4. Introduction
Liver cancer is one of the major
death factors in the world.
Early detection and accurate staging
of liver cancer is an important issue
in practical radiology.
Currently, the confirmed diagnosis
used widely for the liver cancer is
needle biopsy and it is an invasive
technique and not recommended.
5. Introduction
Therefore, Computed Tomography
(CT) has been identified as accurate
and non-invasive imaging modalities
in the diagnosis of hepatic lesions.
Manual segmentation of this CT
scans are tedious and prohibitively
time-consuming for a clinical setting.
6. Introduction
An application of abdominal CT
imaging has been chosen and
segmentation approach has been
applied to see their ability and
accuracy to segment abdominal CT
images.
An improved segmentation approach
based on Neutrosophic sets ( NS) and
fuzzy c-mean clustering (FCM) is
proposed.
7. Introduction
The abdominal CT image is
transformed into NS domain, which is
described using three subsets.
The percentage of truth in a subset T,
the percentage of indeterminacy in a
subset I, and the percentage of falsity in
a subset F.
Threshold for subset T,I and F is
adapted using Fuzzy C-mean algorithm.
8. Liver Segmentation:
Problems and challenging
Automatic segmentation is a very
challenging task due to the various
factors:
CT abdominal images are represented in gray level
rather than color.
Liver stretch over 150 slices in a CT image.
Irregularity in the liver shape and size between
patients and the similarity with other organs making it
harder to clearly identify the liver.
Indefinite shape of the lesions.
Low intensity contrast between lesions and similar to
those of nearby tissues.
9. Cross section of abdomen
region,
liver has similar intensity
with neighbour organs.
Liver Segmentation:
Problems and challenging
CT scan for patient has four
phases and four orientations
make segmentation very
difficult.
Algorithms used to analyze
abdominal CT can be both time
consuming and error regions.
10. Methodology
Applied pre-processing median filter
technique for CT liver images to enhance,
remove noise that caused by defects of CT
scanner and improve the quality.
Develop hybrid technique to segment liver
from abdominal CT using:
Fuzzy C-mean algorithm (FCM)
Neutrosophic Sets and its operations
Connected Component Labeling
algorithm(CCL)
11. Fuzzy C-mean algorithm (FCM)
FCM is an unsupervised clustering algorithm
that has been successfully applied to a
number of problems
FCM adopts fuzzy partitions to make each
given value of data input between 0 and 1 in
order to determine the degree of its belonging
to a group.
FCM is a fuzzy clustering method allowing a
piece of data to belong to two or more
clusters.
12. Fuzzy C-mean algorithm (FCM)
Fuzzy set has been applied to handle
uncertainty.
In applications such as expert system, Medical
system, should consider not only the truth
membership, but also the falsity membership
and the indeterminacy of the two memberships.
It is hard for fuzzy set to solve such problems.
13. Neutrosophic Sets (NS)
Neutrosophy is a branch of philosophy,
introduced by Florentin in 1995.
It can solve some problems that can not be
solved by fuzzy logic.
NS values studies the propositions. Each
proposition is estimated to have three
components: the percentage of truth in a
subset T, the percentage of indeterminacy
in a subset I, and the percentage of falsity
in a subset F
14. Neutrosophic Sets (NS)
The main distinction between neutrosophic
logic (NL) and fuzzy logic (FL) is that the NL
is a multiple value logic based on neutrosophy.
FL extends classical logic by assigning a
membership function ranging in degree
between 0 and 1 to variables.
Neutrosophic logic introduces a new
component called “indeterminacy” and carries
more information than fuzzy logic.
15. Neutrosophic Sets (NS)
Example:
Ahmed wants to invite Karim to the lunch.
Karim may or may not accept the invitation.
In neutrosophic terms, the statement “Karim
will accept the invitation” can be described
in the following way: it is 60% true, 40%
indeterminate, and 30% false.
Neutrosophic logic is close to human
reasoning in the way that it considers the
uncertain character of real life.
16. Neutrosophic Sets (NS)
Convert Image to Neutrosophic
The image is transformed from image domain
to neutrosophic domain.
Each pixel in the neutrosophic domain can be
represented as T,I, and F which means the
pixel is t% true, i% indeterminate and f%
false.
The pixel P(i, j) in the image domain is
transformed into neutrosophic domain
PNS (i, j) ={T(i, j), I(i, j), F(i, j)}.
17. T(i, j), I(i, j) and F(i, j) are defined as:
is the local mean value of the pixels of the
window size , Ho(i, j) is the homogeneity value of T
at (i, j) which is described by the absolute value of
difference between intensity g(i, j) and its local mean
value
Neutrosophic Sets (NS)
Convert Image to Neutrosophic
18. The image is divided into three parts: objects (O),
edges ( E ), and background ( B ).
T(x, y) represents the degree of being an object
pixel,
I (x, y) is the degree of being an edge pixel, and
F(x, y) is the degree of being a background pixel
for pixel P(x, y).
This three parts are defined as follows:
Neutrosophic Sets (NS)
Convert Fuzzy NS to binary image
19. Neutrosophic Sets (NS)
Convert Fuzzy NS to binary image
Where tt, tf and ii are the thresholds computed from
subsets {T,I,F} using adaptive FCM. The objects and
background are mapped to 1, and the edges are mapped
to 0 in the binary image.
21. Proposed approach
Phases
The first phase searches for suitable slices in CT
DICOM file because liver intensity distribution is
different between slices. Liver parenchyma is the largest
abdominal object in middle slices. These slices are
suitable for segmentation and gives high accuracy.
The second phase pre-processing algorithm is used
before the segmentation phase to enhance contrast,
remove noise and emphasize certain features that affect
segmentation algorithms and morphology operators.
22. Proposed Technique
Phases
The third phase CT image is transformed into NS domain.
Each pixel in the NS domain represented by T,I, and F
which means the pixel is t% true, i% indeterminate and f
% false.
In fourth phase binarize neutrosophic image PNS(T,I,F)
using Fuzzy C-means algorithm.
Finally, the fifth phase is the post-processing CCL
algorithm used to remove small objects, false positive
regions and focused on liver parenchyma.
25. Experimental Results
CT DataSet
CT dataset collected from radiopaedia divided into seven
categories depends on the tumor type of benign and malignant,
each of these categories have more than fifteen patients, each
patient has more than one hundred fifty slices, and each
patient has more than one phases of CT scan (arterial, delayed,
portal venous, non-contrast), also this dataset has a diagnosis
report for each patient.
26. Experimental Results
Hybrid technique based on Neutrosophic set and Fuzzy C-means
clustering algorithm to segment liver automatically from
abdominal CT is proposed.
A middle slice was selected as suitable abdominal CT image of a
patient from DICOM file.
Pre-processing median filter3x3 window was used to enhance,
remove noise and emphasize certain features that affect
segmentation process.
The image is transformed into neutrosophic set based on T,I,F.
27. Experimental Results
NS image is enhanced using intensification transformation to
improve the quality and emphasizes certain features of CT
image to makes segmentation easier and more effective.
A new a adaptive threshold based on FCM are employed to
reduce the indeterminacy degree of the image for three
membership sets T, I and F.
NS CT image is converted to binary image based on
threshold extracted from FCM for T, I and F.
28. Experimental Results
Finally, Post-processing CCL is applied with 8-connected
objects to search for the largest connected region, remove
false positive regions and focus on the ROI.
29. Experimental Results
The performance and accuracy of the
proposed approach was evaluated by Jaccard
Index and Dice coefficient techniques
between automated segmented images and
manual segmented images.
The proposed approach applied on 30
abdominal CT images
The accuracy obtained from Jaccard Index is
88% and accuracy obtained from Dice
coefficient is 94% .
31. Experimental Results
Liver Segmentation Process
Results of NS algorithm based on FCM for Liver segmnetation (a) Original,(b) T-domain, (c) F-domain, (d) Enhanced,
(e) FCM for T-Image, (f) FCM for F-Image, (g) Homogeneity image, (h) Indeterminate image, (i) FCM for I-Image, (j)
Binary Image based on T,I,F and finally (k) Segmentation of abdominal CT organs, and (l) Liver segmentation.
32. Results of NSFCM for segmenting noisy CT liver image (a) Original,(b) T-domain, (c) F-domain, (d) Enhanced, (e)
FCM for T-Image, (f) FCM for F-Image, (g) Homogeneity image, (h) Indeterminate image, (i) FCM for I-Image, (j)
Binary Image based on T,I,F , (k) Segmentation of abdominal CT organs, and (l) Liver segmentation.
Experimental Results
Liver Segmentation Process for Noisy Image
33. The proposed NSFCM approach compared
with some of existing methods such as
classical fuzzy c-mean, local threshold and
global threshold Otsu's method.
From experimental results, the proposed
approach gives clear and well connected
boundaries.
The result gives an improvement better than
those obtained by other methods.
Experimental Results
Comparison between proposed technique with other
methods
34. Comparison between proposed technique with other methods to segment noisy CT liver. (a) Original
Image, (b) Fuzz C-mean Image, (c) Global Threshold - Otsu's, (d) Local Threshold, and (e) proposed
NSFCM.
Experimental Results
Comparison between proposed technique with other
methods
36. Conclusion
The presented approach for liver segmentation is able to
reliably segment liver from abdominal CT in the used
patient database
The image is described as a NS set using three
membership sets T, I and F and thresholded using FCM.
The experiments on abdominal CT images with noise
demonstrate that the proposed technique can reduce the
indeterminate of the CT images and perform optimum
threshold with better results especially in noisy cases.
37. The experiments demonstrate that neutrosophy can reduce
over-segmentation and gives better performance on noisy and
non-uniform images than obtained by using other methods,
since the proposed technique can handle uncertainty and
indeterminacy better.
Conclusion
38. Future Work
The proposed system can be improved by integrating
Neutrosophic logic and Evolutionary algorithms to
better manage inaccuracies.
We plan to assess the performance using a large
dataset to evaluate generalization performance of the
algorithm that includes a number of parameters in
the feature measurement process, which means it
might sensitive to size and characteristics of liver.