Scholarlink

<p>Quantitative histomorphometric analysis of peripheral nerves is essential for assessing axonal regeneration and remyelination, but manual analysis is labor-intensive and impractical for large-scale studies. This study compared three automated morphometric approaches with conventional manual analysis to identify an efficient and reliable alternative. Six rat sciatic nerves, including three naïve nerves and three regenerating nerves following crush injury, were analyzed using semithin transverse sections stained with toluidine blue. Manual analysis by three independent observers served as the reference standard. Automated analyses were performed using Trainable Weka Segmentation, AxonDeepSeg, and AxonDeepSeg with manual refinement. Axon count, axon diameter, axon area, g-ratio, myelin thickness, and analysis time were compared. Manual analysis required the longest processing time (53.1 ± 16.5&#xa0;min for naïve nerves and 63.4 ± 3.9 min for regenerating nerves), whereas automated approaches reduced analysis time by more than half, with AxonDeepSeg being the fastest (10.2 ± 0.5&#xa0;min and 6.7 ± 0.2&#xa0;min, respectively). Agreement between automated and manual measurements was primarily evaluated using Bland–Altman analysis (mean bias and 95% limits of agreement). Fully automated methods demonstrated variable levels of agreement, with wider limits of agreement, particularly in regenerating nerves. A hybrid workflow combining deep learning-based segmentation with selective manual refinement showed reduced bias and relatively narrower limits of agreement compared with other automated approaches, while maintaining improved time efficiency.</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

Deep learning-based hybrid analysis for axonal regeneration and myelination in rat sciatic nerve

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Jun Hong Won,
  • Chawon Yun,
  • So Young Lee,
  • Tae Hyun Kim,
  • Jung Il Lee

摘要

Quantitative histomorphometric analysis of peripheral nerves is essential for assessing axonal regeneration and remyelination, but manual analysis is labor-intensive and impractical for large-scale studies. This study compared three automated morphometric approaches with conventional manual analysis to identify an efficient and reliable alternative. Six rat sciatic nerves, including three naïve nerves and three regenerating nerves following crush injury, were analyzed using semithin transverse sections stained with toluidine blue. Manual analysis by three independent observers served as the reference standard. Automated analyses were performed using Trainable Weka Segmentation, AxonDeepSeg, and AxonDeepSeg with manual refinement. Axon count, axon diameter, axon area, g-ratio, myelin thickness, and analysis time were compared. Manual analysis required the longest processing time (53.1 ± 16.5 min for naïve nerves and 63.4 ± 3.9 min for regenerating nerves), whereas automated approaches reduced analysis time by more than half, with AxonDeepSeg being the fastest (10.2 ± 0.5 min and 6.7 ± 0.2 min, respectively). Agreement between automated and manual measurements was primarily evaluated using Bland–Altman analysis (mean bias and 95% limits of agreement). Fully automated methods demonstrated variable levels of agreement, with wider limits of agreement, particularly in regenerating nerves. A hybrid workflow combining deep learning-based segmentation with selective manual refinement showed reduced bias and relatively narrower limits of agreement compared with other automated approaches, while maintaining improved time efficiency.