Scholarlink

Brain metastases are the most common intracranial tumors. Volumetric assessment of the lesions is crucial for stereotactic radiosurgery planning. However, manual annotation of the lesions is time-consuming and prone to error. Automated, reliable solutions have the potential to overcome those barriers. A DiNTS-based 3D segmentation model of brain metastases MRI cases, trained on the Brain Tumor Segmentation (BraTS) – Metastases 2025 Lighthouse Challenge dataset was developed and presented in this study. MONAI and DiNTS frameworks with combined Dice and Focal loss were utilized to develop the model that achieved mean Dice coefficients of 0.692 (WT), 0.718 (TC), and 0.687 (ET) on a validation set of 130 cases. These scores indicate the model’s performance across tumor subregions.

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

Taking Advantage of MONAI and DiNTS Frameworks to Develop a State-of-the-Art Algorithm for Automatic Segmentation of Brain Metastases

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fabian Umeh,
  • Nikolay Y. Yordanov,
  • Nazanin Maleki,
  • Raisa Amiruddin,
  • Ahmed Moawad,
  • Monika Pytlarz,
  • Crystal Chukwurah,
  • Mariam Aboian

摘要

Brain metastases are the most common intracranial tumors. Volumetric assessment of the lesions is crucial for stereotactic radiosurgery planning. However, manual annotation of the lesions is time-consuming and prone to error. Automated, reliable solutions have the potential to overcome those barriers. A DiNTS-based 3D segmentation model of brain metastases MRI cases, trained on the Brain Tumor Segmentation (BraTS) – Metastases 2025 Lighthouse Challenge dataset was developed and presented in this study. MONAI and DiNTS frameworks with combined Dice and Focal loss were utilized to develop the model that achieved mean Dice coefficients of 0.692 (WT), 0.718 (TC), and 0.687 (ET) on a validation set of 130 cases. These scores indicate the model’s performance across tumor subregions.