Scholarlink

<p>This paper presents ChromoGAT, an advanced deep learning framework designed for precise chromosome segmentation, combining the strengths of U-Net and Graph Attention Networks (GAT). ChromoGAT effectively addresses the significant challenges of cytogenetic image analysis, including the segmentation of overlapping chromosomes and the accurate delineation of complex chromosomal morphologies. Evaluated on the Cell Image Library and Bio Image Chromosome Classification datasets, ChromoGAT demonstrates state-of-the-art performance, achieving a Dice coefficient of 0.92 and an F1-score of 0.91. The model consistently delivers robust segmentation across all 23 chromosome pairs, with Intersection over Union (IoU) scores exceeding 0.75 and Pixel Accuracy ranging from 0.85 to 0.97. Ablation studies highlight the critical contributions of both U-Net and GAT components, underscoring their synergistic integration. Qualitative analyses reveal ChromoGAT’s superior capability in resolving overlapping chromosomes, significantly outperforming traditional methods. This framework holds substantial promise for advancing automated cytogenetic analysis, with potential applications in the diagnosis of genetic disorders and cancer research, particularly in enhancing karyotype analysis and the detection of chromosomal abnormalities.</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

ChromoGAT: Precision chromosome segmentation through U-Net and GAT integration

  • 期刊论文

  • 发表日期：

    2025年2月27日

查看全文
  • A. G. Dinu,
  • V. G. Biju,
  • B. R. Vinod,
  • Nonam Chellapan,
  • Smitha G. Raj

摘要

This paper presents ChromoGAT, an advanced deep learning framework designed for precise chromosome segmentation, combining the strengths of U-Net and Graph Attention Networks (GAT). ChromoGAT effectively addresses the significant challenges of cytogenetic image analysis, including the segmentation of overlapping chromosomes and the accurate delineation of complex chromosomal morphologies. Evaluated on the Cell Image Library and Bio Image Chromosome Classification datasets, ChromoGAT demonstrates state-of-the-art performance, achieving a Dice coefficient of 0.92 and an F1-score of 0.91. The model consistently delivers robust segmentation across all 23 chromosome pairs, with Intersection over Union (IoU) scores exceeding 0.75 and Pixel Accuracy ranging from 0.85 to 0.97. Ablation studies highlight the critical contributions of both U-Net and GAT components, underscoring their synergistic integration. Qualitative analyses reveal ChromoGAT’s superior capability in resolving overlapping chromosomes, significantly outperforming traditional methods. This framework holds substantial promise for advancing automated cytogenetic analysis, with potential applications in the diagnosis of genetic disorders and cancer research, particularly in enhancing karyotype analysis and the detection of chromosomal abnormalities.