Scholarlink

The study of white blood cells (WBC) is crucial in clinical practice due to their role in estimating the immune system general health state. However, the manual evaluation of blood smears is often tedious and susceptible to errors, making it a prime candidate for computer automation. This paper proposes an alternative method for WBC nuclei segmentation using blue-to-green and green-to-blue ratio subtraction, followed by Otsu’s thresholding. Furthermore, a size filtering and nuclei regions of interest identification algorithm, designed for FPGA acceleration, is applied to obtain the final binary mask. Additionally, blue-to-red ratio computation reduces segmentation errors. The nuclei segmentation method achieves a Dice similarity coefficient (DSC) of 0.8444 without correction and 0.8777 with correction. The nuclei identification algorithm tested on LISC, and a private dataset obtained an accuracy of 0.9982.

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

White Blood Cell Nuclei Detection and Segmentation on Ratio Channels G/B, B/G, and B/R for FPGA Implementation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Roberto C. Araujo-Parra,
  • Aura M. Jiménez-Garduño,
  • Jorge F. Martinez-Carballido

摘要

The study of white blood cells (WBC) is crucial in clinical practice due to their role in estimating the immune system general health state. However, the manual evaluation of blood smears is often tedious and susceptible to errors, making it a prime candidate for computer automation. This paper proposes an alternative method for WBC nuclei segmentation using blue-to-green and green-to-blue ratio subtraction, followed by Otsu’s thresholding. Furthermore, a size filtering and nuclei regions of interest identification algorithm, designed for FPGA acceleration, is applied to obtain the final binary mask. Additionally, blue-to-red ratio computation reduces segmentation errors. The nuclei segmentation method achieves a Dice similarity coefficient (DSC) of 0.8444 without correction and 0.8777 with correction. The nuclei identification algorithm tested on LISC, and a private dataset obtained an accuracy of 0.9982.