Fast automatic quantitative cell replication with fluorescent live cell imaging
 ChingWei Wang^{1}Email author
https://doi.org/10.1186/147121051321
© Wang; licensee BioMed Central Ltd. 2012
Received: 20 July 2011
Accepted: 31 January 2012
Published: 31 January 2012
Abstract
Background
live cell imaging is a useful tool to monitor cellular activities in living systems. It is often necessary in cancer research or experimental research to quantify the dividing capabilities of cells or the cell proliferation level when investigating manipulations of the cells or their environment. Manual quantification of fluorescence microscopic image is difficult because human is neither sensitive to fine differences in color intensity nor effective to count and average fluorescence level among cells. However, autoquantification is not a straightforward problem to solve. As the sampling location of the microscopy changes, the amount of cells in individual microscopic images varies, which makes simple measurement methods such as the sum of stain intensity values or the total number of positive stain within each image inapplicable. Thus, automated quantification with robust cell segmentation techniques is required.
Results
An automated quantification system with robust cell segmentation technique are presented. The experimental results in application to monitor cellular replication activities show that the quantitative score is promising to represent the cell replication level, and scores for images from different cell replication groups are demonstrated to be statistically significantly different using ANOVA, LSD and Tukey HSD tests (pvalue < 0.01). In addition, the technique is fast and takes less than 0.5 second for high resolution microscopic images (with image dimension 2560 × 1920).
Conclusion
A robust automated quantification method of live cell imaging is built to measure the cell replication level, providing a robust quantitative analysis system in fluorescent live cell imaging. In addition, the presented unsupervised entropy based cell segmentation for live cell images is demonstrated to be also applicable for nuclear segmentation of IHC tissue images.
Keywords
1 Background
However, manual quantification is subjective and results tend to be poorly reproducible. Worse, the ability of manual quantification is limited as human eye is not sensitive to fine differences in color intensity, and only restricted semiquantitative mode can be provided. As a result, an automated quantification approach of live cell imaging is needed for monitoring cell proliferation activities.
where n is the number of cells, A is the set of cell locations, and f_{ j }is the fluorescence level at location j.
Another popular approaches for nuclear detection are watershed algorithms [12]. As in practice, the VincentSoille watershed tends to produce an oversegmentation(Figure 2(e)), we tested a watershed algorithm (adapted from [13]) and markercontrolled watershed method [14] with optimized empiricallyset parameters (Figure 2(g,f)). However, the watershed methods still produce many false positives and false negatives. As a result, a robust method for nuclear segmentation in challenging IHC tissue images is needed, and in this study, an entropybased method is presented, which can also be extended for nuclear segmentation in IHC tissue images (an output by the proposed method is shown in Figure 2(h)).
In this paper, an entropy based cell segmentation method is developed for fluorescent microscopic images and an automated quantification system of live cell imaging is built to analyze the cell replication level. The method is invariant to the camera sampling location and the amount of cells appearing in the image. In addition, it takes less than 0.5 second to process each image with dimension 2560 × 1920.
2 Results and Discussion
ANOVA Test on the quantification score
sum of squares  df  mean square  F  Significance  

Between groups  1429.6  2  714.8  43.754  < 0.001 
Within groups  98.022  6  16.337  
Total  1527.622  8 
Multiple Comparison using Tukey HSD and LSD tests
Dataset (I)  Dataset (J)  mean diff. (IJ)  std. error  Significance  

Tukey HSD  p63  BRCA1  15.29  3.3  0.008* 
Scr  30.87  3.3  < 0.001*  
BRCA1  p63  15.29  3.3  0.008  
Scr  15.579  3.3  0.008*  
Scr  p63  30.87  3.3  < 0.001*  
BRCA1  15.58  3.3  0.008*  
LSD  p63  BRCA1  15.29  3.3  0.004* 
Scr  30.87  3.3  < 0.001*  
BRCA1  p63  15.29  3.3  0.004  
Scr  15.579  3.3  0.003*  
Scr  p63  30.87  3.3  < 0.001*  
BRCA1  15.58  3.3  0.003* 
2.1 Results of the extension to IHC tissue images
3 Conclusions
It is often necessary in experimental research to quantify the dividing capabilities of cells when investigating manipulations of the cells or their environment. The presented technique provides a new type of information in monitoring cell replication activity and greatly empowers live cell imaging in studying cellular process. The availability of this novel technique should facilitate a more precise and comprehensive evaluation of cell proliferation and aid in the interpretation of results.
4 Methods
4.1 Materials in fluorescence microscopy images
4.2 Materials in immunohistochemistry tissue images
An extended version of the proposed entropybased segmentation technique is built for nuclear segmentation of IHC images. A tissue microarray (TMA) slide was scanned using Aperio Scanscope CS2 (Aperio Technologies Inc. San Diego USA), at 40x objective magnification. Nine different tissue cores were randomly selected for evaluation; the image size of individual tissue cores is around 2896 × 2756 and the nuclear areas of each tissue core were manually marked to produce ground truth data. A quantitative performance evaluation was conducted by comparing the ground truth data and the system output.
4.3 Automatic cell based quantification approach in fluorescent microscopic images
In general image data, the color of the image pixels represents the appearance of the surface of an object. This however is not always applicable to the microscopic images. In the fluorescent microscopic images, the gray color of the background regions can not represent the true appearance of specimens. These background areas are in fact transparent and become gray when captured using a digital microscopic system. Hence, segmentation by general unsupervised clustering techniques, such as Otsu or Kmeans, based on the raw digital image appearance information does not produce good segmentation results because those information can be misleading.
4.3.1 Extraction of Foreground Color Information
Color representation
where I_{1}(λ) is the intensity of light of wavelength λ transmitted through the specimen (the intensity of light detected), I_{0}(λ) is the intensity of light of wavelength λ entering the specimen, α is the amount of stain per unit area of the specimen, and c(λ) is a wavelengthdependent factor reflecting the absorption characteristics of the particular stain.
where each row represents a specific stain and each column represents the OD as detected by RGB channels for individual stain.
Color deconvolution [17] can be used to obtain independent information about each stain's contribution based on orthonormal transformation of the RGB information, and the transformation has to be normalized to achieve correct balancing of the absorbtion factor for separate stains. For normalization, each OD vector is divided by its total length to obtain a normalized OD array A. If C is the 2 × 1 vector for amounts of the two stains at a particular pixel, then the vector of OD levels detected at that pixel is D = CA. Defining K = A^{1} as the colordeconvolution array, we can therefore obtain individual stain information by C = KD.
4.3.2 Local Entropybased Cell Segmentation
Given the extracted color image, I_{ B }, the cell is segmented by local entropybased segmentation. According to Shannon's theorem [18], if the event i occurs from a set of valid events with the probability p_{ i }, the amount of uncertainty related to the event is equal to H_{ i }= log(p_{ i })(bits/symbol), and the amount of the uncertainty that the source of the events generates is equal to H =  Σ (p_{ i }log(p_{ i }))(bits). The idea behind local entropy method is to divide the processed image into separate regions and then to analyze each region separately as information source.
where j ∈ {0...2^{ c } 1}, A = {0...j} and B = {2^{ c } 1...j}.
where i ∈ {0...2^{ c } 1}.
4.3.3 Quantification Function
4.4 Software
The developed software is platform independent and thus can be executed in different operation systems such as Windows, Linux or Mac. The software with some test images can be downloaded from the author's website (http://wwwo.ntust.edu.tw/~cweiwang/Cell/).
4.5 Extension to IHC tissue images
Next, a multistage entropybased segmentation of nuclei is applied. After calculating 2D image histogram entropy function, we first apply an eight stage maximum entropy function to automatically separate input image into eight layers, and then a two stage entropy function to extract potential regions of nuclei, which is then processed by morphological operations to produce final nuclear segmentation results. The algorithm is described below.

divide histogram into four equal subhistograms P_{1}, P_{2}, P_{3}, P_{4}, obtaining j_{1}, j_{3}, j_{5} where j ∈ 0...2^{ c } 1

compute maximum entropy points j_{0}, j_{2}, j_{4}, j_{6} for the four different P intervals, where j_{0} = arg max H(P_{1}), j_{2} = arg max H(P_{2}), j_{4} = arg max H(P_{3}), j_{6} = arg max H(P_{4}).

use j_{0}...j_{6} to categorize input image into eight layers

calculate new histogram P*

compute j* = arg max H(P*) and categorize input image into 2 categories, including nuclei and nonnuclei.

apply the morphological operations described below
where K = {x  r,..., x + r}, L = {y  r,..., y + r}, and r is empirically set as 3.
Declarations
Authors’ Affiliations
References
 Frigault M, Lacoste J, Swift J, Brown C: Livecell microscopy  tips and tools. Cell Science 2009, 122: 753–767. 10.1242/jcs.033837View ArticleGoogle Scholar
 Kitamura E, Blow J, Tanaka T: LiveCell Imaging Reveals Replication of Individual Replicons in Eukaryotic Replication Factories. Cell 2006, 125(7):1297–1308. 10.1016/j.cell.2006.04.041PubMed CentralView ArticlePubMedGoogle Scholar
 Drost R, Jonkers J: Preclinical mouse models for BRCA1associated breast cancer. British Journal of Cancer 2009, 101: 1651–1657. 10.1038/sj.bjc.6605350PubMed CentralView ArticlePubMedGoogle Scholar
 Wang X, Mori I, Tang W, Nakamura M, Nakamura Y, Sato M, Sakurai T, Kakudo K: P63 expression in normal, hyperplastic and malignant breast tissues. Breast Cancer 2002, 9(3):216–219. 10.1007/BF02967592View ArticlePubMedGoogle Scholar
 Huang F, Reeves K, Han X, et al.: Identification of candidate molecular markers predicting sensitivity in solid tumors to dasaranib: rationale for patient selection. Cancer Research 2007, 65: 2226–2238.View ArticleGoogle Scholar
 Buckley N, Conlon S, Jirstrom K, Kay E, Crawford N, et al.: The ANp63 proteins are key allies of BRCA1 in the prevention of basallike breast cancer. Cancer Research 2011, 71: 1933–1944. 10.1158/00085472.CAN102717View ArticlePubMedGoogle Scholar
 Wu HS, Deligdisch L, Gil J: Segmentation of microscopic nuclear image  A review Recent Res. Devel Electronics 2004., 2:Google Scholar
 Rexhepaj E, Brennan DJ, Holloway P, Kay EW, McCann AH, Landberg G, Duffy M, Jirstrom K, Gallagher W: Quantification of histochemical staining by color deconvolution. Breast Cancer Research 2008., 10(5):Google Scholar
 Mao K, Zhao P, Tan PH: Supervised LearningBased Cell Image Segmentation for P53 Immunohistochemistry. IEEE Trans Biomedical Engineering 2006, 53(6):1153–1163. 10.1109/TBME.2006.873538View ArticleGoogle Scholar
 Otsu N: A threshold selection method from graylevel histograms. IEEE Trans Sys Man Cyber 1979, 9: 62–66.View ArticleGoogle Scholar
 Lloyd SP: Least squares quantization in PCM. IEEE Transactions on Information Theory 1982, 28(2):129–137. 10.1109/TIT.1982.1056489View ArticleGoogle Scholar
 Vincent L, Soille P: Watersheds in digital spaces: An efficient algorithm based on immersion simulations. IEEE Trans Pattern Anal Mach Intell 1991, 13(6):583–598. 10.1109/34.87344View ArticleGoogle Scholar
 Eddins S: Cell segmentation.[http://blogs.mathworks.com/steve/2006/06/02/cellsegmentation/]
 Matlab : MarkerControlled Watershed Segmentation. Image Processing Toolbox 7.0.Google Scholar
 SPSS Inc: SPSS for Windows, Rel.17.0.1. 2008. Chicago: SPSS Inc;Google Scholar
 Van Der Laak JA, Pahlplatz MM, Hanselaar AG, de Wilde PC: Huesaturationdensity (HSD) model for stain recognition in digital images from transmitted light microscopy. Cytometry 2000., 39(4):Google Scholar
 Ruifrok AC, Johnston DA: Quantification of histochemical staining by color deconvolution. Analytical and Quantitative Cytology and Histology 2001, 23: 291–299.PubMedGoogle Scholar
 Shannon C: A mathematical theory of communication. Bell system technical journal 1948, 27: 379–423.View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.