Scholarlink

<p>Surgical site infection (SSI) is one of the most common and costly healthcare-associated infections, and surgical wound care remains a significant challenge for preventing SSIs and improving patient outcomes. Although deep learning has been explored for preliminary surgical wound screening, progress is limited by data privacy concerns and the high cost of expert annotation. Consequently, no publicly available dataset or benchmark currently covers diverse surgical wound types, hindering the development of open-source surgical wound screening tools. To address this gap: (1) we present SurgWound, the first open-source dataset featuring diverse surgical wound types. It contains 686 surgical wound images annotated by three professional surgeons with eight fine-grained clinical attributes. (2) Based on SurgWound, we introduce the first benchmark for surgical wound diagnosis, which includes visual question answering (VQA) and report generation tasks for comprehensive evaluation. (3) Furthermore, we propose a three-stage learning framework, WoundQwen, for surgical wound diagnosis. The first stage predicts detailed wound characteristics using multiple MLLMs. The second stage uses these predictions as additional knowledge to assess infection risk and clinical urgency. The third stage integrates the diagnostic results to produce a comprehensive report. Our dataset and code are publicly available.</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

SurgWound-Bench: a benchmark for surgical wound diagnosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月23日

查看全文
  • Jiahao Xu,
  • Changchang Yin,
  • Odysseas P. Chatzipanagiotou,
  • Diamantis I. Tsilimigras,
  • Kevin Clear,
  • Bingsheng Yao,
  • Weidan Cao,
  • Dakuo Wang,
  • Timothy M. Pawlik,
  • Ping Zhang

摘要

Surgical site infection (SSI) is one of the most common and costly healthcare-associated infections, and surgical wound care remains a significant challenge for preventing SSIs and improving patient outcomes. Although deep learning has been explored for preliminary surgical wound screening, progress is limited by data privacy concerns and the high cost of expert annotation. Consequently, no publicly available dataset or benchmark currently covers diverse surgical wound types, hindering the development of open-source surgical wound screening tools. To address this gap: (1) we present SurgWound, the first open-source dataset featuring diverse surgical wound types. It contains 686 surgical wound images annotated by three professional surgeons with eight fine-grained clinical attributes. (2) Based on SurgWound, we introduce the first benchmark for surgical wound diagnosis, which includes visual question answering (VQA) and report generation tasks for comprehensive evaluation. (3) Furthermore, we propose a three-stage learning framework, WoundQwen, for surgical wound diagnosis. The first stage predicts detailed wound characteristics using multiple MLLMs. The second stage uses these predictions as additional knowledge to assess infection risk and clinical urgency. The third stage integrates the diagnostic results to produce a comprehensive report. Our dataset and code are publicly available.