Scholarlink

<p>Machine learning has shown significant potential in the medical field as computer technology has advanced. Deep learning models are extensively utilized in medical image analysis and have demonstrated superior performance. In radiation therapy for cervical cancer, radiation oncologists are required to delineate the contours of tumors. To alleviate the burden on radiation oncologists, automated tumor contour segmentation would be beneficial. In this study, T2 weighted images (T2WI) from 151 cervical cancer (CC) patients were collected, and the dataset has been made publicly accessible. We trained an automatic cervical tumor contour segmentation model based on nnU-Net to explore their potential clinical applications. Dice Similarity Coefficient (DSC), Intersection over Union (IoU), recall, accuracy, and precision metrics were employed to evaluate the differences between the tumor contours delineated by radiation oncologists and those generated by the models. The nnU-Net-3D achieved a DSC of 0.8181, an IoU of 0.7096, a Recall of 82.52%. This study indicates that the nnU-Net model can accurately and rapidly outline cervical cancer tumors with high reproducibility, thereby reducing the workload of radiation oncologists. The MRI dataset and code are publicly available at <a href="https://github.com/peekaboo990/CCMD-Cervical-Cancer-MRI-Dataset">https://github.com/peekaboo990/CCMD-Cervical-Cancer-MRI-Dataset</a>.</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 open MRI dataset for cervical cancer with tumor segmentation using nnU-net

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月5日

查看全文
  • Genyun Liu,
  • Yi Tang,
  • Lin Guo,
  • Liwei Zhang,
  • Qisong Zheng,
  • Hui Hui,
  • Li Cai,
  • Yun Zhou

摘要

Machine learning has shown significant potential in the medical field as computer technology has advanced. Deep learning models are extensively utilized in medical image analysis and have demonstrated superior performance. In radiation therapy for cervical cancer, radiation oncologists are required to delineate the contours of tumors. To alleviate the burden on radiation oncologists, automated tumor contour segmentation would be beneficial. In this study, T2 weighted images (T2WI) from 151 cervical cancer (CC) patients were collected, and the dataset has been made publicly accessible. We trained an automatic cervical tumor contour segmentation model based on nnU-Net to explore their potential clinical applications. Dice Similarity Coefficient (DSC), Intersection over Union (IoU), recall, accuracy, and precision metrics were employed to evaluate the differences between the tumor contours delineated by radiation oncologists and those generated by the models. The nnU-Net-3D achieved a DSC of 0.8181, an IoU of 0.7096, a Recall of 82.52%. This study indicates that the nnU-Net model can accurately and rapidly outline cervical cancer tumors with high reproducibility, thereby reducing the workload of radiation oncologists. The MRI dataset and code are publicly available at https://github.com/peekaboo990/CCMD-Cervical-Cancer-MRI-Dataset.