Scholarlink

<p>Consistency between paraffin blocks and corresponding histological slides is a key component of pathological quality control. In routine practice, verification is still largely based on manual visual comparison, which is both time-consuming and susceptible to human error. This study aims to develop a deep learning-based automated verification framework for reliable morphological matching between paraffin blocks and their histological slides. We retrospectively collected 2,220 paired paraffin block and histological slide images from the Second Affiliated Hospital of Soochow University. The AI framework integrates YOLOv11 for tissue detection and a Siamese network for feature matching. The AI framework demonstrated strong discriminative performance, achieving a mean AUC of 0.9809 in fivefold cross-validation. On the independent test set, the AI framework achieved an overall accuracy of 95.05%, significantly exceeding the average of 91.52% for manual verification. Subgroup analysis reached an accuracy of 98.01% for surgical resection specimens, and 88.77% for small biopsy specimens with limited morphological landmarks. Regarding efficiency, the AI processing time was &lt; 0.05&#xa0;s per slide, which was approximately 80 times faster than that of human verification. The AI framework also identified several mismatches that were missed by human reviewers. The proposed AI framework provides an efficient and more objective approach for block-slide consistency verification. The framework may serve as an automated screening tool that can support routine pathological quality control.</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 AI framework for automated quality control of paraffin block and slide consistency: a clinical evaluation and human–machine comparison study

  • 期刊论文

  • 发表日期：

    2026年4月18日

查看全文
  • Linfeng Tang,
  • Liwei Xie,
  • Xiaoling Yang,
  • Jinqiu Liu,
  • Xinru Zou,
  • Wei Xia

摘要

Consistency between paraffin blocks and corresponding histological slides is a key component of pathological quality control. In routine practice, verification is still largely based on manual visual comparison, which is both time-consuming and susceptible to human error. This study aims to develop a deep learning-based automated verification framework for reliable morphological matching between paraffin blocks and their histological slides. We retrospectively collected 2,220 paired paraffin block and histological slide images from the Second Affiliated Hospital of Soochow University. The AI framework integrates YOLOv11 for tissue detection and a Siamese network for feature matching. The AI framework demonstrated strong discriminative performance, achieving a mean AUC of 0.9809 in fivefold cross-validation. On the independent test set, the AI framework achieved an overall accuracy of 95.05%, significantly exceeding the average of 91.52% for manual verification. Subgroup analysis reached an accuracy of 98.01% for surgical resection specimens, and 88.77% for small biopsy specimens with limited morphological landmarks. Regarding efficiency, the AI processing time was < 0.05 s per slide, which was approximately 80 times faster than that of human verification. The AI framework also identified several mismatches that were missed by human reviewers. The proposed AI framework provides an efficient and more objective approach for block-slide consistency verification. The framework may serve as an automated screening tool that can support routine pathological quality control.