Scholarlink

Gliomas are among the most aggressive and difficult brain tumors to diagnose, particularly in Africa. Effective and accurate segmentation of tumor regions using MRI images enables automated diagnosis and improves the therapeutic management of patients. This study proposes an optimized 3D U-Net model, named SenTumorNet, tailored for glioma segmentation and addressing challenges such as computational efficiency, generalization, and adaptability to resource-limited environments like those in Africa. The proposed SenTumorNet model integrates a simplified encoder-decoder block and an optimized skip connection, allowing for accurate segmentation while reducing computation time. The SenTumorNet model was validated on the BraTS-Africa 2024 multimodal MRI dataset after training. The model achieved an overall Dice score and IoU of 79.92% and 68.98%, respectively. These results highlight the potential of the optimized SenTumorNet as a robust, accurate, and efficient model for glioma segmentation. Future work will focus on comparing our model, SenTumorNet, with state-of-the-art (SOTA) models, clinically validating the model in the African context, and extending it toward a multimodal approach by integrating more diverse clinical data. The implementation of SenTumorNet is publicly available at: https://github.com/SPARK-Academy-2025/SPARK-2025/tree/main/SPARK2025_BraTs_MODELS/Team_Jolof_HealthIA .

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

SenTumorNet: A Lightweight 3D U-Net Model for Brain Tumor Segmentation in Sub-Saharan African MRI Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Papa Seydou Wane,
  • Abdourahamane Balde,
  • Guy Mbatchou,
  • Dieu-Donné Okalas Ossami,
  • Mariama Dione,
  • Dieumbe Khoule,
  • Mor Diop,
  • Ndeye Maty Bousso,
  • Adama Traore,
  • Aondona Iorumbur,
  • Raymond Confidence,
  • Udunna Anazodo

摘要

Gliomas are among the most aggressive and difficult brain tumors to diagnose, particularly in Africa. Effective and accurate segmentation of tumor regions using MRI images enables automated diagnosis and improves the therapeutic management of patients. This study proposes an optimized 3D U-Net model, named SenTumorNet, tailored for glioma segmentation and addressing challenges such as computational efficiency, generalization, and adaptability to resource-limited environments like those in Africa. The proposed SenTumorNet model integrates a simplified encoder-decoder block and an optimized skip connection, allowing for accurate segmentation while reducing computation time. The SenTumorNet model was validated on the BraTS-Africa 2024 multimodal MRI dataset after training. The model achieved an overall Dice score and IoU of 79.92% and 68.98%, respectively. These results highlight the potential of the optimized SenTumorNet as a robust, accurate, and efficient model for glioma segmentation. Future work will focus on comparing our model, SenTumorNet, with state-of-the-art (SOTA) models, clinically validating the model in the African context, and extending it toward a multimodal approach by integrating more diverse clinical data. The implementation of SenTumorNet is publicly available at: https://github.com/SPARK-Academy-2025/SPARK-2025/tree/main/SPARK2025_BraTs_MODELS/Team_Jolof_HealthIA .