Scholarlink

<p>Bowel preparation quality significantly affects the diagnostic accuracy of colonoscopy. Inadequate preparation can obscure mucosal visualization, leading to repeated procedures and increased costs. Traditionally, bowel cleanliness is manually assessed by endoscopists, which introduces subjectivity and variability. To address this, we propose an artificial intelligence-based decision support system for objective and standardized evaluation. This study investigates two deep learning models using the Nerthus dataset: a frame-based classification approach with SqueezeNet and a video-based classification method using the R(2 + 1)D network. Video inputs were generated by grouping consecutive frames into fixed-length sequences. The R(2 + 1)D model achieved 97.78% accuracy, 96.88% sensitivity, 98.91% precision, and a 97.78% F1-score, demonstrating its high performance. Additionally, a user-friendly graphical user interface (GUI) was developed to facilitate seamless integration into clinical workflows. These results highlight the advantages of video-based AI methods for robust and consistent bowel preparation assessment.</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

Artificial intelligence supported colonoscopy bowel preparation assessment: a video-based approach

  • 期刊论文

  • 发表日期：

    2025年7月30日

查看全文
  • Faruk Enes Oğuz,
  • Ahmet Alkan,
  • Artur Klepaczko,
  • Pawel Strumillo,
  • Murat İspiroğlu

摘要

Bowel preparation quality significantly affects the diagnostic accuracy of colonoscopy. Inadequate preparation can obscure mucosal visualization, leading to repeated procedures and increased costs. Traditionally, bowel cleanliness is manually assessed by endoscopists, which introduces subjectivity and variability. To address this, we propose an artificial intelligence-based decision support system for objective and standardized evaluation. This study investigates two deep learning models using the Nerthus dataset: a frame-based classification approach with SqueezeNet and a video-based classification method using the R(2 + 1)D network. Video inputs were generated by grouping consecutive frames into fixed-length sequences. The R(2 + 1)D model achieved 97.78% accuracy, 96.88% sensitivity, 98.91% precision, and a 97.78% F1-score, demonstrating its high performance. Additionally, a user-friendly graphical user interface (GUI) was developed to facilitate seamless integration into clinical workflows. These results highlight the advantages of video-based AI methods for robust and consistent bowel preparation assessment.