Scholarlink

<p>Nuclear mechanical properties influence transcription and cell signaling. Atomic Force Microscopy (AFM) provides high-resolution topography and mechanical properties of living cells. Here, we introduce AFMap-UNet, a deep learning architecture that integrates AFM-derived and optical microscopy data to achieve precise nuclear segmentation. AFM topography maps were combined with enhanced optical channels and processed through a two-stage, region-guided U-Net pipeline, achieving a precision-recall curve area with an average precision of 99% and a median Dice coefficient of 96%. Moreover, AFMap-UNet retains its performance even on small training sets of 15 images, highlighting scalability for data-constrained settings typical of AFM studies. To our knowledge, this represents the first high-performance deep learning model for spatially resolved quantification of nuclear mechanics, enabling new applications in AFM-based cell analysis and disease modeling.</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

AFMap-UNet enables accurate nuclear segmentation of atomic force microscopy images with minimal training data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月12日

查看全文
  • Arthur Henrique Rocha,
  • Cleyton Alexandre Biffe,
  • Ed Carlos Santos e Silva,
  • José Salvatore Leister Patane,
  • Carlos Alberto Rodrigues Costa,
  • Marco Antônio Gutierrez,
  • José Eduardo Krieger,
  • Ayumi Aurea Miyakawa

摘要

Nuclear mechanical properties influence transcription and cell signaling. Atomic Force Microscopy (AFM) provides high-resolution topography and mechanical properties of living cells. Here, we introduce AFMap-UNet, a deep learning architecture that integrates AFM-derived and optical microscopy data to achieve precise nuclear segmentation. AFM topography maps were combined with enhanced optical channels and processed through a two-stage, region-guided U-Net pipeline, achieving a precision-recall curve area with an average precision of 99% and a median Dice coefficient of 96%. Moreover, AFMap-UNet retains its performance even on small training sets of 15 images, highlighting scalability for data-constrained settings typical of AFM studies. To our knowledge, this represents the first high-performance deep learning model for spatially resolved quantification of nuclear mechanics, enabling new applications in AFM-based cell analysis and disease modeling.