Scholarlink

<p>Brain tumor segmentation (BTS) is a critical task for the accurate diagnosis and treatment of brain tumors. Manual detection poses significant challenges due to the complex anatomy and the variations in tumor types, sizes, shapes, and locations. Computer-aided diagnostic techniques have gained popularity for assisting healthcare providers in diagnosing diseases and improving the consistency and accuracy of their findings. This study introduces a novel hybrid PA-ResNet50 deep learning approach for explainable panoptic brain tumor segmentation (PBTS), integrating both instance and semantic segmentation to achieve detailed tumor boundary delineation while addressing uncertainties in brain imaging. Unlike conventional segmentation methods, our approach leverages panoptic segmentation with uncertainty modeling, enhancing both interpretability and robustness. The primary objective is to reduce the uncertainties in brain imaging, thereby increasing tumor identification accuracy and boosting the confidence of medical professionals. Evaluation results demonstrate that the proposed framework enhances both the interpretability of the results and the precision of brain tumor segmentation. Our model achieved an accuracy of 99.3 and 98.7%, Dice scores of 99.29 and 98.85% and computational times of 13 and 27&#xa0;s for the BraTS 2019 and BraTS 2021 datasets, respectively. This method not only improves segmentation precision but also enhances the interpretability and reliability of tumor diagnoses, providing a trustworthy, explainable AI-driven solution for clinical decision-making.</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

An explainable deep learning-based panoptic segmentation for brain tumor diagnosis

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Berlin Shaheema,
  • Naresh Babu Muppalaneni,
  • K. Suganya Devi

摘要

Brain tumor segmentation (BTS) is a critical task for the accurate diagnosis and treatment of brain tumors. Manual detection poses significant challenges due to the complex anatomy and the variations in tumor types, sizes, shapes, and locations. Computer-aided diagnostic techniques have gained popularity for assisting healthcare providers in diagnosing diseases and improving the consistency and accuracy of their findings. This study introduces a novel hybrid PA-ResNet50 deep learning approach for explainable panoptic brain tumor segmentation (PBTS), integrating both instance and semantic segmentation to achieve detailed tumor boundary delineation while addressing uncertainties in brain imaging. Unlike conventional segmentation methods, our approach leverages panoptic segmentation with uncertainty modeling, enhancing both interpretability and robustness. The primary objective is to reduce the uncertainties in brain imaging, thereby increasing tumor identification accuracy and boosting the confidence of medical professionals. Evaluation results demonstrate that the proposed framework enhances both the interpretability of the results and the precision of brain tumor segmentation. Our model achieved an accuracy of 99.3 and 98.7%, Dice scores of 99.29 and 98.85% and computational times of 13 and 27 s for the BraTS 2019 and BraTS 2021 datasets, respectively. This method not only improves segmentation precision but also enhances the interpretability and reliability of tumor diagnoses, providing a trustworthy, explainable AI-driven solution for clinical decision-making.