Escolar Documentos
Profissional Documentos
Cultura Documentos
18
doi: 10.1111/jmi.12292
Institute of Clinical Physiology (IFC), National Research Council of Italy (CNR), Pisa, Italy
Department of Veterinary Science, University of Pisa, Pisa, Italy
Summary
HisTOOLogy is an open-source software for the quantification
of digital colour images of histological sections. The simple
graphical user interface enables both expert and non-expert
users to rapidly extract useful information from stained tissue
sections. The softwares main feature is a generalizable colour
separation algorithm based on k-means clustering which accurately and reproducibly returns the amount of colour per
unit area for any stain, thus allowing the quantification of tissue components. Here we describe HisTOOLogys algorithms
and graphical user interface structure, showing how it can be
used to separate different dye colours in several classical stains.
In addition, to demonstrate how the tool can be employed to
obtain quantitative information on biological tissues, the effect
of different hepatic tissue decellularization protocols on cell removal and matrix preservation was assessed through image
analysis using HisTOOLogy and compared with conventional
DNA and total protein content assays. HisTOOLogys performance was also compared with ImageJs colour deconvolution
plug-in, demonstrating its advantages in terms of ease of use
and speed of colour separation.
Introduction
Differential staining is widely used in histological samples to
investigate tissue architecture because it enables specific features or compounds of interest to be highlighted (Kiernan,
1999; Horobin & Kiernan, 2002; Coleman, 2006). Specific
chemical dyes can be used to stain tissue slices in which each
component is represented by a different colour (e.g. extracellular matrix, proteins, polysaccharides, cell cytoplasm, cell
nuclei). Histological sections can then be easily analysed using
a microscope, recording the size, shape and number of tissue
components.
Correspondence to: Chiara Magliaro, Research Center E. Piaggio, Largo Lucio
Lazzarino, 1, 56122, Pisa, Italy. Tel: +39-0502217061; fax: +39-0502217051;
e-mail: chiara.magliaro@googlemail.com
C 2015 The Authors
C 2015 Royal Microscopical Society
Journal of Microscopy
C. MAGLIARO ET AL.
H IST O O L OGY
Fig. 1. The k-means clustering colour separation algorithm for histochemical stains. In the first step, 24-bit colour images are represented in the RGB
and CYMK colour models. The algorithm transforms the image matrix into an array of size (number of pixels number of colours). For a given input
value of k (number of colours to separate), k-means clustering is performed to partition the array into clusters, returning the IDX vector containing
cluster indices (i.e. k value) for each row. In the second step, the average Silhouette value Sv is calculated for each colour model, and the model with the
highest average Sv is selected for subsequent steps. The third step involves pixel colour assignment, whereby each pixel is allocated a specific colour (from
1 to k) according to whether it falls within a predetermined selection (see text for selection criterion). An optional multiband thresholding procedure is
also available for fine tuning of pixel colour assignment. Finally, the fourth step involves image display and pixel counting. Simplified images showing all
k colours are displayed and the algorithm returns the total area occupied by each colour.
for all the channels ch. For example, in the RGB colour
model, a pixel with I = (IR , IG , IB ) is assigned to a colour
i = 1: k if m R < I R < MR mG < I G < MG m B <
I B < MB . For each colour, subsequent counting of the
pixels satisfying this criterion gives the amount of area
covered by that colour.
For low contrast images or images with a low signalto-noise ratio or with artefacts, the user may feel it
C. MAGLIARO ET AL.
r
r
r
Fig. 2. HisTOOLogy index GUI showing the principal functions for loading, sizing, and cropping an image. The image loaded is H&E stained
porcine liver, with a pixel size of 0.22 m.
HisTOOLogy: an outline
The colour separation algorithm described above has been
implemented as an open-source software named HisR
TOOLogy developed in Matlab
(The MathWorks-sTM, Inc.,
Natick, Massachussets, USA). These and other standard image analysis algorithms are built into a series of GUIs designed
for quantification of histological stains. Here we report a detailed description of all the functions and GUIs implemented in
HisTOOLogy.
Load an image and set pixel size.
Users can load 24-bit colour
images saved in the most common formats (i.e. *.tiff, *.gif and
*.bmp). The image is visualized in the HisTOOLogy index GUI
(Fig. 2) and a structure (the datamatrix) which stores all the
information related to the image, such as pathname and filename is created. The datamatrix is given the same name as
the image file and stored in the same folder. A pop-up menu
enables users to assign the pixel size and the magnification
used.
Region of interest selection.
An image may need to be cropped
to remove regions of noninterest. A tool to select a region of
interest (ROI) enables the user to choose specific areas or to
eliminate others [e.g. to remove vignetting artefacts (Inou`e &
Spring, 1997)]. The tool opens a new window, in which the
user can select the region to crop using a mouse cursor.
Fig. 3. Once the image in Figure 2 is analysed using the k-means colour
separation algorithm (for k = 3), the RGB GUI appears showing the original and Simplified Merged Image and the channel sliders for multiband
thresholding.
Start analysis.
Clicking on the Start Analysis button, a
window appears prompting the user to insert the number of
colours to separate. Assuming that in a specific image the reference ranges of the intensity values on each channels are
the same, the algorithm speed is improved by analysing a
representative crop in which all the colours to detect are represented. For example, for H&E staining, the colours to detect
are the blue-violet of the haematoxylin, the red-pink of the
Eosin and, if needed, the white of FSA. The first and the second
steps of the colour separation algorithm described above are
automatically performed on the image crop in order to choose
the optimal colour model.
Colour model dedicated GUIs. At this point a GUI specific for
the colour channel assigned according to the Sv analysis automatically appears and the colour separation algorithm is
applied to the whole original image (shown in the first panel
on the left in Fig. 3). The Simplified Merged Image representing
C 2015 The Authors
C 2015 Royal Microscopical Society, 00, 18
Journal of Microscopy
H IST O O L OGY
Image
area
ROI
Colour Colour
Mean
area
#1
covered
value
area
Ch1
10
11
Mean
value
Chn
Object
counting
12
13
Colour
#2
14
15
Colour
#k
Column 1 to 5: staining-independent parameters; Columns 6 to the end: staining dependent parameters. The number of columns after #5 in the datamatrix
are variable.
C. MAGLIARO ET AL.
Fig. 4. Detection of eosin, haematoxylin, and tissue free areas for porcine
liver (the same image as in Figs. 1 and 2) using HisTOOLogy and the
ImageJ deconvolution plugin. Note that the ImageJ plugin does not return
the simplified image representing the free space regions.
Fig. 5. The final Simplified Merged Image of H&E stained porcine liver as
returned by HisTOOLogy after processing.
C 2015 The Authors
C 2015 Royal Microscopical Society, 00, 18
Journal of Microscopy
H IST O O L OGY
Table 2. Results of dye quantification with HisTOOLogy for porcine liver slices treated with different decellularization protocols
ECA (mm2 )
FSA (mm2 )
NC
481.2 20.03
379.6 95.66
2.2 1.3
1.4 1.6
Fig. 6. Comparison of HisTOOLogy results and standard biochemical assays for pig liver subject to different decellularization treatments: (A) ECA vs. TPC
and (B) NC vs. DNAc.
C. MAGLIARO ET AL.
Conclusion
Histochemical staining procedures use coloured dyes to
highlight a compound. To exploit stain based histological
techniques for quantification of tissue components, we developed HisTOOLogy, an open-source software tool with a
simple, user-friendly interface. As colour quantification requires multiple parameters, we first implemented a method
based on k-means clustering which automatically chooses the
best colour-model for colour separation and clusters image
pixels covered by the same colour. The algorithm also indicates the best channel pair for multiband thresholding should
users wish to improve colour selection using sliders in the
GUI.
Using this method, as well as classical image processing
methods such as cropping and object counting, HisTOOLogy extracts quantitative parameters (e.g. cell density, collagen content) and stores all the measurements in a structured
format (i.e. datamatrix). The data thus stored can be further
used with statistical tools.
In this work, HisTOOLogy was applied to analyse the effects
of different decellularization protocols, comparing the results
with complementary techniques (e.g. DNA and TPC). The results show positive and linear correlation between the two
pairs of variables, and attest to the validity of image based
quantification of stains for assessing tissue composition.
HisTOOLogy can be downloaded at http://www.centro
piaggio.unipi.it/content/histoology-download.
Acknowledgement
This work was partially funded by the European Union Seventh Framework Program (FP7/20072013) under the grant
agreement no. 304961, ReLiver. The A. Tirellas current address is: annalisa.tirella@manchester.ac.uk.
Disclosures
The authors have nothing to disclose.
References
Abr`amoff, M.D., Magalhaes, P.J. & Ram, S. J. (2004) Image processing
with ImageJ. Biophoto. Intl. 11, 3643.
Arthur, David, & Sergei, Vassilvitskii (2007) k-means++: The advantages of careful seeding. Proceedings of the eighteenth annual ACMSIAM symposium on Discrete algorithms. Soc. Ind. Appl. Math.,
10271035.
Belsare, A. & Mushrif, M. (2012) Histopathological image analysis using image processing techniques: an overview. Signal Image Process. 3,
2336.
C 2015 The Authors
C 2015 Royal Microscopical Society, 00, 18
Journal of Microscopy