04-08-2014, 11:32 AM
Segmentation of CT Brain Images Using K-means and EM Clustering
[attachment=66540]
Abstract
The combination of the different approaches for the
segmentation of brain images is presented in this paper.
The system segments the CT head images into 3 clusters,
which are abnormal regions, cerebrospinal fluid (CSF)
and brain matter. Firstly we filter out the abnormal
regions from the intracranial area by using the decision
tree. As for the segmentation of the CSF and brain
matter, we employed the Expectation-maximization (EM)
algorithm. The system has been tested with a number of
real CT head images and has achieved some promising
results
Introduction
Medical images are produced rapidly recent years.
Almost every day, huge medical visual data are produced
from X-ray, Computed Tomography (CT) scanner,
Magnetic Resonance Imaging (MRI) scanner and so on.
The images, if processed appropriately can provide very
useful information to assist doctors in diagnosis.
Image segmentation is the process of partitioning a
digital image into sets of pixels. Image segmentation is
important for classification and analysis. Manual brain
segmentation probably is more accurate than fully
automated segmentation ever likely to achieve. However,
the major drawbacks of manual image segmentation are
time consuming and subjectivity of human segmentation.
Therefore, it is significant to develop a reliable
automated segmentation to overcome the drawbacks of
manual segmentation. The challenges for automatic
segmentation of the CT head images have given rise to
many different approaches. The techniques of
segmentation developed so far include statistical pattern
recognition techniques[1,2,3], morphological processing
with thresholding [4,5], clustering algorithm[6,7] and
active contour[8,9].
In this paper, we segment the intracranial area into 3
clusters which are abnormal regions, CSF and brain
matter. The obtained results from image segmentation
play important role for annotation and image retrieval.
However, these are not within the scope of this paper.
Section 2 of this paper gives the brief overview picture
of the whole process. Section 3 elaborates the details of
the entire process. In Section 4, we show the
experimental results of the segmentation. In the final
section, conclusions and possible future work are
discussed.
. Binarization of the image
For the binarization of the enhanced image from
Section 3.3, we have attempted on three methods which
are k-means segmentation, EM segmentation and Otsu
threshold. The results for respective segmentation are
shown as in Figure 7. We convert the image into binary
image in order to apply the connected component
analysis to detect all the regions inside the brain.
From the obtained results, we discovered that for
Otsu threshold and EM segmentation the abnormal
regions are merged with the normal region. This makes it
impossible to filter out the abnormal region thoroughly.
On the contrary, k-means segmentation yields the best
result as the significant regions are separated from the
normal regions. This is crucial to grant that the abnormal
regions are able to be distinguished out from the normal
regions.
6. EM segmentation
After classification of the abnormal regions, EM
segmentation is adopted to segment the intracranial area
into two clusters which are CSF and brain matter.
Basically, EM algorithm[10] is a statistical estimation
algorithm used for finding maximum likelihood
estimates of parameters in probabilistic models.
The procedure to segment the CSF and brain matter
work as below:
Conclusions
We have elaborated our concept for the real CT head
image segmentation. The system performs image
segmentation based on different approaches for the
different regions of the brain. The method works quite
effectively for detection of the abnormal regions in the
brain.
Plans for future work include the annotation of the
abnormal regions such as hemorrhage, calcification and
lesion.