-driven MRI interpretation, is often time-consuming and prone to subjectivity. Motivated by this, we propose, PRO_SPEEDY, a hybrid deep learning technique designed to address these challenges by combining the strengths of Convolution Neural Networks (CNNs) and Vision Transformers (ViTs). The proposed architecture integrates a ResNet backbone for localized feature extraction and a MobileViT component for capturing broader spatial dependencies. This fusion is predicated on the complementary capabilities of CNNs, which efficiently recognize fine-grained patterns, and ViTs, which excel in modeling long-range contextual relationships. The model was trained on the BRISC 2025 dataset, with extensive pre-processing and data augmentation employed to enhance robustness. The framework achieved a classification accuracy of 99.1%, surpassing several state-of-the-art baselines. These results indicate that a balanced CNN-Transformer hybrid can serve as an effective diagnostic tool, particularly in environments where speed, precision, and computational efficiency are critical. The proposed model shows strong potential for integration into real-time clinical decision-support systems.

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

PRO_SPEEDY: Achieving Speed with Precision Using a Lightweight CNN–Mobile Vision Transformer Model for Brain Tumor Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Parth Mishra,
  • Anushka Nehra,
  • Sandeep Verma

摘要

Brain tumors represent a significant neurological burden, characterized by their heterogeneity and diagnostic complexity. Traditional diagnosis, which relies on expert-driven MRI interpretation, is often time-consuming and prone to subjectivity. Motivated by this, we propose, PRO_SPEEDY, a hybrid deep learning technique designed to address these challenges by combining the strengths of Convolution Neural Networks (CNNs) and Vision Transformers (ViTs). The proposed architecture integrates a ResNet backbone for localized feature extraction and a MobileViT component for capturing broader spatial dependencies. This fusion is predicated on the complementary capabilities of CNNs, which efficiently recognize fine-grained patterns, and ViTs, which excel in modeling long-range contextual relationships. The model was trained on the BRISC 2025 dataset, with extensive pre-processing and data augmentation employed to enhance robustness. The framework achieved a classification accuracy of 99.1%, surpassing several state-of-the-art baselines. These results indicate that a balanced CNN-Transformer hybrid can serve as an effective diagnostic tool, particularly in environments where speed, precision, and computational efficiency are critical. The proposed model shows strong potential for integration into real-time clinical decision-support systems.