Scholarlink

<p>In recent times, considerable advances have been seen in the diagnosis of brain tumors due to unceasing improvements in computer-aided diagnostic techniques. The current study aims to systematically examine the evolution of various methodologies, from machine learning to deep transfer learning, and the most emerging self-attention-based vision transformer architectures. This literature review considers 161 articles from 2016 to 2025 for in-depth analysis, with an aim to present a widespread investigation of classification pipeline workflows, highlighting the shift from handcrafted feature-based models to automated deep feature extraction and hybrid learning frameworks. A noteworthy focus is given to self-attention-based models, with a capability for augmenting contextual feature representation and improving generalization in brain tumor diagnostics. To investigate effectiveness and workflow designs of the models, the authors framed significant research questions addressed through deep analysis of published peer-reviewed studies. Furthermore, the study also presents prevailing challenges, limitations, and opportunities within the field, with an aim to provide future research directions to improve robust, interpretable, and clinically viable computer-aided diagnosis models for brain tumor analysis. The presented study serves as an important resource for researchers with the aim of understanding cutting-edge developments and nurturing enhancements in automated brain tumor classification using deep neural networks.</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 Learning Based Brain Tumor Diagnosis with Pre-Trained and Self-Attention Based Models Using MRI Scans: A Systematic Literature Review

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Bindu Goyal,
  • Rahul Hans,
  • Sanjeev Kumar Sharma,
  • Harpreet Singh

摘要

In recent times, considerable advances have been seen in the diagnosis of brain tumors due to unceasing improvements in computer-aided diagnostic techniques. The current study aims to systematically examine the evolution of various methodologies, from machine learning to deep transfer learning, and the most emerging self-attention-based vision transformer architectures. This literature review considers 161 articles from 2016 to 2025 for in-depth analysis, with an aim to present a widespread investigation of classification pipeline workflows, highlighting the shift from handcrafted feature-based models to automated deep feature extraction and hybrid learning frameworks. A noteworthy focus is given to self-attention-based models, with a capability for augmenting contextual feature representation and improving generalization in brain tumor diagnostics. To investigate effectiveness and workflow designs of the models, the authors framed significant research questions addressed through deep analysis of published peer-reviewed studies. Furthermore, the study also presents prevailing challenges, limitations, and opportunities within the field, with an aim to provide future research directions to improve robust, interpretable, and clinically viable computer-aided diagnosis models for brain tumor analysis. The presented study serves as an important resource for researchers with the aim of understanding cutting-edge developments and nurturing enhancements in automated brain tumor classification using deep neural networks.