Scholarlink

<p>The heterogeneous and infiltrative nature of brain tumors poses significant diagnostic and therapeutic challenges, and the Brain Tumor Segmentation (BraTS) dataset has thus become a pivotal benchmark for developing Deep Learning (DL) applications in computational neuro-oncology. This systematic review synthesizes research on DL methodologies leveraging the BraTS dataset to investigate model architectures, performance, limitations, and clinical translation. In this regard, a comprehensive literature search of PubMed and the MICCAI proceedings through November 2025 was conducted, yielding 145 eligible studies. Findings indicate the dominance of convolutional-based neural networks for segmentation tasks, with widespread use of Transfer Learning (TL) and data augmentation to strengthen performance. While DL approaches lead to advantages in computational efficiency and robustness, key limitations include a performance gap on external data, data imperfections, and a lack of prospective clinical validation. The review concludes that, despite the indispensable role of BraTS in advancing DL innovations, future work should prioritize the development of computationally efficient, interpretable, and generalizable models that are validated within clinical workflows to bridge the gap between research and practice.</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 for Brain Tumor Analysis: A Systematic Review of the BraTS Dataset

  • 期刊论文

  • 发表日期：

    2026年1月31日

查看全文
  • Oktay Fasihi Shirehjini,
  • Farshid Babapour Mofrad

摘要

The heterogeneous and infiltrative nature of brain tumors poses significant diagnostic and therapeutic challenges, and the Brain Tumor Segmentation (BraTS) dataset has thus become a pivotal benchmark for developing Deep Learning (DL) applications in computational neuro-oncology. This systematic review synthesizes research on DL methodologies leveraging the BraTS dataset to investigate model architectures, performance, limitations, and clinical translation. In this regard, a comprehensive literature search of PubMed and the MICCAI proceedings through November 2025 was conducted, yielding 145 eligible studies. Findings indicate the dominance of convolutional-based neural networks for segmentation tasks, with widespread use of Transfer Learning (TL) and data augmentation to strengthen performance. While DL approaches lead to advantages in computational efficiency and robustness, key limitations include a performance gap on external data, data imperfections, and a lack of prospective clinical validation. The review concludes that, despite the indispensable role of BraTS in advancing DL innovations, future work should prioritize the development of computationally efficient, interpretable, and generalizable models that are validated within clinical workflows to bridge the gap between research and practice.