Scholarlink

The automatic segmentation and measurement of intrapartum ultrasound images are crucial for surgical planning in obstetrics and gynecology. IUGC 2024 challenge aims to accurately classify standard planes in ultrasound videos (including those with important anatomical landmarks for fetal biometry measurement) and perform segmentation of the fetal head and pubic symphysis on the classified standard planes for precise fetal biometry measurement. Utilizing the IUGC dataset, we trained multiple deep learning models with specific loss functions tailored for accurate phase classification and ultrasound segmentation. These models not only facilitate the categorization of various phases during childbirth but also enable the accurate measurement of the progress angle. Through rigorous comparative experiments, we demonstrated the robust competitiveness and efficacy of our adopted models, highlighting their potential for enhancing clinical decision-making in obstetrics and gynecology. In this paper, the ResNet implemented in the classification task achieved an accuracy rate of 0.996, and the DeepLabV3 implemented in the segmentation task obtained an IoU of 0.917. The source code is available at https://github.com/tyb311/IUGC2024 .

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

Classification and Segmentation of Intrapartum Ultrasound Images with Deep Learning Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yubo Tan,
  • Shiye Wang,
  • Wen-Da Shen,
  • Wang-Wang Yu,
  • Yihao Li,
  • Philippe Zhang,
  • Weili Jiang,
  • Yong-Jie Li

摘要

The automatic segmentation and measurement of intrapartum ultrasound images are crucial for surgical planning in obstetrics and gynecology. IUGC 2024 challenge aims to accurately classify standard planes in ultrasound videos (including those with important anatomical landmarks for fetal biometry measurement) and perform segmentation of the fetal head and pubic symphysis on the classified standard planes for precise fetal biometry measurement. Utilizing the IUGC dataset, we trained multiple deep learning models with specific loss functions tailored for accurate phase classification and ultrasound segmentation. These models not only facilitate the categorization of various phases during childbirth but also enable the accurate measurement of the progress angle. Through rigorous comparative experiments, we demonstrated the robust competitiveness and efficacy of our adopted models, highlighting their potential for enhancing clinical decision-making in obstetrics and gynecology. In this paper, the ResNet implemented in the classification task achieved an accuracy rate of 0.996, and the DeepLabV3 implemented in the segmentation task obtained an IoU of 0.917. The source code is available at https://github.com/tyb311/IUGC2024 .