You are on page 1of 6

Shekhar Singh et al.


Vol No. 6, Issue No. 1, 004 - 009

Breast Cancer detection and Classification using

Neural Network
Shekhar Singh1, Dr P. R. Gupta2

Abstract— Breast cancer detection, classification, scoring and attention in the construction of the expert supporting diagnosis
grading of histopathological images is the standard clinical system has to be paid to the segmentation or Breast Cancer
practice for the diagnosis and prognosis of breast cancer. In a detection stage. The main difficulty of the segmentation
large hospital, a pathologist typically handles number of cancer process is due to the incompleteness and uncertainty of the
detection cases per day. It is, therefore, a very difficult and time- information contained in the histopathological image. The
consuming task. This paper proposes a method for automatic
imperfection of the data acquisition process in the form of
Breast cancer detection, classification, scoring and grading to
noise, chromatic distortion and deformity of histopathological

assist pathologists by providing second opinions and reducing
their workload. A computer-aided Breast cancer detection, material caused by its preparation additionally increases the
classification, scoring and grading system for tissue cell nuclei in problem complexity. The nature of image acquisition (3D to
histological image is introduced and validated as part of the 2D transformation) and the method of scene illumination also
Biopsy Analysis System. Cancer cell nuclei are selectively stained affect the image luminance and sharpness and quality [13]-
with monoclonal antibodies, such as the ant estrogen receptor [15]. Until now many segmentation methods have been
antibodies, which are widely used as part of assessing patient proposed (Carlotto, 1987; Chen et al., 1998; Kass et al., 1987;
prognosis in breast cancer. This paper also presents the
classification of micro cancer object of breast tumor based on
feed forward back propagation Neural Network (FNN). Twenty
six hundred sets of cell nuclei characteristics obtained by
applying image analysis techniques to microscopic slides. The
Otsu, 1979; Su and Chou, 2001; Vincent and Soille, 1991)
but, unfortunately, each of them introduces numerous
additional problems and usually works in practice under given
assumptions and/or needs the end-user’s interaction/co-
dataset consist of eight features which represent the input layer operation (Lee and Street, 2000; Street, 2000; Wolberg et al.,
to the FNN. The FNN will classify the input features into type4, 1993; Zhou and Pycock, 1997) [19]-[21].Since nowadays
type3, type2 and type1 cancer affected objects. The sensitivity, many histopathological projects assume full automation and
specificity and accuracy were found to be equal 99.10%, 95.70% real-time operation with a high degree of efficiency, a method
and 98.60% respectively. It can be concluded that FNN gives fast free of drawbacks of the already known approaches has to be
and accurate classification and it works as promising tool for constructed. Breast cancer detection, classification, scoring
classification of breast cell nuclei.
and grading of histopathological images is the standard
Keywords- color conversion, Adaptive histogram equalization, clinical practice for the diagnosis and prognosis of breast
Adaptive thresholding based segmentation, Watershed cancer. Pathologists perform cancer detection manually under
segmentation a microscope. Their experience directly influences the
accuracy of cancer detection. Variability among pathologists
I. INTRODUCTION has been observed in clinical practice [4]-[6]. In a large
hospital, a pathologist typically handles number of cancer
The number of research works conducted in the area of breast detection cases per day. It is, therefore, a very difficult and

cancer detection, classification, scoring and grading. Many time-consuming task. A Computer system that performs
university centers, research centers and commercial automatic cancer detection can assist the pathologists by
institutions are focused on this issue because of the fact that providing second opinions, reducing their workload and
breast cancer is becoming the most common form of cancer alerting them to cases that require closer attention, allowing
disease of today’s female population. Thus, the construction them to focus on diagnosis and prognosis. This paper presents
of a fully automatic cancer detection system supporting a a multi-segmentation method for breast cancer detection and
human expert has become a challenging and difficult task. neural network for classification. In this paper a group of
Nowadays many camera-based automatic breast cancer modified versions of histopathological image segmentation
detection systems have to face the problem of cells and their methods adopted for fine needle biopsy images are presented,
nuclei separation from the rest of the image content (Lee and that is, the adaptive thresholding based segmentation and
Street, 2000; Pena-Reyes and Sipper, 1998; Setiono, 1996; watershed algorithm. One can also find here a description of
Wolberg et al., 1993) [4]- [15]. Nuclei separation process is denoising and contrast enhancement techniques, pre-
very important because the nucleus of the cell is the place segmentation and fully automatic nuclei localization
where breast cancer malignancy can be observed. Thus, much mechanism used in our approach. Cancer detection and

ISSN: 2230-7818 @ 2011 All rights Reserved. Page 4

Vol No. 6, Issue No. 1, 004 - 009

classification systems, however, although dedicated, do not data is saturated at low and high intensities of Image. This
use the manual approach of counting individual cancer cell increases the contrast of the output image.
nuclei, but rather work with specific areas of stained nuclei
[10]–[13]. This is because the identification of individual C. Contrast limited Adaptive histogram equalization
cancer cell nuclei makes the accumulation of a reasonable process
sample size tedious and time consuming, since it is difficult to
decide automatically or by visual inspection, whether cancer Contrast-limited adaptive histogram equalization (CLAHE)
cell nuclei touch or are partially overlapping [14]-[20]. To enhances the contrast of the grayscale image by transforming
overcome these difficulties methods based on assessment of the values using contrast-limited adaptive histogram
areas of interest (e.g. specifically stained nuclei) have been equalization (CLAHE).CLAHE operates on small regions in
developed, which do not rely on identifying individual nuclei. the image, called tiles, rather than the entire image. Each tile's
Additionally, methods depending on adaptive thresholds to contrast is enhanced, so that the histogram of the output
distinguish between stained tissue (specific staining) and region approximately matches the histogram specified by the
background (nonspecific staining), are based on the 'Distribution' parameter. The neighboring tiles are then
assumption that local variations due to preparation or imaging combined using bilinear interpolation to eliminate artificially
do not significantly influence the measurements [10]–[14]. induced boundaries. The contrast, especially in homogeneous
Furthermore, one of the main objectives of computer-aided areas, can be limited to avoid amplifying any noise that might
biopsy analysis is to minimize some of the variability’s that be present in the image.
occur as a consequence of the manual microscopically

inspection of stained slides. In addition, computer-aided D. Adaptive Thresholding based Segmentation
nuclei analysis also has to be efficient, since pathologists are
unlikely to spend more time on evaluating a specimen than Adaptive Thresholding based Segmentation process operates
that required for the routine manual assessments. As already on small regions in the image, called tiles, rather than the
mentioned, in the manual assessment of biopsy slides one entire image. For each region computes a threshold (level) that
strategy has been to utilize a semi-quantitative scheme to can be used to convert an intensity image to a binary image.
make the assessment more accurate and more objective. To
the best of our knowledge no systems as yet exist which
attempt to aid the expert in the detection, counting, and
classification of individual nuclei using the semi-quantitative
Level is a normalized intensity value that lies in the range [0,
1]. For each region uses Otsu's method, this chooses the
threshold to minimize the intra class variance of the black and
white pixels. Multidimensional arrays are converted
scheme. automatically to 2-D arrays using reshape. The Adaptive
Thresholding function ignores any nonzero imaginary part of
II. BREAST CANCER DETECTION ALGORITHM Image. Adaptive Thresholding returns the effectiveness
metric, EM, as the second output argument. The effectiveness
Our breast cancer detection system adopts a adaptive metric is a value in the range [0 1] that indicates the
histogram equalization and multi-segmentation approach. effectiveness of the thresholding of the input image. The
First, a low-resolution global image of the whole histological lower bound is attainable only by images having a single gray
slide is reconstructed by scaling frames. These frames are level, and the upper bound is attainable only by two-valued
captured at 40× magnification from a patient’s histological images.
sample stained with H&E marker. Typically, down and tiling After finding threshold value converts the grayscale image to
high-resolution micro-image many image frames are a binary image. The output image replaces all pixels in the
generated from a patient’s sample. input image with luminance greater than level with the value 1
(white) and replaces all other pixels with the value 0 (black).
A. Preprocessing & Color Conversion Level specify in the range [0, 1], regardless of the class of the

input image. The Adaptive Thresholding function can be used

In Gray Scale conversion process converts the true color to compute the level argument automatically. Use these level
image RGB to the grayscale intensity image. Gray Scale segments the Biopsy Image. The original image is divided into
conversion process converts RGB images to grayscale by an array of overlapping sub-images. A gray-level distribution
eliminating the hue and saturation information while retaining histogram is produced for each sub-image, and the optimal
the luminance. threshold for that sub-image is calculated based on this
Algorithm: Gray Scale conversion process converts RGB histogram. Since the sub-images overlap, it is then possible to
values to grayscale values by forming a weighted sum of the produce a threshold for each individual pixel by interpolating
R, G, and B components: 0.2989 * R + 0.5870 * G + 0.1140 * the thresholds of the sub-images. An alternative approach is to
B. statistically examine the intensity values of the local
neighborhood of each pixel. The first problem facing us when
B. Adjust Image Intensity value choosing this method is the choice of statistic by which the
In Adjust Image Intensity value process maps the intensity measurement is made. The appropriate statistic may vary from
values in grayscale image to new values in such that 1% of

ISSN: 2230-7818 @ 2011 All rights Reserved. Page 5

Vol No. 6, Issue No. 1, 004 - 009

one image to another, and is largely dependent on the nature image border. The output image is grayscale or binary,
of the image. respectively. The connectivity is 8 for two dimensions.
Removing small objects: Objects removes from a binary
E. Morphological Operation after adaptive thresholding image all connected components (objects) that have fewer
based segmentation than P pixels, producing another binary image. The
connectivity is 8 for two dimensions.
Morphological operation dilation, erosion, bottom-hat Perimeter of objects: this process returns a binary image
filtering, opening, closing, top-hat filtering and filling holes containing only the perimeter pixels of objects in the input
used after adaptive thresholding based segmentation. image. A pixel is part of the perimeter if it is nonzero and it is
connected to at least one zero-valued pixel.
F. Watershed Segmentation Blob labeling: Blob labeling process returns a matrix, of the
same size as Binary image, containing labels for the connected
The watershed segmentation algorithm is inspired by natural objects in Binary image. The elements of matrix are integer
observations, such as a rainy day in the mountains (Gonzalez values greater than or equal to 0. The pixels labeled 0 are the
and Woods, 2002; Pratt, 2001; Russ, 1999). A given image background. The pixels labeled 1 make up one object; the
can be defined as a terrain on which nuclei correspond to pixels labeled 2 make up a second object, and so on.
valleys (upside down the terrain modeled in previous steps).
The terrain is flooded by rainwater and arising puddles start to III. CLASSIFICATION ALGORITHM
turn into basins. When the water from one basin begins to

pour away to another, a separating watershed is created. The In the present work, the neural networks are used for cancer
flooding operation has to be stopped when the water level affected micro object classification purposes. The neural
reaches a given threshold θ. The threshold should preferably networks derive their power due to their massively parallel
be placed somewhere in the middle between the background structure, and an ability to learn from experience. Neural
and a nucleus localization point. network can be used for fairly accurate classification of input
In my approach used the distance transformation function for data into categories, provided they are previously trained to do
so. The accuracy of the classification depends on the
segmentation. The areas where the cells are located are
analyzed by dilating the region maximum of the distance
transforms for each cell nucleus. To visualize the individual
mountain peaks, markers are applied to every peak. The
efficiency of training. The knowledge gained by the learning
experience is stored in the form of connection weights, which
are used to make decisions on fresh input.
watershed function starts at the peaks expands in every Three issues need to be settled in designing an ANN for a
direction until it reaches an edge from another peak or the specific application:
edge of the picture. The watershed division lines are applied • Topology of the network;
to the segmented image to make a clear separation of the • Training algorithm;
touching cell nuclei. The watershed lines superimposed to the • Neuron activation functions.
segmented image and in the other picture, the lines are In our topology, the number of neurons in the input layer is 8
removed leaving segmented cell nuclei only. neurons for the ANN classifier. The output layer was
Segmenting the cells in the manner described above not only determined by the number of classes desired. The outputs are
improves the accuracy of the cell count, but also the accuracy type4, type3, type2, type1 micro object therefore; the output
of the cell nucleus area statistics. In order to make statistics layer consists of 4 neurons. The hidden layer consists of 20
with regards to cell nucleus area even more accurate, we neurons. Before the training process is started, all the weights
remove any cell nuclei that are touching the border, since are initialized to small random numbers. This ensured that the
most of those cells are more likely to be incomplete cells. classifier network was not saturated by large values of the
Again, after the average nucleus size is calculated, excluding weights. In this experiment, the training set was formed by
the incomplete border nuclei, it is possible to estimate the choosing 390 data sets for the testing process.

equivalent total number of border cell nuclei by adding all Cancer object (nucleus) is classified into one of four type
partial cell nucleus areas to be divided by the Average size, if object (nucleus) using the feed forward back propagation
the border cell count is needed in analyzing different types of neural network classifier. After the classification of each
tissue sample. It is also possible to double count the projected cancer objects (nucleus), score and grade are computed. The
overlapping nuclei area (intersection) to improve the accuracy classifier was trained and tested on the images created by one
of the cell nucleus area statistics. of the experts. The main steps of the classification algorithm
are summarized as follows:
G. Morphological Operation after watershed segmentation Extract Features for Each Nucleus (micro object): The
Clearing Border object: following local and global features are extracted: Area,
Diameter, Solidity, Eccentricity, Extent, Perimeter,
Clearing Border object process, suppresses structures that are Roundness, and Compactness.
lighter than their surroundings and that are connected to the

ISSN: 2230-7818 @ 2011 All rights Reserved. Page 6

Vol No. 6, Issue No. 1, 004 - 009

Classify Each Cancer objects: A Cancer objects (nucleus) is

classified into one of four type cancer objects using the feed VI. RESULTS AND DISCUSSION
forward back propagation neural network classifier.
Figure 1 shows a sample high-resolution original image.
Figure 2 illustrates detected cancer cells for nuclear scoring.
IV. TRAINING AND TESTING Nevertheless, scoring is still reliable as long as enough type-2
and type-3 cells are detected. This is analogous to clinical
The proposed network was trained with all 1820 Micro practice in which a pathologist just performs a quick scan of at
objects (tumor) data cases. These 1820 cases are fed to the most 5 image frames to pick up enough type-2 and type-3
FNN with 8 input neurons, one hidden layer of 20 neurons and cells to make an assessment. Table II compares the grading
four outputs neuron. MATLAB software package version 8 is results of the algorithm and a pathologist. It can be seen the
used to implement the software in the current work. When the system’s scores are very close to the pathologist’s scores. The
training process is completed for the training data (1820 system’s scores tend to be slightly lower than the pathologist’s
cases), the last weights of the network were saved to be ready scores. This could be due to the slightly more stringent criteria
for the testing procedure. The time needed to train the training taken by the system. Given these encouraging results, we are
datasets was approximately 1.80 second. confident that an automatic breast cancer detection and
The testing process is done for 390 cases. These 390 cases are grading system can be developed to assist the pathologists by
fed to the proposed network and their output is recorded for providing second opinions and alerting them to cases that
calculation of the sensitivity, specificity and accuracy of require further attention. Nevertheless, more comprehensive

prediction. tests are needed to provide better evaluation of the algorithm’s
performance. The performance of the classification algorithm
V. BIOPSY SCORING AND GRADING was evaluated by computing the percentages of Sensitivity
(SE), Specificity (SP) and Accuracy (AC), the respective
Scoring according to positive cells: The scoring of the cancer definitions are as follows:
detected biopsy image was evaluated by computing the
percentages of positive cancer cells, the respective definitions SE=TP/ (TP+FN)*100 (1)
are as follows:

% of cells positive Score

ES SP=TN/ (TN+TP)*100 (2)

0 0 AC= (TP+TN)/ (TN+TP+FN+FP)*100 (3)

1-25% 1
26-50% 2 Where TP is the number of true positives, TN is the number of
51-75% 3 true negatives; FN is the number of false negatives, and FP is
>=76% 4 the number of false positives. Since it is interesting to estimate
the performance of classifier based on the classification of
Scoring based on the micro objects: benign and malignant breast cell nuclei, the true positives
All the detected cells (micro object) in the high resolution (TP), false positives (FP), true negatives (TN), and false
images are classified into one of the four type’s micro objects. negatives (FN) are defined appropriately as shown below:
Based on the type of object scoring perform.
Overall Grading: The overall grade of the patient is defined in FP: Predicts benign as malignant.
terms of the total. Then, the overall grade is assigned as TP: Predicts malignant as malignant.
follows: FN: Predicts malignant as benign
• Grade I (low grade): G = 3, 4, 5 TN: Predicts benign as begin.

• Grade II (intermediate grade): G = 6, 7, 8

• Grade III (high grade): G = 9, 10, 11 Sensitivity, specificity and accuracy of prediction have been
calculated according to the above formals for all of the testing
TABLE II data (390 micro object cases). Table 1 shows the resulted SE,
GRADE. P: PATHOLOGIST, S: SYSTEM. SP and AC for testing data of the proposed networks.
Patient Image1 P 3
Patient Image1 S 3
Patient Image2 P 2 No Sensitivity Specificity Accuracy
Patient Image2 S 2 of
Patient Image3 P 3 cases
Patient Image3 S 2 390 99.10% 95.70% 98.60%
Patient Image4 P 2
Patient Image4 S 1
Patient Image5 P 1
Patient Image5 S 1

ISSN: 2230-7818 @ 2011 All rights Reserved. Page 7

Vol No. 6, Issue No. 1, 004 - 009


This paper presented a multi-segmentation method for

automatic breast cancer detection and grading of
histopathological images. The individual cells are detected
and classified in the high-resolution image frames. They are
then scored according to the three criteria of the Nottingham
system. Given the encouraging test results, we are confident
that an automatic grading system can be developed to assist
the pathologists by providing second opinions and alerting
them to cases that require further attention. FNN has been
implemented for classification of micro object of breast cancer
tumor. Twenty six hundred sets of cell nuclei characteristics Figure3 Classified Cancer Detected Image
obtained by applying image analysis techniques to Red, Magenta, Blue, Green Objects are type4, type3, type2, type1 Objects.
microscopic slides of H & E stained samples of breast biopsy
have been used in the current work. MATLAB software
package version 8 is used to implement the software in the REFERENCES
current work. These feature vectors which consist of eight
[1] Ajay Nagesh Baravanhalty, Shridar Gancean, Shannan Agner, James

image analysis features each were carried out to generate Peter Monaco, “Coputerised image based detection and grading of
training and testing of the proposed NN. The accuracy is lymphocytic infiltration in HER2 breast cancer histopathology”,IEEE
calculated to evaluate its effectiveness of the proposed Transactions on biomedical engineering vol 57,no-3 march -2010.
network. [2] Jean Roman Dalle, Wee Kheng Leow, Daniel Recoeanis, Adima
Eunice Tatac, Thomas C.Putti, “Automatic breast cancer grading of
The obtained accuracy of the network was 98.60% whereas histopathological images”, 30th annual international IEEE EMBS
the sensitivity and specificity were found to be equal 99.10% conference van couver British Columbia, Canada, august , 2008.
and 95.70% respectively. We conclude that that the proposed
system gives fast and accurate classification of breast tumors.

Ali H. Al-Timemy, Fawzi M. Al-Naima ,Nebras H. Qaeeb,
“Probabilistic Neural Network for Breast Biopsy Classification”,
MASAUM Journal of Computing, Volume 1 Issue 2, September 2009 .
F. Schnorrenberg , C.S. Pattichis , K. Kyriacou , M.Vassiliou , C.N.
Schizas , “Computer–aided classification of breast cancer nuclei” ,
Accepted for publication in the journal Technology & HealthCare,
Elsevier Science B.V.,Amsterdam, Netherlands, 1996.
[5] Spiros Kostopoulos, Dionisis Cavouras, Antonis Daskalakis, Panagiotis
Bougioukos, Pantelis Georgiadis, George C. Kagadis, Ioannis Kalatzis,
Panagiota Ravazoula, George Nikiforidis, “Colour-Texture based
image analysis method for assessing the Hormone Receptors status in
Breast tissue sections”, Proceedings of the 29th Annual International
Conference of the IEEE EMBS Cité International, Lyon, France
August 23-26,2007.
[6] Sokol Petushi, Fernando U Garcia, Marian M Haber,Constantine
Katsinis and Aydin Tozeren, “Large-scale computations on histology
images reveal grade-differentiating parameters for breast cancer”,
BMC Medical Imaging 2006, 6:14 doi:10.1186/1471-2342-6-14.
Figure1 Biopsy Image [7] Tao Shi , David Seligson, Arie S Belldegrun, Aarno Palotie and Steve
Horvath ,” Tumor classification by tissue microarray profiling: random
forest clustering applied to renal cell carcinoma”, Modern
Pathology (2005) 18, 547–557.
[8] Anthony McCabe , Marisa Dolled-Filhart , Robert L. Camp , David L.

Rimm,” Automated Quantitative Analysis (AQUA) of In Situ Protein

Expression, Antibody Concentration,and Prognosis”, Journal of the
National Cancer Institute, Vol. 97, No. 24, December 21, 2005.
[9] James W. Bacus,2 Charles W. Boone, James V. Bacus,Michele Follen,
Gary J. Kelloff, Valery Kagan, and Scott M. Lippman,” Image
Morphometric Nuclear Grading of Intraepithelial Neoplastic Lesions
with Applications to Cancer Chemoprevention Trials”, Cancer
Epidemiology, Biomarkers & Prevention Vol. 8, 1087–1094,
December 1999
[10] Hussain Fatakdawala, Jun Xu, Ajay Basavanhally, Gyan Bhanot,
Shridar Ganesan, Michael Feldman, John E. Tomaszewski, and Anant
Madabhushi,” Expectation Maximization driven Geodesic Active
Contour with Overlap Resolution (EMaGACOR): Application to
Figure2 Cancer Detected Image Lymphocyte Segmentation on Breast Cancer Histopathology”, IEEE

ISSN: 2230-7818 @ 2011 All rights Reserved. Page 8

Vol No. 6, Issue No. 1, 004 - 009

[11] Scarff RW, Torloni H: Histological typing of breast tumors.

International histological classification of tumors. World Health
Organization 1968, 2(2):13-20.
[12] Elston CW, Ellis IO: Pathological prognostic factors in breast cancer.
The value of histological grade in breast cancer: experience from a
large study with long-term follow-up. Histopathology 1991, 19:403-
[13] C. Demir and B.Yener, “Automated cancer diagnosis based on
histopathological images: a systematic survey,” Rensselaer Polytechnic
Institute, Tech. Rep., 2005.
[14] A. Tutac, D. Racoceanu, T. Putti, W. Xong, W.-K. Leow, and V.
Cretu, “Knowledge-guided semantic indexing of breast cancer
histopathology images,” in Proc. Int. Conf. on Biomedical Engineering
and Informatics, 2008.
[15] S. Petushi, F. U. Garcia, M. M. Haber, C. Katsinis, and A. Tozeren,
“Large-scale computations on histology images reveal
gradedifferentiating parameters for breast cancer,” BMC Medical
Imaging, vol. 6, no. 14, 2006.
[16] S. Doyle, M. Hwang, M. Feldman, and J. Tomaszeweski, “Automated
grading of prostate cancer using architectural and textural image
features,” in Proc. of 4th IEEE Int. Symp. on Biomedical Imaging,
2007, pp. 1284 – 1287.
[17] H. Soltanian-Zadeh and K. Jafari-Khouzani, “Multiwavelet grading of
prostate pathological images,” IEEE Trans. on Biomedical

Engineering, vol. 50, pp. 697–704, 2003.
[18] A. Nedzved, S. Ablameyko, and I.Pitas, “Morphological segmentation
of histology cell images,” in Proc. Int. Conf. Pattern Recognition,
2000, pp. 1500–1503.
[19] H. Jeong, T.-Y. Kim, H.-G. Hwang, and H.-J. Choi, “Comparison of
thresholding methods for breast tumor cell segmentation,” in Proc. of
7th Int. Workshop on Enterprise networking and Computing in
Healthcare Industry, 2005, pp. 392–395.

F. Schnorrenberg, “Comparison of manual and computer-aided breast
cancer biopsy grading,” in Proc. of IEEE EMBS, 1996.
M. E. Adawi, Z. Shehab, H. Keshk, and M. E. Shourbagy, “A fast
algorithm for segmentation of microscopic cell images,” in Proc. Of
4th Int. conf. on Information & Communications Technology, 2006.
[22] Latson L, Sebek B, Powell KA: Automated cell nuclear segmentation
in color images of hematoxylin and eosin-stained breast biopsy.
Analytical and Quantitative Cytology and Histology 2003, 25(6):321-
[23] Schnorrenberg F, Pattichis C, Kyriacou K,Schizas C: Computer-
Aided Detection of Breast Cancer Nuclei. IEEE Trans on Information
Tech in Biomedicine 1997, 1(2):128-140.
[24] Schupp S, Elmoataz A, Herlin P, Bloyet D: Mathematical
morphologicsegmentation deticated to quantitative
immunohistochemistry. Analytical and Quantitative Cytology and
Histology 2001, 23(4):257-267.
[25] Bacus JW, Boone Charles W, Bacus James V, Michele Follen, Kelloff
Gary J, Valery Kagan, Scott Lippman M: Image Morphometric
Nuclear Grading of Intraepithelial Neoplastic Lesions with
Applications to Cancer Chemoprevention Trials. Cancer Epidemiol
Biomarkers Prev 1999, 8:1087-1094.
[26] Hoque A, Lippman Scott M, Boiko Iouri V, Atkinson Edward N, Nour

Sneige, Aysegul Sahin, Weber Diane M, Seymon Risin, Lagios

Michael D, Roland Schwarting, Colburn William J, Kapil Dhingra,
Michele Follen, Kelloff Gary J, Boone Charles W, Hittelman Walter N:
Quantitative Nuclear Morphometry by Image Analysis for Prediction
of Recurrence of Ductal Carcinoma in Situ of the Breast. Cancer
Epidemiol Biomarkers Prev 2001, 10:249-259.
[27] Wolberg WH, Street WN, Heisey DM, Mangasarian OL: Computer-
derived nuclear grade and breast cancer prognosis. Analytical
Quantitative Cytology Histology 1995, 17:257-64.
[28] Petushi S, Katsinis C, Coward C, Garcia F, Tozeren A: Automated
identification of microstructures on histology slides. IEEE International
Symposium on Biomedical Imaging: Macro to Nano 2004, 1:424-427

ISSN: 2230-7818 @ 2011 All rights Reserved. Page 9