17-11-2012, 02:42 PM
Detection of Small Bowel Polyps and Ulcers in Wireless Capsule Endoscopy Videos
1Detection of Small Bowel.pdf (Size: 829.37 KB / Downloads: 33)
Abstract
Over the last decade, wireless capsule endoscopy
(WCE) technology has become a very useful tool for diagnosing
diseases within the human digestive tract. Physicians using WCE
can examine the digestive tract in aminimally invasive way searching
for pathological abnormalities such as bleeding, polyps, ulcers,
andCrohn’s disease. To improve effectiveness ofWCE, researchers
have developed softwaremethods to automatically detect these diseases
at a high rate of success. This paper proposes a novel synergistic
methodology for automatically discovering polyps (protrusions)
and perforated ulcers in WCE video frames. Finally, results
of the methodology are given and statistical comparisons are also
presented relevant to other works.
INTRODUCTION TO WIRELESS CAPSULE ENDOSCOPY
AN ESTIMATED 19 million people in the United States
suffer from diseases of the small intestine, such as obscure
bleeding, Crohn’s disease, chronic diarrhea, or cancer [1].
Wireless capsule endoscopy (WCE) is a technology that offers
medical doctors (gastroenterologists) the ability to view the interior
of the small intestine with a noninvasive procedure.
WCE was invented by a group of researchers in Baltimore in
1989, and later introduced by Given Imaging Ltd., Yoqneam,
Israel, as a commercial tool. Given Imaging Ltd. has developed
a swallowable wireless capsule PillCam SB2 (see Fig. 1) that
has been the only available product on the market since 2002.
A competitor, Olympus Corporation, has been selling the EndoCapsule
(see Fig. 2)in Europe since 2005, and received Food
and Drug Administration approval in September 2007. In general
terms, the operational differences between the two products
are minor.
Ulcers
In regard to ulcers, the definition of a peptic ulcer is given
as an area where tissue has been destroyed by gastric juices.
Gastric juices are produced by the stomach and the intestine to
digest the starch, fat, and protein in food. Since the intestine
and the stomach also consist of proteins, they are protected by
1) mucous layer, 2) bicarbonate, which neutralizes acid, and 3)
prostaglandins, which are hormones to boost bicarbonate and
mucus production.
Although most peptic ulcers appear in the stomach (gastric ulcers)
and the duodenum (duodenal ulcers) they may also appear
in the small bowel.
DETECTING POLYPS
It is worth mentioning at this point that polyp detection in
CTC is a very difficult task and all the previous CTC methodologies
are complex and sophisticated in order to reach a high
level of accuracy. However difficult this task might be though,
suspicious regions can be found on the boundary of the colon
lumen (see Fig. 4), whereas in WCE videos polyps can be found
anywhere (see Fig. 5). Extraction of polyp candidates in WCE
cannot be based on the lumen boundary because in most cases
it cannot be seen due to camera’s perspective. Therefore, in the
case of polyp identification in WCE the most crucial part is
the preprocessing step which is the segmentation scheme. The
segmentation process must fulfill following two goals.
1) Maintain details of object boundaries.
2) Extract only crisp segments.
However, these two goals cannot be satisfied successfully at
the same time using traditional segmentation algorithms based
on color or texture values, because these algorithms can either
focus on high detail, producing many segments, or focus on crisp
segments, losing boundary details.
Log Gabor Filters and Segmentation
Gabor filters have been widely used in image processing over
the last two decades. In [17], Daugmann and in [18] Webster
and De Valois showed that Gabor wavelet kernels have many
common properties with mammalian visual cortical cells. These
properties are orientation selectivity, spatial localization, and
spatial frequency characterization. In this sense, Gabor filters
offer the best simultaneous localization of spatial and frequency
information [19].
However, while Gabor filters are very successful they suffer
from bandwidth limitation. To obtain larger spectral information
while maintaining maximum spatial localization log Gabor
filters have been introduced. Log Gabor filters have a response
that is Gaussian when viewed on a logarithmic frequency scale
instead of a linear one like Gabor filters. Log Gabor filters can
be constructed with arbitrary bandwidth and the bandwidth can
be optimized to produce a filter with minimal spatial extent [19].
Machine Learning
Support vector machines (SVMs) are supervised learning
methods widely used to classify data. The basic concept is that
an SVMmaps the input data to an n-dimensional space, where it
tries to find the optimal hyperplane to separate the datasets [26].
The popularity of SVMs lies on their generalization ability for
a wide range of pattern recognition problems. As well put by
John Shawe-Taylor and Nello Cristianini: the key features of
SVMs are 1) the use of kernels, 2) the absence of local minima,
3) the sparseness of the solution, and 14) the capacity control
obtained by optimizing the margin.
CONCLUSION AND DISCUSSION
In this paper, a novel synergistic methodology was proposed
that deals with the detection of polyps and small bowel ulcers
(perforated). More specifically, our proposed methodology is
based on the robust and promising log Gabor filters, which behave
successfully as a segmentation scheme. With log Gabor
filters we managed to extract the dominating texture segments
and leave out the background “less meaningful” textures, which
is the goal of finding polyp candidates in WCE images.