International Journal of Computer Applications (0975 – 8887)

Volume 76– No.9, August 2013

Tibia Bone Segmentation in X-ray Images - A

Comparative Analysis

Nathanael .E. Jacob M.V. Wyawahare

Department of E & Tc. Department of E & Tc.
VIT, Pune. Country- India VIT, Pune. Country- India

ABSTRACT time, sensitivity, specificity, Jaccard and Dice coefficients as

Segmentation techniques in the medical field are used to performance metrics.
segment anatomical structures or other region of interest from
medical images obtained from different modalities. This paper
2. RELATED WORK
deals with segmentation techniques like manual thresholding, Segmentation of bones plays an important role not only in
Otsu thresholding, watershed, traditional active contours and fracture detection [5] but also surgery, quantitative analysis
growcut in X-ray modality, for segmenting the tibia bone. and planning for surgery. Segmentation has been researched
This paper analyzes the performance of these algorithms on a in different modalities like CT, MRI and X-ray. In case of X-
database of 48 clinical X-ray images. The images have been ray images the segmentation is quite challenging. This is
obtained from different X-ray machines and vary in their because of bone boundaries being less clear in X-ray images
resolution and dimensions. The performance of the algorithms as compared to images in CT or MRI [6].
have been measured and validated empirically. Region based algorithm involving region growing, region
merging and region labelling has been applied by [7] Manos
General Terms et al. to segment hand and wrist bones. [8] El-Feghi et al. used
Medical imaging, machine vision, biomedical image analysis. a fuzzy set algorithm to segment bone in lateral skull x-ray
images. The algorithm however suffers from the problem of
Keywords disjoined segmented regions. [9] Vinhais et al. segmented the
X-ray, tibia bone, segmentation, growcut, validation. rib cage in posterior-anterior chest X-ray images using a
deformable prior model, which is deformed using a
1. INTRODUCTION deformation grid. The segmented output defines the lung
A common ailment that affects the tibia bone is fractures and region, which is used in Computer aided diagnosis system. On
account for approximately 20% occupancy in hospitals [1]. the other hand [10] Zhanjun Yue et al., rib finding algorithm
The fractures that occur in tibia bones are varied and pose uses Hough transform to approximate and finally localize the
difficulties to a doctor in finding and assessing them rib bones using active contour model in chest radiographs.
accurately. Misdiagnosis of fractures can occur in clinical
setting due to factors such as a tired radiologist, huge volume Geodesic active contour incorporating prior shape information
of data to be analyzed; satisfaction of search etc [2].This can has been used by [11] Yuchong Jiang et al. to segment the leg
result in loss of money, time and litigations. Therefore there is bone. The algorithm is robust to background noise of the
a need to design systems that can aid experts in assessing casting material overlaying on the fractured leg. [12] Ying
bone anomalies. Chen et al. worked on developing a model based code to
automatically extract femur bone from X-ray images. Initially
Development in machine vision can enable doctors to use a model femur contour is registered to the x-ray image,
computers as second opinion to diagnose fractures in bone. followed by active contour with shape constrains to refine the
[3]. Such systems called Computer aided diagnosis (CAD) contour. [13] G. Behielset al. uses active shape model (ASM),
systems can prove very useful to analyze large volumes of involving a regularizing smoothness constrain to segment
medical data, as well as improve the accuracy of interpretation femur, humer and calcaneus bones in the human body.
while reducing time for diagnosis. A CAD for fracture
detection system consists of four blocks: preprocessing, 3. PREPROCESSING
segmentation, fracture detection and location of fracture [4]. The database consists of x-ray images obtained from different
This is shown in the Fig 1 below. x-ray machines and is very large in size. The size falls in
following range: height = 2068 – 3072 (in pixels) and width =
956 – 2844 (in pixels). To process such images is extremely
time consuming and real time implementation of
segmentation algorithms for such images is not possible. Also
Fig 1: Block diagram of bone fracture CAD system the orientation of the leg in the images is non uniform and
need to be corrected for uniformity and to aid automation. To
This paper deals with the first and second blocks. It focuses address these issues a pre-processing technique that corrects
on applying threshold based, region based, deformable model alignment and also reduces the size of the images has been
and cellular automata based segmentation techniques to solve devised. The method takes care that important information is
the problem of segmenting tibia bone accurately from X-ray not lost from the x-ray images.
images. This is followed by validation of segmentation using

34
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

Algorithm: The segmentation algorithms adopted must be robust to such

interpolated images, obtained due to pre-processing. Once the
1. Start ROI is obtained from segmentation, the binary mask of the
2. Read input image, I ROI is rescaled and rotated back to its original size and
inclination. The binary mask is then used to crop the ROI
3. Resize the input image I, by a scale of 0.15 to get the from the original, unaltered medical image. This ensures that
resized image Iresized. the final ROI obtained contains pixels having intensities that
have been originally provided to the segmentation system and
4. Apply Otsu segmentation [14] to the image, Iresized. the system is in no way modifying the intensities of the ROI
by processing it.
5. Calculate area of every region in the threshold image,
Iotsu. Prior to segmentation, the images are rotated and scaled.
Scaling and rotation of images involve interpolation. The less
6. Select the region with maximum area as the ROI. frequently this technique is used, the lesser is the distortion in
the image. This is because interpolation never adds additional
7. Calculate its inclination.
details to an image other than what is already present.
8. Correct the inclination of the bone using the formula Interpolation adopts the strategy of guessing pixel values at
additional new locations based on the neighbouring pixel
8.1. If the bone is inclined to the right, then it is rotated values of those locations. This causes loss of quality of
in anti-clockwise direction by an angle equal to images and hence is generally avoided on sensitive medical
“90-angle” (i.e. Anglecorrected = 90 - Angleoriginal) images
8.2. But if the bone is inclined to the left, then it is 4. SEGMENTATION METHODS
rotated in clockwise direction by an angle equal to The methods discussed in this section are categorized in Table
“-90-angle” (i.e. Anglecorrected = - 90 - 2.
Angleoriginal). Here the value obtained will be Table 2. Segmentation Techniques
negative. This denotes that the image should be
rotated in clockwise direction. No Method Category

9. The new image obtained is Irot. Next, the co-ordinates of 1. Manual thresholding
Pixel based
the bounding box of the maximum area region in the 2. Otsu
above image are obtained.
3. Watershed Region based
10. Using these coordinates a rectangle containing the bone
region (ROI) is cropped from the image Irot. The new 4. Active contour Deformable model
image obtained is Icrop, which is the final pre-processed 5. Growcut Cellular automata
image Inew. The following section explains the concepts behind the
11. Stop segmentation techniques.

The size of test images obtained after pre-processing fall in 4.1 MANUAL THRESHOLDING
the following range: height = 367 – 530 (in pixels) and width Manual thresholding requires user effort for selecting the
= 50 – 191 (in pixels).One of the results is shown in Table 1. right threshold value and is usually done with the help of a
histogram, using trial and error method. It involves converting
Table 1. Preprocessing result the input image into binary image based on a fixed threshold.
Before After The input image pixels having intensities greater than the
threshold level are assigned the value 1 and all the other
pixels are assigned the value 0 in the output image. The
manual thresholding does not hold good since the intensity
distribution range for the 3 classes (Skin, bone and
background) is not unique and varies with different images as
well as the X-ray modality used. Often the three classes are
Image found to overlap for a given area on the histogram. Due to
these reasons this method is not suitable for tibia bone
segmentation problem. The segmentation result for a
threshold value of 153 is shown in Table 3.
4.2 OTSU THRESHOLDING
Otsu thresholding [14] attempts to find that threshold value
Size (in pixels) 2446 x 2010 461 x 100 for which the sum of variances of the foreground and
Angles (in degree) 60.22543 0 background pixels is at its minimum. It is an automated
technique and is easy to implement. However the regions in
the image need to be well separated in the histogram to give
perfect results.

35
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

Table 3. Manual thresholding results Table 4. Otsu segmentation results

Segmented Tibia Segmented Segmented Segmented
Input Image Input
bone Tibia bone Tibia bone Tibia bone
Image
for T1 for T2 for T3
X-ray Machine 1

4.3 WATERSHED SEGMENTATION [15]

The watershed segmentation is known to face over
segmentation error due to irregularities in the gradient of the
image to be segmented. To overcome this markers are defined
on the image from where, the topological surface is flooded.
The algorithm used in Matlab for watershed segmentation is
the Meyer‟s flooding algorithm developed in the early 1990‟s.
A disc marker of radius 5 or 10 has been used to obtain the
X-ray Machine 2

results. The size of marker to be used for a given image is

empirically set. The results obtained using the above approach
is shown below.
Different regions of the Watershed output for one of the input
images have been indicated by different colours in Table 5.
Table 5. Watershed segmentation results
Over- Segmented Segmented
Input
segmented Tibia bone Tibia bone
Image
Image (D=5) (D=10)

Here three thresholds are found using Otsu method. The first
threshold (T1) is only good enough to segment the leg from
the background. So in order to get the bone the leg region is
provided to the Otsu algorithm to get the second threshold
(T2). But this results in over segmentation. So the region
having intensities between T1 and T2 is provided to the
algorithm to get a third threshold (T3). This has been tested to
give better results than T1 and T2.

The Threshold values can be shown on the intensity axis as

follows:-

At the end of the program, a user intervention is needed in

order to indicate the actual region of the watershed transform
Fig 2: Threshold values on the intensity axis that one is interested in. This can be avoided by determining
the region properties of the segmented regions and then
The threshold values are: T1 = 108, T2 = 191 and T3 = 151 classifying a region as bone if it satisfies the region properties
for the example shown in Table 4. of a bone. Also there is a need to automatically determine the
radius of the marker disc. The code for marker controlled
watershed is part of the demos section of Matlab 2012a and
can be run by typing ipexwatershed in the command window.

36
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

4.4 ACTIVE CONTOUR However the active contour method is known to face
problems at concavities. Furthermore the snake needs to be
SEGMENTATION initialized close to the ROI. This is often done manually. The
Active contours [16] or Snakes are widely used for edge snake is found to be highly sensitive to its parameters which
detection in the field of image processing. A snake can be are initialized empirically to values that vary for different
defined as a spline curve which tends to minimize its energy. types of images. The result for AC model is given in Table 6.
The energy of the snake is dependent on its shape (internal
energy) and its location (external energy). The total energy of 4.5 GROWCUT
the snake is given by:- GrowCut [17] segmentation is an interactive process,
modelled by evolution of cellular automata. Here every pixel
in an image can be considered as a cell. The cells can be
initialized as foreground, background, or undefined. As the
The above equation involves a parametric snake given by s(p)
algorithm proceeds, these cells compete to dominate the
= (x(p),y(p)). Eint and Eext are the internal and external
image domain. Some cells capture their neighbours, replacing
energies of the snake.
their labels. The ability of a cell to spread depends on its
Eint can is given by the formula, feature vector and strength. There can be more than 2 class
labels in an image. It can be extended to N dimensions.
The growcut involves „for‟ loops which prove to be
computationally expensive in Matlab. To solve this problem,
It controls the mechanical behaviour of the snake by the code for growcut has been written in C in Matlab (i.e. as a
determining its shape. Here α and β are constants. If alpha is binary mex file). Mex stands for Matlab executable.
less, the snake can elongate more and if beta is less the snake The most important challenged faced in applying the growcut
can bend more. The internal energy is meant to force the algorithm is the computation time for large medical images.
snake to be small and smooth. The results reported in the original paper of growcut [17],
Eext is given by the equation, mention a segmentation time of 4 sec for a 256x256 image on
a 2.5 GHz processor. However for large images the
computation time required is 10 minutes, which is quite large.
To address this problem the images were resized. The mex
It depends on the properties of the image. The simplest form code for growcut has been implemented by Shawn Lankton
of external energy is the inverse of the gradient magnitude of and can be downloaded from [18].
the input image. The results obtained are given in Table 7.
Internal and external forces determine the shape and position Table 7. Active contour segmentation results
of the snake. The internal force keeps the snake smooth
whereas the external force guides the snake towards the image Input image Image with seeds Output Image
features.
The output of the snake model is a closed contour. This is
very advantages if the region boundary has discontinuities.
The snake tends to take the general shape of the image
boundary. Also the edges of output of the snake model are
very fine. This will prove to be advantages if ones interest lies
in extracting features pertaining to the edges.
Table 6. Active contour segmentation results
External
Input Spline Segmented
Force
Image Image Tibia bone
Field

(Blue: ROI and

Red: Background)

5. VALIDATION
It is not possible to visually inspect the subtle differences
between an expert segmented image and a user segmented
image at all times [19]. In order to compare different
segmentation techniques and to encourage programmers to

37
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

develop more better and efficient codes, evaluation 1. Sensitivity (S)

parameters will be needed [20] .These parameters help to
evaluate the quality or “goodness” of image segmented by the It measures the proportion of tibia bone pixels that have been
above algorithms. In general validation techniques can be accurately segmented by the segmentation algorithm. Using
categorized as follows:- sensitivity, false negative rate (fnr) is obtained and is given
by:-
1. Analytical
False negative rate (%) = (1-sensitivity) x 100
2. Empirical
Since sensitivity gives the amount of foreground pixels
a. Empirical Goodness correctly segmented as foreground pixels, false negativity rate
can be defined at the percentage of foreground pixels
b. Empirical Discrepancy classified as background pixels or in other words false
The paper uses empirical discrepancy method for validation negative rate is the percentage of under segmentation error.
and analysis.
2. Specificity (Sp)
Since the output of a segmentation algorithm is affected by Similarly using specificity, false positive rate (fpr) is defined
multiple parameters, it is not possible that a single evaluation as the percentage of over segmentation error and is given by:-
parameter will prove effective in evaluating the “goodness” of
segmentation. Therefore four metrics have been used to False positive rate (%) = (1-specificity) x 100
validate the segmentation results. They include: Sensitivity,
Specificity, Jaccard and Dice co-efficient. The parameters are 3. Similarity metrics
chosen, such that they complement each other in their ability a) Jaccard co-efficient (J)
to evaluate a segmented output. The ground truth images have
been produced manually, by tracing a polygon on the The Jaccard co-efficient measures the ratio of area of overlap
boundaries of the tibia bone in the X-ray images. between the segmented output and the ground truth to the
union of their areas. This is defined as:-
The Venn diagram below will help to visualize the spatial
differences between a Segmented output (Results set) and its J = |A ∩ B| / |A ∪ B|
corresponding gold standard (Truth set) [21] b) Dice co-efficient (D)
The dice co-efficient is another metric used to define
segmentation quality or measure of similarity between the
segmented image and the gold standard image. It is defined
as:-
J = 2 |A ∩ B| / (|A| ∪ |B|)
The dice co-efficient is more commonly used parameter to
measure spatial overlap and is used when the number of non-
ROI pixels is greater than the number of ROI pixels [22].
Using the Jaccard co-efficient, the dice co-efficient can be
easily obtained, since the dice co-efficient and Jaccard co-
efficient are equivalent to each other and there exists a
Fig 3: Venn diagram monotonic relationship between them i.e.
Let X be the set of all pixels in the image. Then set A is the D=2J / (1+J)
set of pixels identified as tibia bone by gold standard Both the coefficients give value between 0 and 1. Here 1
segmentation and B is the set of pixels identified as tibia bone implies perfect similarity of the segmented image to its gold
by a segmentation algorithm. True positive set (TP) is then standard and 0 implies total dissimilarity between the two.
defined as the set of pixels identified as tibia bone by both
The Jaccard co-efficient gives a low percentage of similarity
gold standard and user defined segmentation. as compared to Dice co-efficient for cases where the spatial
i.e. TP = A ∩ B overlap is very less [23].
True Negative set (TN) is the set of pixels identified as non- However both the Jaccard and dice coefficients are unable to
tibia bone by both the algorithms. distinguish between over and under segmentation errors and
assume equal cost for both these errors. Also small errors in
TN = (A ∪ B)’ the segmented image cannot be identified by these parameters
Similarly False Positive (FP) and False Negative (FN) are [22].
given as, The segmentation time (t) along with the four evaluation
FP = (A‟ ∩ B) and FN = (A ∩ B‟) parameters have been calculated for 48 images. The average
values of the five parameters for every method are given in
Using these regions four evaluation parameters are defined as, Table 8. The segmentation algorithms have been coded in
Matlab R2012a (7.14.0.739) and run on a machine with the
1. Sensitivity = TP/(TP + FN)
following configuration:- Intel(R) Core(TM)2 Duo CPU
2. Specificity = TN/(TN + FP) T6600 @ 2.20GHz, 2.20Ghz with RAM: 4.00GB and
Windows 7 (64bit) as operating system.
3. Jaccard Co-efficient = TP/(FP + TP + FN)
4. Dice Co-efficient = 2TP/(FP + 2TP + FN)

38
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

Table 8. Averages of validation parameters and specificity as compared to watershed which means that its
accuracy is higher compared to watershed. Its relatively
AVERAGES higher value for specificity as compared to sensitivity implies
Methods t (sec) fnr fpr J D that it suffers more from under segmentation error as
(%) (%) compared to over segmentation error. Growcut performs the
Manual Thresholding 0.40 54 12 0.30 0.44 best among the 5 techniques and records a low value of 6%
and 1% of under and over segmentation error respectively.
Otsu Thresholding 0.48 17 25 0.44 0.61
Watershed 0.89 17 18 0.60 0.73
Active Contours 8.69 13 2 0.82 0.90
Growcut 3.78 6 1 0.90 0.95

For the sake of analysis, the average values of every

parameter are plotted against the segmentation methods in
Figs 4 to 8.

Fig 6: Graph of fpr (%) of Segmentation

Fig 4: Graph of Segmentation time

The first three segmentation algorithms give results in very
less time as compare to Active contour and Growcut.
However their accuracy is less. Active contour and Growcut
however give better segmentation results with an increase in
computation time. Among active contour and growcut, Fig 7: Graph of Jaccard Coefficient for Segmentation
growcut is less computationally expensive with higher
accuracy and is worth considering for automation.

Fig 8: Graph of Dice Co-efficient for Segmentation

The average error information for the segmentation methods
Fig 5: Graph of fnr (%) of Segmentation
are provided by the Jaccard and Dice similarity metrics. Their
Sensitivity gives an idea of accuracy as well as segmentation ideal value is one for perfect segmentation and less than one if
error. Manual segmentation gives a very low value which either one of the under or over segmentation errors occur.
implies poor segmentation. This is because of an under They record a high value only if both the segmentation errors
segmentation (fnr) error of 54% and over segmentation (fpr) are low.
error of 12%. Otsu, watershed, active contour and growcut
give a reasonably good value in the range 0.8 – 0.9. However The Jaccard and Dice coefficients (Figs. 7 and 8) record a
Otsu gives a low value of specificity in the range 0.75 ± 0.07 lower value of similarity for manual, Otsu and watershed
and records a maximum over segmentation error (fpr) of 25%. algorithms, which implies unsatisfactory results. The active
contour method gives a good value of 80-90 % similarity,
Watershed gives the same value of 0.8 for sensitivity and while the growcut gives the best value of 90-95%. The trend
specificity, which implies a reasonably good segmentation. was predicted by the sensitivity and specificity parameters,
Active contour has recorded a higher value for both sensitivity and is confirmed by the Jaccard and Dice coefficients.

39
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

6. CONCLUSIONS Pune) for their support in providing us the X-ray image

Based on the analysis the following points can be concluded:- database. Also a special thanks to the technicians Mr.Sanatan,
Mr.Vinay and Miss Usha.
1. Intensity based algorithms fail to successfully
segment bone from X-ray images because of the 9. REFERENCES
intensity distribution of the bone and skin classes. [1] Chipchase, L. S., K. McCaul, and T. C. Hearn. "Hip
There is no perfect separation of classes occurring fracture rates in South Australia: Into the next
in the histogram of the X-ray images. This cases century." Australian and New Zealand Journal of
over or under segmentation, which cannot be Surgery 70.2 (2000): 117-119.
corrected even by normal morphological operations
like region filling, image close, image open etc. [2] Berbaum, Kevin S., and Edmund A. Franken.
"Satisfaction of Search in Radiographic
2. Although the watershed segmentation gives results Modalities." Radiology 261.3 (2011): 1000-1001.
in very less time, the accuracy offered by the
technique is unsuitable for medical applications. [3] Giger, Maryellen L. "Computer-aided diagnosis in
Also the selection of marker size has to be medical imaging-a new era in image
automated. interpretation." Medical Imaging Ultrasound (2000): 75-
79.
3. Active contour in its most basic form seems to give
accurate result, but the process of selecting the [4] Umadevi, N., and S. N. Geethalakshmi. "Enhanced
snake parameters to give such results is tedious and Segmentation Method for bone structure and diaphysis
is empirical in nature. Also it suffers from the extraction from x-ray images." International Journal of
problem of detecting concave boundaries at the end Computer Applications 37.3 (2012): 30-36.
of the tibia bone. Therefore a better implementation [5] Nathanael E Jacob and M V Wyawahare. “Survey of
of active contours is needed that can solve the Bone Fracture Detection Techniques.” International
issues. One such method is the GVF implementation Journal of Computer Applications 71.17(2013):31-34.
of active contours.
[6] Ding, Feng, Wee Kheng Leow, and Tet Sen Howe.
4. Growcut has been proved to give better results as "Automatic segmentation of femur bones in anterior-
compared to Otsu, watershed and active contours posterior pelvis x-ray images." Computer Analysis of
both in terms of accuracy and speed of Images and Patterns. Springer Berlin Heidelberg, 2007.
segmentation. Although the method appears
promising, automation of growcut segmentation is [7] Manos, G. K., et al. "Segmenting radiographs of the hand
essential if the technique is to have practical and wrist." Computer Methods and Programs in
significance in the medical field. Automating the Biomedicine 43.3 (1994): 227-237.
seed selection step will prove to be challenging. [8] El-Feghi, I., et al. "X-ray image segmentation using auto
Also the automation process will tend to reduce the adaptive fuzzy index measure." Circuits and Systems,
accuracy of the algorithm. 2004. MWSCAS'04. The 2004 47th Midwest
Symposium on. Vol. 3. IEEE, 2004.
7. FUTURE SCOPE
The importance of bone segmentation can be understood in [9] Vinhais, Carlos, and Aurélio Campilho. "Genetic model-
clinical analysis, where it can be used for computer aided based segmentation of chest x-ray images using free
diagnosis and surgery [24]. Future work can include designing form deformations." Image Analysis and Recognition.
a computer aided diagnosis system (CAD). It is unlikely that Springer Berlin Heidelberg, 2005. 958-965.
such systems can replace surgeons because CAD is a system
where both physicians and computers play complementary [10] Yue, Zhanjun, Ardeshir Goshtasby, and Laurens V.
roles. Such systems will definitely help in assisting doctors in Ackerman. "Automatic detection of rib borders in chest
radiographs." Medical Imaging, IEEE Transactions
reducing diagnosis time and increasing accuracy of diagnosis.
on 14.3 (1995): 525-536.
In future CAD and PACS (Picture archiving and
communication systems) may be combined and used as a [11] Jiang, Yuchong, and Paul Babyn. "X-ray bone fracture
single unit for clinical diagnosis [25]. segmentation by incorporating global shape model priors
into geodesic active contours."International Congress
Future work can include designing a bone fracture computer Series. Vol. 1268. Elsevier, 2004.
aided diagnosis system (CAD). In such systems after
successfully segmenting the bone image from X-ray, the next [12] Chen, Ying, et al. "Automatic extraction of femur
step for a fracture detection system would be to extract contours from hip x-ray images." Computer Vision for
features from the segmented region. Once the features of Biomedical Image Applications. Springer Berlin
interest unique to a normal and fractured bone have been Heidelberg, 2005. 200-209.
obtained, various classifiers can then be used to classify the [13] Behiels, Gert, et al. "Active shape model-based
segmented bone as normal or fractured. A fractured bone can segmentation of digital X-ray images." Medical Image
be further classified based on the type of fracture. Some of the Computing and Computer-Assisted Intervention–
classifiers that can be implemented are Bayesian, Neural MICCAI‟99. Springer Berlin Heidelberg, 1999.
networks, SVM etc.
[14] Otsu, Nobuyuki. "A threshold selection method from
8. ACKNOWLEDGMENTS gray-level histograms."Automatica 11.285-296 (1975):
The authors would like to thank Dr. Kapil Kapadnis (Poona 23-27.
Hospital), Dr. Hemant Deshmukh (KEM Hospital, Mumbai)
[15] Meyer, Fernand. "Topographic distance and watershed
and Dr. Ashwini Kulkarni (Deenanath Mangeshkar Hospital,
lines." Signal processing 38.1 (1994): 113-125.

40
 International Journal of Computer Applications (0975 – 8887)
Volume 76– No.9, August 2013

[16] Kass, Michael, Andrew Witkin, and Demetri [21] 52. Shattuck, David W., et al. "Online resource for
Terzopoulos. "Snakes: Active contour validation of brain segmentation
models." International journal of computer vision 1.4 methods." NeuroImage 45.2 (2009): 431-439.
(1988): 321-331.
[22] Tahmasebi, Amir M., et al. "A validation framework for
[17] Vezhnevets, Vladimir, and Vadim Konouchine. probabilistic maps using Heschl's gyrus as a
"GrowCut: Interactive multi-label ND image model." Neuroimage 50.2 (2010): 532-544.
segmentation by cellular automata." proc. of Graphicon.
2005. [23] Manning, Christopher D., and Hinrich
Schütze. Foundations of statistical natural language
[18] http://www.shawnlankton.com/?s=growcut. processing. Vol. 999. Cambridge: MIT press, 1999.
[19] Torsney-Weir, Thomas, et al. "Tuner: Principled [24] Pham, Dzung L., Chenyang Xu, and Jerry L. Prince.
parameter finding for image segmentation algorithms "Current methods in medical image segmentation
using visual response surface exploration."Visualization 1." Annual review of biomedical engineering 2.1 (2000):
and Computer Graphics, IEEE Transactions on 17.12 315-337.
(2011): 1892-1901.
[25] Doi, Kunio. "Computer-aided diagnosis in medical
[20] Zhang, Yu Jin. "A survey on evaluation methods for imaging: historical review, current status and future
image segmentation."Pattern recognition 29.8 (1996): potential." Computerized medical imaging and graphics:
1335-1346. the official journal of the Computerized Medical Imaging
Society 31.4-5 (2007): 198.

41
IJCATM : www.ijcaonline.org