Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
ABSTRACT
LIST OF FIGURES
LIST OF ABBREVATIONS
CHAPTER 1 : INTRODUCTION
1.1 GENERAL
1.1.1 THE IMAGE PROCESSING SYSTEM
2.1 INTRODUCTION
2.
3. 3.2 EXPRESSIONS
3.2.1 VARIABLES
3.2.2 NUMBERS
3.2.3 OPERATORS
3.2.4 FUNCTIONS
THE ELEMENTS
3.3.3 CONCATENATION
3.4 GRAPHICS
MANIPULATIONS
VECTORS
3.8.3 PROBLEMS
3.9.1 2 D-ROUTINES
3.9.2 3D-ROUTINES
CHAPTER 4 : IMPLEMENTATION
4. 4.1 GENERAL
4.3 SNAPSHOTS
5. 5.1 CONCLUSION
5.2 REFERENCES
LIST OF FIGURES
INTRODUCTION
1.1 GENERAL
The term digital image refers to processing of a two dimensional picture by a digital computer.
In a broader context, it implies digital processing of any two dimensional data. A digital image is
an array of real or complex numbers represented by a finite number of bits. An image given in the
form of a transparency, slide, photograph or an X-ray is first digitized and stored as a matrix of
binary digits in computer memory. This digitized image can then be processed and/or displayed
on a high-resolution television monitor. For display, the image is stored in a rapid-access buffer
memory, which refreshes the monitor at a rate of 25 frames per second to produce a visually
continuous display.
Hard Copy
Device
Display
A digitizer converts an image into a numerical representation suitable for input into a
digital computer. Some common digitizers are
1. Microdensitometer
2. Flying spot scanner
3. Image dissector
4. Videocon camera
5. Photosensitive solid- state arrays.
IMAGE PROCESSOR:
Knowledge Result
Preprocessing Recognition &
interpretation
Base
As detailed in the diagram, the first step in the process is image acquisition by an imaging
sensor in conjunction with a digitizer to digitize the image. The next step is the preprocessing step
where the image is improved being fed as an input to the other processes. Preprocessing typically
deals with enhancing, removing noise, isolating regions, etc. Segmentation partitions an image
into its constituent parts or objects. The output of segmentation is usually raw pixel data, which
consists of either the boundary of the region or the pixels in the region themselves. Representation
is the process of transforming the raw pixel data into a form useful for subsequent processing by
the computer. Description deals with extracting features that are basic in differentiating one class
of objects from another. Recognition assigns a label to an object based on the information provided
by its descriptors. Interpretation involves assigning meaning to an ensemble of recognized objects.
The knowledge about a problem domain is incorporated into the knowledge base. The knowledge
base guides the operation of each processing module and also controls the interaction between the
modules. Not all modules need be necessarily present for a specific function. The composition of
the image processing system depends on its application. The frame rate of the image processor is
normally around 25 frames per second.
DIGITAL COMPUTER:
MASS STORAGE:
The secondary storage devices normally used are floppy disks, CD ROMs etc.
The hard copy device is used to produce a permanent copy of the image and for the storage
of the software involved.
OPERATOR CONSOLE:
The operator console consists of equipment and arrangements for verification of intermediate
results and for alterations in the software as and when require. The operator is also capable of
checking for any resulting errors and for the entry of requisite data.
1.1.2 IMAGE PROCESSING FUNDAMENTAL:
Digital image processing refers processing of the image in digital form. Modern cameras
may directly take the image in digital form but generally images are originated in optical form.
They are captured by video cameras and digitalized. The digitalization process includes sampling,
quantization. Then these images are processed by the five fundamental processes, at least any one
of them, not necessarily all of them.
Image Enhancement
Image Restoration
IP
Image Analysis
Image Compression
Image Synthesis
IMAGE ENHANCEMENT:
Image enhancement operations improve the qualities of an image like improving the
image’s contrast and brightness characteristics, reducing its noise content, or sharpen the details.
This just enhances the image and reveals the same information in more understandable image. It
does not add any information to it.
IMAGE RESTORATION:
Image restoration like enhancement improves the qualities of image but all the operations
are mainly based on known, measured, or degradations of the original image. Image restorations
are used to restore images with problems such as geometric distortion, improper focus, repetitive
noise and camera motion. It is used to correct images for known degradations.
IMAGE ANALYSIS:
IMAGE COMPRESSION:
Image compression and decompression reduce the data content necessary to describe the image.
Most of the images contain lot of redundant information, compression removes all the
redundancies. Because of the compression the size is reduced, so efficiently stored or transported.
The compressed image is decompressed when displayed. Lossless compression preserves the exact
data in the original image, but Lossy compression does not represent the original image but provide
excellent compression.
IMAGE SYNTHESIS:
Image synthesis operations create images from other images or non-image data. Image
synthesis operations generally create images that are either physically impossible or impractical to
acquire.
APPLICATIONS OF DIGITAL IMAGE PROCESSING:
Digital image processing has a broad spectrum of applications, such as remote sensing via
satellites and other space crafts, image transmission and storage for business applications, medical
processing, radar, sonar and acoustic image processing, robotics and automated inspection of
industrial parts.
MEDICAL APPLICATIONS:
SATELLITE IMAGING:
COMMUNICATION:
Radar and sonar images are used for detection and recognition of various types of targets or in
guidance and maneuvering of aircraft or missile systems.
DOCUMENT PROCESSING:
It is used in scanning, and transmission for converting paper documents to a digital image form,
compressing the image, and storing it on magnetic tape. It is also used in document reading for
automatically detecting and recognizing printed characteristics.
DEFENSE/INTELLIGENCE:
Accuracy depends upon the accuracy of image binarizations as the topographic surfaces for
segmentation are built from distance maps of binary image.
Detection is not automatic.
We propose a novel method based on an image processing algorithm to analyze large sets of
digital micrographs. The nucleus detector design is based on a Laplacian of Gaussian filter.
We use the leave-one-out cross validation method for estimating the generalization error,
which is then used to choose the model and parameters of the proposed nucleus detector with
both fluorescent and dye stained images. We also evaluate the performance of a nucleus
detector by comparing the results with manual counts. When our nucleus detector is applied to
previously unanalyzed images of feline retina, it correctly counts nuclei within the outer
nuclear layer (ONL) with an average error of 3.67% ranging from 0 to 6.07%, and nuclei within
the inner nuclear layer (INL) with an average error of 8.55% ranging from 0 to 13.76%. Our
approach accurately identifies the location of cell bodies. Even though we have a relatively
large error in the INL due to the large intra-observer variation, both manual counting and
nucleus detector result in the same conclusion. This is the first time that cell death in the INL
in response to retinal detachment is analyzed quantitatively. We also test the proposed tool
with various images and show that it is applicable to a wide range of image types with nuclei
varying in size and staining intensity.
An automated technique for nuclear atypia score (NAS) calculation is proposed. The proposed
technique uses sophisticated digital image analysis and machine-learning methods to measure
the NAS for haematoxylin and eosin stained images. The proposed technique first segments
all nuclei regions. A set of morphology and texture features is extracted from presegmented
nuclei regions. The histogram of each feature is then calculated to characterize the statistical
information of the nuclei. Finally, a support vector machine classifier is applied to classify a
high-power field image into different nuclear atypia classes. A set of 1188 digital images was
analysed in the experiment. We successfully differentiated the high-power field image with
NAS1 versus non-NAS1, NAS2 versus non-NAS2 and NAS3 versus non-NAS3, with area
under receiver-operating characteristic curve of 0.90, 0.86 and 0.87, respectively. In three
classes evaluation, the average classification accuracy was 78.79%. We found that texture-
based feature provides best performance for the classification.
Extraction of cell nuclei from hematoxylin and eosin (H&E)-stained histopathological images
is an essential preprocessing step in computerized image analysis for disease detection,
diagnosis, and prognosis. We present an automated cell nuclei segmentation approach that
works with H&E-stained images. A color deconvolution algorithm was first applied to the
image to get the hematoxylin channel. Using a morphological operation and thresholding
technique on the hematoxylin channel image, candidate target nuclei and background regions
were detected, which were then used as markers for a marker-controlled watershed transform
segmentation algorithm. Moreover, postprocessing was conducted to split the touching nuclei.
For each segmented region from the previous steps, the regional maximum value positions
were identified as potential nuclei centers. These maximum values were further grouped into
k -clusters, and the locations within each cluster were connected with the minimum spanning
tree technique. Then, these connected positions were utilized as new markers for a watershed
segmentation approach. The final number of nuclei at each region was determined by
minimizing an objective function that iterated all of the possible k -values. The proposed
method was applied to the pathological images of the tumor tissues from The Cancer Genome
Atlas study. Experimental results show that the proposed method can lead to promising results
in terms of segmentation accuracy and separation of touching nuclei.
[4] S. Petushi et al., “Large-scale computations on histology images reveal grade-
differentiating parameters for breast cancer,” BMC medical imaging, vol. 6, no. 1, p. 1,
2006.
Image processing of histology slide images was used to detect and identify adipose tissue,
extracellular matrix, morphologically distinct cell nuclei types, and the tubular architecture.
The texture parameters derived from image analysis were then applied to classify images in a
supervised classification scheme using histologic grade of a testing set as guidance. The
histologic grade assigned by pathologists to invasive breast carcinoma images strongly
correlated with both the presence and extent of cell nuclei with dispersed chromatin and the
architecture, specifically the extent of presence of tubular cross sections. The two parameters
that differentiated tumor grade found in this study were (1) the number density of cell nuclei
with dispersed chromatin and (2) the number density of tubular cross sections identified
through image processing as white blobs that were surrounded by a continuous string of cell
nuclei. Classification based on subdivisions of a whole slide image containing a high
concentration of cancer cell nuclei consistently agreed with the grade classification of the entire
slide.
Input Image
Nuclear Seed
Detection
Nuclear Boundarty
Delieation
Segmented Output
PROJECT DESCRIPTION
2.1 INTRODUCTION
Evaluation of cell nuclei plays an important role in the histopathological examination and analysis
of digitized slides. The parameters like cell size, shape, contours and presence or absence of
nucleoli are generally used by pathologists for various cancer diagnosis. The abnormal cell
distributions and morphological changes often indicate the malignancy of the lesion [1], [2].
However, manual counting and segmentation of cell nuclei is a very tedious work, and also prone
to inter- or intra-observer variability due to poor contrast and clumping of nuclei. To reduce the
workload and obtain more objective results, many efforts have been devoted to develop automatic
nuclear segmentation techniques. Threshold-based techniques have been widely used for nuclear
segmentation in histological images. These works are related to segmentation of cell nuclei in the
digitized neuroblastoma slides [3], breast histopathological images [4] and skin histopathological
images.
In order to address this issue, a number of techniques for cellular or nuclear clumps segmentation
have been reported. These techniques mainly includes two categories: concavity analysis based
techniques [6]–[12], and “seed” (or “marker”) based techniques [13]–[18]. For concavity analysis
based techniques, the first step is to identify valid concave points for splitting. Several techniques
exist for identifying concave points on the contour. For example, the concave points have been
detected based on computing the cross product of vectors defined by three consecutive points [6],
or minimizing a cost function to find the bottleneck positions on the contour [7]. With detected
concave points, the next step is to connect them and split cellular or nuclear clumps into isolated
ones. This step can be achieved iteratively by connecting the shortest path between concave points
until all split regions are considered to be isolated [11]. Although concavity analysis based
techniques have been reported to provide a good performance, the performance
is overly sensitive to image binarizations, and the predefined thresholds (usually several
thresholds) that are used for finding valid concave points [7], [10].
For “seed” ( or “marker”) based techniques, the first step is to identify “seeds” for nuclear regions.
These detected “seeds” are usually close to nuclear centers, and used as the starting points for
active contours or watershed segmentation. The segmentation performance depends critically on
the accuracy and reliability of initial seed points. Because of the advantages of simplicity, speed
and absence of adjustable parameters, several marked watershed algorithms have been developed
for cellular or nuclear segmentation. For example, the markers of watershed algorithm are detected
by condition erosion [15], Hminima transform [16], [17], improved voting algorithm [14] and
radial symmetry transform [18]. These watershed based segmentations are heavily replied on the
accuracy of image binarizations as the topographic surfaces for segmentation are built from
distance maps of binary images. In order to get more accurate segmentations, Al-Kofahi et al. [19]
proposed a multi scale Laplacian of Gaussian (mLoG) filter based technique, which detects nuclei
by mLoG filters and segments nuclei based on size-constrained clustering with further refinement
by a graph cuts based algorithm. Qi et al. [13] proposed to detect cell seeds in breast
histopathological images by a single pass voting algorithm and delineate cell contours by a
repulsive level set model [20]. Xing et al. [21] proposed a cell detection and segmentation
technique for evaluating Ki- 67 proliferation index in gauge neuroendocrine tumor images.
Recently a few techniques based on active shape models and deep learning (DL) have been
proposed for nuclear segmentation. Plissiti et al. [23] proposed a technique that segments
overlapping cell nuclei by driving the deformable model based on physical principals, but this
technique only works automatically for the image with a known number of cell nuclei. Lu et al.
[24] proposed a technique to perform cervical cell segmentation by joint optimization of multiple
level set functions. This technique has a high computational complexity and its performance is
critically replied on initial segmentations for level set functions. Janowczyk et al. [25] proposed a
resolution adaptive deep hierarchical (RADHical) learning scheme for nuclear segmentation in
histological images. The RADHical employs the AlexNet network [26] in a multi-resolution
framework, which classifies most of image pixels at low resolutions and only processes uncertain
regions at high resolutions. The RADHical provides a similar performance with the naive DL
model [27] that processes the high-resolution images directly, but it achieves about 6 times speed
improvement. Although the DL methods are very powerful to learn image features, they are
generally quite time consuming to train and apply.
Besides the above mentioned methods, graph theories are also used for medical tissue
segmentations. Li et al. [28] proposed a segmentation method for volumetric images (e.g.,
pulmonary CT images) by computing a minimum s-t cut in a derived arc-weighted directed graph.
Zhang et al. [29] proposed a graph search based method for intestinal gland segmentation in
histological images. The method uses a graph search algorithm to generate the gland boundary
probability map. The watershed algorithm is then applied on the boundary probability map to
perform segmentations. The graph search methods have the potential to be used for nuclear
segmentations, but it is difficult to determine an optimal graph search radius due to nuclear size
variations and existence of clustered nuclei.
2.2 RELATED WORK
MATLAB
INTRODUCTION TO MATLAB
3.1 What is Matlab?
Matlab is an interactive system whose basic data element is an array that does not require
dimensioning. This allows the user to solve many technical computing problems, especially
those with matrix and vector operations, in less time than it would take to write a program
in a scalar non-interactive language such as C or FORTRAN.
3.2. EXPRESSIONS
Like most other programming languages, Matlab provides mathematical expressions, but unlike
most programming languages, these expressions involve entire matrices. The building blocks of
expressions are
Variables
Numbers
Operators
Functions
3.2.1 Variables
Matlab does not require any type declarations or dimension statements. When a new
variable name is introduced, it automatically creates the variable and allocates the appropriate
amount of memory. If the variable already exists, Matlab changes its contents and, if necessary,
allocates new storage.
For example
>> books = 10
Creates a 1-by-1 matrix named books and stores the value 10 in its single element. In the
expression above, >> constitutes the Matlab prompt, where the commands can be entered.
Variable names consist of a string, which start with a letter, followed by any number of
letters, digits, or underscores. Matlab is case sensitive; it distinguishes between uppercase and
lowercase letters. A and a are not the same variable. To view the matrix assigned to any variable,
simply enter the variable name.
3.2.2 Numbers
Matlab uses the conventional decimal notation. A decimal point and a leading plus or minus
sign is optional. Scientific notation uses the letter e to specify a power-of-ten scale factor.
Imaginary numbers use either i or j as a suffix. Some examples of legal numbers are:
7 -55 0.0041 9.657838 6.10220e-10 7.03352e21 2i -2.71828j 2e3i 2.5+1.7j.
3.2.3 Operators
Expressions use familiar arithmetic operators and precedence rules. Some examples are:
+ Addition
- Subtraction
* Multiplication
/ Division
’ Complex conjugate transpose
( ) Brackets to specify the evaluation order.
3.2.4 Functions
Matlab provides a large number of standard elementary mathematical functions, including
sin, sqrt, exp, and abs. Taking the square root or logarithm of a negative number does not lead to
an error; the appropriate complex result is produced automatically. Matlab also provides a lot of
advanced mathematical functions, including Bessel and Gamma functions. Most of these functions
accept complex arguments. For a list of the elementary mathematical functions, type
>> help elfun
Some of the functions, like sqrt and sin are built-in. They are a fixed part of the Matlab core so
they are very efficient. The drawback is that the computational details are not readily accessible.
Other functions, like gamma and sinh, are implemented in so called M-files. You can see the code
and even modify it if you want.
3.3.3 Concatenation
Concatenation is the process of joining small matrices to make bigger ones. In fact, the first matrix
A was created by concatenating its individual elements. The pair of square brackets, [ ], is the
concatenation operator. For an example, start with the 3-by-3 matrix A, and form
>> F = [A A+10; A*2 A*4].
The result is a 6-by-6 matrix, obtained by joining the four sub matrices.
F= 1 2 3 11 12 13
8 6 4 18 16 14
3 6 9 13 16 19
2 4 6 4 8 12
16 12 8 32 24 16
6 12 19 12 24 38
Examples
>> d = [1 2 3; 4 5 6]
d= 123
456
>> e = [2 2 2; 3 3 3]
e= 222
333
>> f = d+e adds d to e on an element-by-element basis
f= 345
789
>> g = 2*d-e multiplies d by two and subtracts e from the result
g= 024
579
Element-by-element multiplication and division work similarly, but the notation is slightly
different:
>> h = d.*e
h= 2 4 6
12 15 18
The element-by-element multiplication uses the dot multiplication symbol.*, the element-by-
element array division uses either. / or.\
>> d./e
ans = 0.500 1.000 1.500
1.333 1.666 2.000
>> e.\d
ans = 0.500 1.000 1.500
1.333 1.666 2.000
In both cases, the elements of the array in front of the slash are divided by the elements of the array
behind the slash. To compute a matrix multiplication only the asterisk * must be used, e.g.
>> C = A * B
3.4 GRAPHICS
Matlab offers extensive facilities for displaying vectors and matrices as graphs, as well as
annotating and printing these graphs. This section describes some of the most important graphics
functions and gives some examples of some typical applications.
0.5
0
y
-0.5
-1
0 1 2 3 4 5 6 7
x
y = sin(x)
1
0.5
0
y
-0.5
-1
0 1 2 3 4 5 6 7
x
0.5
cos(t)
-1
-3 -2 -1 0 1 2 3
- t
To take a closer look at an interesting part of a plot, the zoom command can be used. Afterwards
it is possible to zoom by marking this part with the mouse. The grid command is used to turn a
grid on and off.
vector e.
The fprintf() command used above creates a formatted string. The principle construction rule of
this command is well known from standard programming languages like C, FORTRAN etc.
t=fprintf(’text %format string(s)\n’, variable(s));
where t is a string variable (which may be omitted), the text string for output and the format
string(s) and perhaps a line break must be enclosed in apostrophes, the format string(s) always
start with a % character followed by the total number of characters to be printed, the . separator
which separates the integer values from the fractions and the number of characters to be printed
after the . separator. The character f, which follows in the format string defines the floating point
format. \n carries out one line break. Separated by a comma, the variable(s) follow(s) in the same
order and with as many variables (or expressions) as format strings are defined.
Greek characters within the text string of fprintf() are generated by the LATEX-notation starting
with a \followed by the name of the Greek character; like for example \Omega, \omega or \Gamma,
\gamma for the corresponding upper- and lower case characters.
%square the elements of vector e or build any power of the
%vector elements of e.
f=e.*e
f=e.^2
subplot(212);
plot(f);
title(’Plot of vector f’);
grid on;
e.*e and e.ˆ 2 show the same results but e.ˆ x is more universal because x may be any power of the
elements of vector e and x can be any real number.
The plot of natural functions requires usually a much more narrow spacing of the horizontal
axis. If we use for example our original vector a with eleven elements and plot sin(a), then we get
a polygon which does not very much resemble the sine function.
If we clear all vectors, redefine vector a with 100 intervals between 0 and 2_ and plot it we get a
much better resolution.
The notation plot(a,sin(a)) plots sin(a) versus vector a, which is now a properly scaled function.
“set(gca,’Xlim’,[0,2*pi]);” sets the entire horizontal range of the plot.
“set(gca,’Ylim’,[-2,2]);” would set the vertical range.
Alternatively, we could use the axis() command to fix the vertical and horizontal scale of the plot,
like for example:
axis([0,2*pi,-2,2]) i.e. axis([xmin,xmax,ymin,ymax])
Matlab provides all basic mathematical functions like all trigonometric and exponential functions
and even more specific functions like Bessel, Gamma, Beta functions etc. Type “help elfun” or
“help specfun” to the command window to find the notation for all the build in functions.
plot(y1); Plots the first output vector versus its indices and connects
subsequent points with lines.
plot1=plot(t,y1); Plots the first output vector versus input vector. Since we
want to change the appearance of this plot later, we need to
define the handle ’plot1’.
hold on; The active figure will not be overwritten by the next plot
command.
grid on; Activates grid of both x- and y-axis (syntax like ’hold’).
leg=legend([plot1,plot2],’A1=1 Displays a legend for the curves with handles ’plot1’ and
V, f1=200 Hz’,’A2=1.5 V,
f2=50 Hz’); ’plot2’.
sub1=subplot(211); ‘subplot (nr, nc, counter)’ plots a subfigure within the active
figure. The figure is divided into ’nr’ rows and ’nc’ columns
and the last parameters determines the position of the
subfigure (the numbering is row-wise and
1≤counter≤Nr*Nc).
set(plot1,’linewidth’,2);
grid on;
axis([Tst,Te,-1.7,1.7]);
leg2 = legend([plot2],’A2=2 V,
3.8.3 PROBLEMS
3.8.3.1 Sampling of ’continuous’ waveform
Given is the Matlab routine ‘[s]=waveform(t)’ which generates the periodical signal s(t) with
period T = 20 ms. It is sampled in the time interval t=[0,65]ms with four different sampling
frequencies f1 = 100Hz, f2 = 250Hz, f3 = 500Hz and f4 = 1000Hz. Display in four subfigures the
’continuous’ waveform and the corresponding sampled waveform, respectively. Use for the
sampled waveform the command ’stem’. Moreover, consider following properties of the figures:
Grid: on
3.8.3.2 Simple 3D-Plots
Plot a two-dimensional Gaussian probability density function fx,y(x, y) with zero mean and
variance 1 within the range x [−5, 5], y [−5, 5]:
1 2 x 2 y 2
1
f x , y x, y e (4)
2
a=rand(1,100) Rand (x,y) generates a matrix with x rows and y columns, whose
elements are uniformly distributed within the range [0,1].
hist(a); hist(a) plots the histogram. The value range of a is divided into 10
intervals. The number of elements in these intervals is plotted.
a=rand(1,1e5);
hist(a);
hist(a,100); hist(a,N) plots the histogram. The value range of a is divided into
a=rand(1,1e5);
hist(a,100); a −→ a is RV [0, 1]
a=rand(1,1e5)+0.5;
a=2*rand(1,1e5);
hist(a,100); a −→ a is RV [0, 2]
a=floor(20*rand(1,1e5)+1);
hist(a,100) Now you can see that values can only take integer values
Gaussian distribution
In the following, continuous, Gaussian distributed random numbers are generated and their
histograms plotted.
hist(b,100) Histogram
x=sqrt(5)*randn(1,1e5)+1; Zero Mean and variance 1 is changed into mean 1 and var 5.
Therefore, x is Gaussian distributed now.
hist(x,100) Histogram
CHAPTER 4
IMPLEMENTATION
4.1 GENERAL
5.1 CONCLUSION
This paper presents an automatic technique for nuclear segmentation in skin histopathological
images. The technique relies on a bank of gLoG kernels for nuclear seeds detection, and thereafter
on a mRLS with DP method for nuclear contours delineation. The proposed mRLS based technique
first identifies several candidate boundary contours for each nucleus. The gradient, intensity and
shape information are then integrated to determine the optimal boundary from candidate boundary
contours. The Dice coefficient measure is finally applied to resolve severely overlapped nuclei.
Experiments have been thoroughly performed on two datasets of skin microscopic images, which
indicates that the proposed technique is superior to several existing techniques of nuclear
segmentation. In addition, the proposed technique has the advantage of recognizing occluded
nuclear boundaries for overlapped nuclei. Compared with level sets and active contours, the
proposed technique is very efficient and suitable for parallel implementations, which can be used
for processing the image with a large number of cell nuclei. In future we will analyse
morphological and textural features of cell nuclei based on this work, which will help in skin
cancer diagnosis.
5.2 REFERENCES:
[1] J. Byun et al., “Automated tool for the detection of cell nuclei in digital microscopic images:
application to retinal images,” Molecular Vision, vol. 12, pp. 949–960, 2006.
[2] C. Lu et al., “Automated image analysis of nuclear atypia in high-power field histopathological
image,” Journal of microscopy, vol. 258, no. 3, pp. 233–240, 2015.
[3] M. N. Gurcan et al., “Image analysis for neuroblastoma classification: Segmentation of cell
nuclei,” in Proceedings of IEEE International Conference on Engineering in Medicine and
Biology Society (EMBC), 2006, pp. 4844–4847.
[4] S. Petushi et al., “Large-scale computations on histology images reveal grade-differentiating
parameters for breast cancer,” BMC medical imaging, vol. 6, no. 1, p. 1, 2006.
[5] C. Lu et al., “A robust automatic nuclei segmentation technique for quantitative
histopathological image analysis,” Analytical and Quantitative Cytology and Histology, vol. 34,
pp. 296–308, 2012.
[6] H. Fatakdawala et al., “Expectation–maximization-driven geodesic active contour with overlap
resolution (emagacor): Application to lymphocyte segmentation on breast cancer histopathology,”
IEEE Transactions on Biomedical Engineering, vol. 57, no. 7, pp. 1676–1689, 2010.
[7] H. Wang et al., “Clump splitting via bottleneck detection and shape classification,” Pattern
Recognition, vol. 45, no. 7, pp. 2780–2787, 2012.
[8] Q. Wen et al., “A delaunay triangulation approach for segmenting clumps of nuclei,” in
Proceedings of IEEE International Symposium on Biomedical Imaging: From Nano to Macro
(ISBI), 2009, pp. 9–12.
[9] O. Schmitt and S. Reetz, “On the decomposition of cell clusters,” Journal of Mathematical
Imaging and Vision, vol. 33, no. 1, pp. 85–103, 2009.
[10] X. Bai et al., “Splitting touching cells based on concave points and ellipse fitting,” Pattern
recognition, vol. 42, no. 11, pp. 2434–2446, 2009.
[11] H. Kong et al., “Partitioning histopathological images: an integrated framework for
supervised color-texture segmentation and cell splitting,” IEEE Transactions on Medical Imaging,
vol. 30, no. 9, pp. 1661–1677,2011.
[12] L. Zhang et al., “Segmentation of cytoplasm and nuclei of abnormal cells in cervical cytology
using global and local graph cuts,” Computerized Medical Imaging and Graphics, vol. 38, no. 5,
pp. 369–380, 2014.
[13] X. Qi et al., “Robust segmentation of overlapping cells in histopathology specimens using
parallel seed detection and repulsive level set,” IEEE Transactions on Biomedical Engineering,
vol. 59, no. 3, pp. 754–765, 2012.
[14] H. Xu et al., “An efficient technique for nuclei segmentation based on ellipse descriptor
analysis and improved seed detection algorithm,” IEEE Journal of Biomedical and Health
Informatics, vol. 18, no. 5, pp. 1729–1741, 2014.
[15] X. Yang et al., “Nuclei segmentation using marker-controlled watershed, tracking using
mean-shift, and kalman filter in time-lapse microscopy,” IEEE Transactions on Circuits and
Systems, vol. 53, no. 11, pp. 2405–2414, 2006.