Scholarlink

Automated morphological analysis of human umbilical cord vein endothelial cells (HUVEC) was performed to study angiogenesis, a process in which new capillaries are formed from pre-existing capillaries. The data was classified into circular, elongated deformed (elongated), and slightly elongated deformed (other deformations). Morphological classification was performed using deep neural networks, specifically the Visual Transformers, Swift-Former, RegNet, and ResNet models. Tests were performed on a single-cell set of images created for this work, using Meta’s Segment Anything model. Several experiments were performed in different scenarios, using both real and generated images, to increase the number of images available and ensure optimal training of the deep learning models. This methodology achieved an accuracy of over 99%, which is higher than that obtained for other approaches in the literature, confirming its efficiency and accuracy in morphological analysis.

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 Models for Automatic Morphological Evaluation of Endothelial Cells

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Miriela Escobedo Nicot,
  • W. D. Font,
  • J. J. Luna,
  • L. F. Gomes

摘要

Automated morphological analysis of human umbilical cord vein endothelial cells (HUVEC) was performed to study angiogenesis, a process in which new capillaries are formed from pre-existing capillaries. The data was classified into circular, elongated deformed (elongated), and slightly elongated deformed (other deformations). Morphological classification was performed using deep neural networks, specifically the Visual Transformers, Swift-Former, RegNet, and ResNet models. Tests were performed on a single-cell set of images created for this work, using Meta’s Segment Anything model. Several experiments were performed in different scenarios, using both real and generated images, to increase the number of images available and ensure optimal training of the deep learning models. This methodology achieved an accuracy of over 99%, which is higher than that obtained for other approaches in the literature, confirming its efficiency and accuracy in morphological analysis.