ARUP; ISBN: 978-0-9562121-5-3 - CMBBE 2012 - Cardiff University

3.2 Image Processing
A system that would assist physicians to diagnose GI lesions like cancer, cancer
precursors, polyps, and bleeding would increase the efficiency of the typical screening
methodologies. The capsule endoscopic images possess rich information expressed by
color and texture. Color images acquired during the procedure, are usually recorded for
reevaluation and archiving purposes. Physicians search for changes in the cellular
pattern (pit pattern) of the GI lining, which might be the very earliest sign of polyps.
An automatic detection of anomalies can be based in alterations in the texture of the
small intestine mucosa. This innovative system will support detection of GI lesions by
processing CE video-frames, using texture segmentation, in conjunction with neural
networks techniques. The first part of the system implements the extraction of image
features, where the texture information of the image is represented by a set of
descriptive statistical features calculated on the wavelet transformation of the image. In
the second phase, a neural network undertakes the classification of these features.
Many methods have been reported in the literature for the analysis of texture, which
may be used in automatic detection methods of anomalies, which potentially indicate
disease. For instance, Maroulis et al. [4] and Karkanis et al. [5] proposed two different
methods based on the analysis of textural descriptors of wavelet coefficients in
colonoscopy videos. Kodogiannis et al. [6] proposed two different schemes to extract
features from texture spectra in the chromatic and achromatic domains, namely a
structural approach based in the theory of formal languages and a statistical approach
based in statistical texture descriptors extracted from the different color channels’
histograms. From authors’ previous work, the application of texture analysis techniques
to classify capsule endoscopic frames is feasible and presents promising results. Cunha
et al. [7] and Mackiewicz et al. [8] suggest that a significant reduction of the viewing
time can be achieved by automatic topographic segmentation the capsule endoscopic
videos. Szczypinski et al. [9] have recently proposed a different interesting concept to
aid clinicians in the interpretation of capsule endoscopic videos. They propose the use
of a model of deformable rings to compute motion-descriptive characteristics and to
produce a two-dimensional representation of the GI tract’s internal surface. From these
maps, certain characteristics that indicate areas of bleeding, ulceration and obscuring
froth can be easily recognized, allowing therefore the quick identification of such
abnormal areas. Alexandre et al [10] proposed a method that compares texture based,
color and position based methods for polyp detection in endoscopic video images. Two
methods for texture feature extraction that presented good results in previous studies
were implemented and their performance was compared against a simple combination
of color and position features. For further notes on the available methodologies for CE
image processing, Karargyris and Bourbakis [11] proposed different methods for
classification of capsule endoscopic video frames based on statistical measures taken
from texture descriptors of co-occurrence matrices, using the discrete wavelet transform
to select the bands with the most significant texture information for classification
purposes.
4. ENDOSCOPIC CAPSULE
4.1 Capsule’s architecture
In this section, a capsule’s architecture is proposed. The capsule will be composed by
modules – vision, communication, power and locomotion – and they were thought
according to space constrains (Fig.1).