Scholarlink

<p>Free flap reconstruction is essential for treating intraoral defects; however, failure can lead to complex and prolonged complications. While various monitoring methods have been employed to prevent such situations, they are qualitative and sometimes unfamiliar to novices. The purpose of this study was to develop a user-friendly model using artificial intelligence that quantitatively represents flap status. We analyzed 1877 images from 131 patients who underwent free flap reconstruction for intraoral defects between June 2021 and March 2024. Since patients with vascular damage were very few in number, class weighting and focal loss techniques were used to address this imbalance. The proposed model achieved high overall accuracy and F1 scores of 0.9867 and 0.9863, respectively. This study introduces the first deep learning model for intraoral flaps and demonstrates the possibility of quantitative measurement of flap changes. This tool can assist surgeons in making timely decisions regarding salvage procedures and facilitate easier monitoring for resident care-givers.</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

Machine learning approaches overcome imbalanced clinical data for intraoral free flap monitoring

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月7日

查看全文
  • Hyounmin Kim,
  • Dongwook Kim,
  • Juho Bai

摘要

Free flap reconstruction is essential for treating intraoral defects; however, failure can lead to complex and prolonged complications. While various monitoring methods have been employed to prevent such situations, they are qualitative and sometimes unfamiliar to novices. The purpose of this study was to develop a user-friendly model using artificial intelligence that quantitatively represents flap status. We analyzed 1877 images from 131 patients who underwent free flap reconstruction for intraoral defects between June 2021 and March 2024. Since patients with vascular damage were very few in number, class weighting and focal loss techniques were used to address this imbalance. The proposed model achieved high overall accuracy and F1 scores of 0.9867 and 0.9863, respectively. This study introduces the first deep learning model for intraoral flaps and demonstrates the possibility of quantitative measurement of flap changes. This tool can assist surgeons in making timely decisions regarding salvage procedures and facilitate easier monitoring for resident care-givers.