Scholarlink

We employed an ensemble technique for the BraTS Generalizability Across Tumors (BraTS-GoAT) task of the BraTS-Lighthouse 2025 Challenge, where it performed well on the unseen test validation dataset. To enhance models’ segmentation capabilities, we applied specific preprocessing techniques tailored to BraTS, focused on training different tumor regions, and made several minor adjustments to the pipeline. We also used the BraTS ranking criteria to identify the nnU-Net variant that met the challenge’s requirements most effectively. Our results for the whole tumor, tumor core, and enhanced tumor were Dice similarity coefficients of 0.84, 0.87, and 0.90 and HD95 values of 3.39, 3.12, and 2.64, respectively.

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

Ensemble-Based Generalization for Brain Tumor Segmentation Using NnU-Net Variants and Swin UNETR

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vaidehi Satushe,
  • Madhav Arora,
  • Vibha Vyas,
  • Shilpa Metkar

摘要

We employed an ensemble technique for the BraTS Generalizability Across Tumors (BraTS-GoAT) task of the BraTS-Lighthouse 2025 Challenge, where it performed well on the unseen test validation dataset. To enhance models’ segmentation capabilities, we applied specific preprocessing techniques tailored to BraTS, focused on training different tumor regions, and made several minor adjustments to the pipeline. We also used the BraTS ranking criteria to identify the nnU-Net variant that met the challenge’s requirements most effectively. Our results for the whole tumor, tumor core, and enhanced tumor were Dice similarity coefficients of 0.84, 0.87, and 0.90 and HD95 values of 3.39, 3.12, and 2.64, respectively.