Advanced Search

ISSN1001-3806 CN51-1125/TN Map

Volume 43 Issue 4
Jul.  2019
Article Contents
Turn off MathJax

Citation:

White blood cell segmentation based on RGB and C-Y color space

  • Corresponding author: LÜ Xiaoqi, lxiaoqi@imut.edu.cn
  • Received Date: 2018-08-21
    Accepted Date: 2018-09-14
  • In order to solve the problems of color approximation and uneven brightness between white blood cells and other cells in blood cell images, a white blood cell segmentation method based on C - Y color space was adopted. The original RGB image was transformed into C - Y image. C - Y image was separated to obtain B - Y color component image with all information. Secondly, a complete white blood cell image is obtained by screening the area of the connected region, opening operation and pixel operation. The contrast stretched G image was extracted. The approximate position of nucleus was obtained by threshold segmentation. And then, the complete nuclear image was segmented by area screening and open operation of connected domain. The results show that the algorithm in this paper has good segmentation accuracy for eosinophils, lymphocytes, monocytes and neutrophils, i.e., 94.33%, 91.60%, 97.72% and 98.66% respectively. The algorithm in this paper can segment white blood cells more completely. It lays a foundation for the follow-up classification research.
  • 加载中
  • [1]

    ANGULO J, FLANDRIN G. Microscopic image analysis using mathematical morphology:Application to haematological cytology[J]. Science, Technology and Education of Microscopy: An Overview, 2003, 21(3):304-312.
    [2]

    DORINI L B, MINETTO R, LEITE N J. White blood cell segmentation using morphological operators and scale-space analysis[J]. Brazilian Symposium on Computer Graphics and Image Processing, 2007, 20(6):294-304.
    [3]

    ELDAHSHAN K, YOUSSEF M, MASAMEER E, et al. Comparison of segmentation framework on digital microscope images for acute lymphoblastic leukemia diagnosis using RGB and HSV color spaces[J]. Biotechnology & Bioengineering, 2015, 2(2):27-34.
    [4]

    PIURI V, SCOTTI F. Morphological classification of blood leucocytes by microscope images[C]// IEEE International Conference on Computational Intelligence for Measurement Systems and Applications. New York, USA: IEEE, 2004: 103-108.
    [5]

    JOSHI M D, KARODE A H, SURALKAR S R. White blood cells segmentation and classification to detect acute leukemia[J]. International Journal of Emerging Trends and Technology in Computer Science, 2013, 2(3): 147-151.
    [6]

    RAZZAK M I, NAZ S. Microscopic blood smear segmentation and classification using deep contour aware CNN and extreme machine learning[C]//IEEE Conference on Computer Vision and Pattern Recognition Workshops. Boston, USA: IEEE Computer Society, 2017: 801-807.
    [7]

    SAJJAD M, KHAN S, SHOAIB M, et al. Computer aided system for leukocytes classification and segmentation in blood smear images[C]// International Conference on Frontiers of Information Technology.New York, USA: IEEE, 2017: 99-104.
    [8]

    CAI J, BAO X D, WU L, et al.Automated segmentation of colored leukocyte images based on the active contour model[J].Journal of Biomedical Engineering Research, 2005, 24(4): 218-222 (in Chinese).
    [9]

    YANG J L, ZHENG W L, MA W L, et al.Segmentation for image of white blood cells based on improved BP neural network[J].Microcomputer Information, 2008, 24(33):294-295(in Chinese).
    [10]

    YIN Y, WANG Y, LIU D P.Improved MEANSHIFT method for blood cell image segmentation. Computer Engineering and Applications, 2010, 46(6):178-180(in Chinese).
    [11]

    WANG J, CAO Y P, CHENG X Sh.Leukocyte image segmentation algorithm based on hue component histogram transformation[J].Optics & Optoelectronic Technology, 2013, 11(2):000074(in Chinese).
    [12]

    HOU H, SHI Y X.Application of the improved watershed algorithm based on distance transform in white blood cell segmentation[J].Computing Technology and Automation, 2016, 35(3):81-84(in Chinese).
    [13]

    LI Y, ZHU R, MI L, et al. Segmentation of white blood cell from acute lymphoblastic leukemia images using dual-threshold method[J]. Computational and Mathematical Methods in Medicine, 2016, 10(1):1-12.
    [14]

    HALIM N H A, MASHOR M Y, NASIR A S A, et al. Colour image segmentation using unsupervised clustering technique for acute leukemia images[C]// American Institute of Physics Conference Series. Melville, New York, USA: AIP Publishing LLC, 2015: 243-246.
    [15]

    LI Y.Research on the techniques of dual-threshold based peripheral blood leukocyte microscope image segmentation[D].Xi'an: The University of Chinese Academy of Sciences(Xi'an Institute of Optics & Precision Mechanics, Chinese Academy of Sciences), 2016: 1-76(in Chinese).
    [16]

    GU G H, CUI D.Automatic segmentation algorithm for leukocyte images[J].Chinese Journal of Scientific Instrument, 2009, 30(9):1874-1879(in Chinese).
    [17]

    DOUGHERTY E R. Electronic imaging technology[M]. Miami, USA:SPIE Optical Engineering Press, 1999:564-605.
    [18]

    HUANG Zh, ZHAO J W, CHU J J, et al.Efficient segmentation algorithm for white blood cell. Computer Engineering and Applications, 2016, 52(21):188-194(in Chinese).
    [19]

    TOH L B, MASHOR M Y, EHKAN P, et al. Image segmentation for acute leukemia cells using color thresholding and median filter[J]. Journal of Telecommunication, Electronic and Computer Engineering, 2018, 10(5):69-74.
    [20]

    LIU J, LEONG T Y, CHEE K B, et al. Set-based cascading approaches for magnetic resonance (MR) image segmentation (SCAMIS)[J]. AMIA Annual Symposium Proceedings, 2006, 49(9):504-508.
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(4) / Tables(1)

Article views(4626) PDF downloads(23) Cited by()

Proportional views

White blood cell segmentation based on RGB and C-Y color space

    Corresponding author: LÜ Xiaoqi, lxiaoqi@imut.edu.cn
  • 1. Inner Mongolia Key Laboratory of Pattern Recognition and Intelligent Image Processing, School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou 014010, China
  • 2. Inner Mongolia University of Technology, School of Information Engineering, Hohhot 010051, China
  • 3. School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China

Abstract: In order to solve the problems of color approximation and uneven brightness between white blood cells and other cells in blood cell images, a white blood cell segmentation method based on C - Y color space was adopted. The original RGB image was transformed into C - Y image. C - Y image was separated to obtain B - Y color component image with all information. Secondly, a complete white blood cell image is obtained by screening the area of the connected region, opening operation and pixel operation. The contrast stretched G image was extracted. The approximate position of nucleus was obtained by threshold segmentation. And then, the complete nuclear image was segmented by area screening and open operation of connected domain. The results show that the algorithm in this paper has good segmentation accuracy for eosinophils, lymphocytes, monocytes and neutrophils, i.e., 94.33%, 91.60%, 97.72% and 98.66% respectively. The algorithm in this paper can segment white blood cells more completely. It lays a foundation for the follow-up classification research.

引言
  • 医学图像分析是当前最有价值也最复杂的研究领域之一。医学图像分析是当前非常热门的研究领域,现在学者致力于将人工智能应用于医学图像处理,使其过程更加便捷。常规血检中最重要的一项是白细胞计数,传统的白细胞计数是医生使用光学显微镜观察血液图片来实现,医生的经验、精力等对计数的准确性有至关重要的影响,并且通过分析不同种类白细胞在血液中的含量,是检验白血病的重要手段之一,白细胞在人体中也具有非常重要的防御功能,因此白细胞的研究意义极大。

    目前,国外在白细胞分割方面的研究有很多。ANGULO等人[1]等提出了基于自动阈值和二进制滤波的分割算法,但没有定量分析实验结果;DORINI等人[2]提出了一种分水岭变换的分割方法;ELDAHSHAN等人[3]将六角锥体模型颜色空间的H通道图像用于图像分割,效果显著,但没有讨论该方法在细胞重叠时的分割精度;PIURI等人[4]提出了一种基于边缘检测的分割方法,并利用形态学特征训练神经网络来识别细胞。JOSHID等人[5]运用Otsu阈值分割算法分割细胞质,方法简单,但是过程繁琐。RAZZAK等人[6]提出一种基于传统轮廓感知算法的卷积神经网络,具有较高的分割精度,但是实验对硬件、数据集要求较高;SAJJAD等人[7]提出一种将色调、饱和度、强度(hue, saturation,intensity, HSI)颜色空间与K-means相结合的聚类分割算法,计算成本低。

    近几年白细胞分割也是国内的研究热点之一。2005年,CAI等人[8]提出基于HSI彩色空间分割细胞核,确定白细胞的大致位置,再结合流域算法、主动轮廓模型分割完整的白细胞图像,取得了不错的效果,但是比较依赖图像的颜色信息。2008年,YANG等人[9]提出了采用Levenberg-Marquardt(L-M)算法的改进多层前馈网络(back propagation,BP),取得较好的分割结果。2010年,YIN等人[10]提出改进MEANSHIFT方法分割出细胞核,再结合区域生长算法完成白细胞分割,但是其算法在分割细胞质时效果较差。2013年,WANG等人[11]提出一种基于H分量直方图变换的分割算法,但是没有定量分析分割结果。2016年,HOU等人[12]提出一种结合最大期望算法(expectation-maximization algorithm,EM)聚类的改进分水岭算法,效果显著。

    上述文献中所用实验数据集中的白细胞图像和背景色差较大,降低了分割难度和复杂度。而本文中提出的基于RGB, C-Y颜色空间的分割算法,在色差较小的图像中也能准确实施分割。

1.   方法
  • 主要利用白细胞RGB图像中G颜色分量将白细胞的细胞核与背景分离,得到比较完整的细胞核图像。该流程主要分为5个步骤(如图 1所示):(1)对原图进行对比度拉伸,得到拉伸后的图像; (2)提取拉伸后图像的G颜色分量图像; (3)二值化G颜色分量图像,标记连通域,根据连通域面积进行粗分割,获得细胞核的二值图; (4)将G颜色分量的二值图像与粗分割得到的图像信息融合,得到完整细胞核图像; (5)在RGB颜色空间中显示结果图像。

    Figure 1.  Segmentation flow chart of nucleus

  • 利用颜色空间分割白细胞[13-16]的方法很多,大多数是提取H通道图像分割白细胞。本文中首次将C-Y颜色空间中的B-Y颜色分量用于分割白细胞图像。该流程主要分为7个步骤(如图 2所示):(1)将原图从RGB颜色空间直接转化为B-Y颜色分量图像; (2)得到B-Y颜色分量的二值化图像; (3)根据连通域的面积粗分割得到大概的白细胞图像; (4)开运算处理,像素点操作,分离边缘粘连部分; (5)进行面积筛选,去除粘连部分,得到白细胞图像; (6)用7×7的中值滤波器平滑图像边缘; (7)在原RGB图像上显示实验结果。

    Figure 2.  Segmentation flow chart of white blood cells

  • 对比度拉伸,也称灰度拉伸,是一种简单的线性点运算。进行对比度拉伸的目的是扩展图像的直方图,使其充满整个灰度等级范围内。公式如下:

    式中,f(x, y)为输入图像, g(x, y)为输出图像; A=min[f(x, y)], 是原图最小灰度级; D=max[f(x, y)], 是原图最大灰度级。

    如果灰度图像中最小值A=0,最大值D=255,则图像没有什么改变。因此,针对本文中所使用数据集,经过多次实验,最终确定G通道图像A=88, D=250;B通道图像A=130, D=250;R通道图像A=50, D=255,此时效果最佳。结果如图 3所示。

    Figure 3.  a—the original RGB image   b—the Contrast stretched RGB images

  • 本文中所使用的C-Y颜色空间,与其它颜色空间类似,都可以通过公式计算将图像从RGB颜色空间转为C-Y颜色空间。C-Y颜色空间包括R-Y, B-Y, G-Y 3个颜色分量和一个亮度分量Y[17]。定义颜色时,只需要3个颜色分量中的任意两个就行。

    RGB颜色空间到C-Y颜色空间需要一个3×3的矩阵,公式如下:

    (2) 式表示,C-Y颜色空间采用的是R-Y, B-Y两个颜色分量和亮度分量Y,而X, Z分别表示R-Y, B-Y的两个颜色分量。

    (3) 式、(4)式中的S表示C-Y颜色空间的饱和度。当饱和度S=0时,R-Y, B-Y两个颜色分量重合,不能区分。

    也可以通过下式将图像从C-Y颜色空间转换到RGB颜色空间:

2.   实验结果与分析
  • 为验证算法的正确性和稳定性, 对白细胞图像应用本文中的算法进行了实验。实验环境为Windows 7, Pentium 3.4GHz, 8GB, Python 3.6。本文中所采用数据集中的图像都为JPEG格式,分辨率为320×240。由于染色种类、时间差异和成像设备的不同,导致所数据集图像中细胞之间、细胞与背景之间色差较小,并且有不同程度的黑色边框,因此,使用本文中数据集分割起来更加困难。

  • 本实验中采用参考文献[18]中分割算法和参考文献[19]中分割算法作为对比算法。参考文献[18]中先在图像中自动检测白细胞,利用灰度直方图分割细胞核,通过GrabCut算法迭代分割出细胞质,此方法延续了GrabCut算法速度快、精度高的特点,但是算法中还存在如何确定检测窗口的大小、提高自适性等问题。参考文献[20]中将图像中的颜色信息与阈值算法结合用于白细胞分割,方法简单、有效,但是对图像的颜色信息过于依赖。实验中使用参考文献[18]中分割算法、参考文献[19]中分割算法、本文中算法分别嗜酸性粒细胞、淋巴细胞、单核细胞和嗜中性粒细胞图像进行分割,将分割后的图像与专家手工标记的图像进行比较,实验结果如图 4所示。从实验结果可以看出,本文中算法对粘连细胞也有较好的分割性能。

    Figure 4.  Experiment results

    本文中采取两种分割误差测量方法、准确度以及分割时间对3种分割方法进行评价,O是过分割率、U是欠分割率[20]C是准确度:

    式中, OP是分割结果中应该包含但没有包含的像素点;UP是分割结果中不应该包含但包含的像素点;DP是手工分割结果中包含的全部像素点;其中下标P表示图像的像素点。表 1中给出了3种分割算法在嗜酸性粒细胞、淋巴细胞、单核细胞和嗜中性粒细胞4种细胞图像中的分割误差和准确度。

    algorithm in reference [18] algorithm in reference [19] our algorithm
    O/% U/% C/% O/% U/% C/% O/% U/% C/%
    eosinophils 18.61 0.30 81.09 2.39 3.93 93.68 2.18 3.49 94.33
    lymphocytes 58.95 0.00 41.96 11.78 4.16 84.06 7.73 0.67 91.60
    monocytes 6.71 0.21 93.08 0.00 14.44 85.56 1.10 1.18 97.72
    neutrophils 44.09 0.00 55.91 0.49 27.17 72.34 0.54 0.79 98.67
    section time/s 0.9340 0.3288 0.3152

    Table 1.  Parameters for three image segmentation algorithms

    表 1中对于嗜酸性粒细胞、淋巴细胞、单核细胞和嗜中性粒细胞图像分割的准确度可以看出,本文中算法明显优于其它两种算法,对于单核细胞和嗜中性粒细胞图像分割的效果明显,但是对于粘连情况明显的淋巴细胞分割性能还需进一步优化。

3.   结论
  • 本文中提出一种基于RGB, C-Y颜色空间的分割算法, 并对血液显微图像的白细胞进行了分割实验研究。通过RGB颜色空间中的G颜色分量,可以准确地分割出细胞核,避免血小板和背景的干扰。同时,首次将C-Y颜色空间的R-Y颜色分量运用于细胞分割,对嗜酸性粒细胞、淋巴细胞、单核细胞和嗜中性粒细胞图像都具有较好的分割精度,分别达到了94.33%, 91.60%, 97.72%, 98.66%的准确率,为后续分类实验提供了很好的支持。

    由于细胞粘连情况复杂多样,因此进一步研究更加有效地分割方法是今后的重要方向。

Reference (20)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return