Scholarlink

<p>In research on cells conducted in vitro, cell viability is determined using staining techniques. However, interference with subsequent observation of live cell growth limits their applicability for real-time or continuous investigation. To address this limitation, we developed a deep learning–based algorithm capable of classifying live and dead cancer cells from microscopic images without staining.&#xa0;In this study, microscopic images were first captured prior to staining, and then the same regions were imaged again after staining to obtain live, dead, and other cell labels using a conventional staining method. The stained images served as ground truth data for supervised training with the corresponding pre-staining images. The proposed model achieved an accuracy of 0.931 after 99 training epochs in distinguishing live and dead cells from unstained images. This framework accurately differentiated live and dead cells directly from pre-staining images, demonstrating performance comparable to conventional stained-image analysis. Moreover, the approach enabled estimation of spatial boundaries between live and dead cell populations. These results demonstrate the potential of this approach as a non-invasive technique for assessing cell viability in in vitro studies.</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

Viability classification of unstained cells in microscopic images using deep learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月20日

查看全文
  • Tomoaki Kyoden,
  • Shunsuke Akiguchi,
  • Ryo Murakami,
  • Tsugunobu Andoh,
  • Noboru Yamada

摘要

In research on cells conducted in vitro, cell viability is determined using staining techniques. However, interference with subsequent observation of live cell growth limits their applicability for real-time or continuous investigation. To address this limitation, we developed a deep learning–based algorithm capable of classifying live and dead cancer cells from microscopic images without staining. In this study, microscopic images were first captured prior to staining, and then the same regions were imaged again after staining to obtain live, dead, and other cell labels using a conventional staining method. The stained images served as ground truth data for supervised training with the corresponding pre-staining images. The proposed model achieved an accuracy of 0.931 after 99 training epochs in distinguishing live and dead cells from unstained images. This framework accurately differentiated live and dead cells directly from pre-staining images, demonstrating performance comparable to conventional stained-image analysis. Moreover, the approach enabled estimation of spatial boundaries between live and dead cell populations. These results demonstrate the potential of this approach as a non-invasive technique for assessing cell viability in in vitro studies.