Pit Pattern Classification Using Multichannel Features and Multiclassification

Pit Pattern Classification Using Multichannel Features and Multiclassification

Michael Haefner (Medical University of Vienna, Austria), Alfred Gangl (Medical University of Vienna, Austria), Michael Liedlgruber (Salzburg University, Austria), A. Uhl (Salzburg University, Austria), Andreas Vecsei (St. Anna Children’s Hospital, Austria) and Friedrich Wrba (Medical University of Vienna, Austria)
DOI: 10.4018/978-1-60566-314-2.ch022
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Wavelet-, Fourier-, and spatial domain-based texture classification methods have been used successfully for classifying zoom-endoscopic colon images according to the pit pattern classification scheme. Regarding the wavelet-based methods, statistical features based on the wavelet coefficients as well as structural features based on the wavelet packet decomposition structures of the images have been used. In the case of the Fourier-based method, statistical features based on the Fourier-coefficients in ring filter domains are computed. In the spatial domain, histogram-based techniques are used. After reviewing the various methods employed we start by extracting the feature vectors for the methods from one color channel only. To enhance the classification results the methods are then extended to utilize multichannel features obtained from all three color channels of the respective color model used. Finally, these methods are combined into one multiclassifier to stabilize classification results across the image classes.
Chapter Preview
Top

Pit Pattern Classification

Polyps of the colon are a frequent finding and are usually divided into metaplastic, adenomatous, and malignant. As resection of all polyps is time-consuming, it is imperative that those polyps which warrant endoscopic resection can be distinguished: polypectomy of metaplastic lesions is unnecessary and removal of invasive cancer may be hazardous. For these reasons, assessing the malignant potential of lesions at the time of colonoscopy is important.

To be able to differentiate between the different types of lesions a classification method is needed.

The most commonly used classification system for distinguishing between non-neoplastic and neoplastic lesions in the colon is the pit pattern classification originally reported by Kudo, Hirota et al. (1994) and Kudo, Tamura et al. (1996).

This system allows a differentiation between normal mucosa, hyperplastic lesions (non-neoplastic), adenomas (a pre-malignant condition), and malignant cancer based on the visual pattern of the mucosal surface. Hence, this classification scheme is a convenient tool to decide which lesions need not, which should, and which most likely can’t be removed endoscopically. The mucosal pattern as seen after dye staining and by using magnification endoscopy shows a high agreement with the histopathologic diagnosis. Furthermore, due to the fact that this method is based on the histopathologic (and therefore visual) structure of the mucosa, it is a convenient choice for a classification using image processing methods.

As illustrated in Figure 1, this classification method differentiates between the five main types I to V according to the mucosal surface of the colon. Type III is divided into two sub-types, III-S and III-L, designating the size of the pit structure. The higher the number of the pit type is, the higher is the risk that the lesion under investigation is malignant.

Figure 1.

Pit pattern classification according to Kudo et al.

Key Terms in this Chapter

Class ification Feature: A numerical or syntactical value used to describe an observed property of an object (e.g., size, color, shape, …).

Wavelet Transform: A transform used to decompose a signal into its frequency components, similar to the Fourier transform. But the time-frequency resolution of the wavelet transform can be adjusted since basis functions with compact support are used, in contrast to the Fourier transform, where sine and cosines are used as basis functions.

Colonoscope: A flexible, lighted instrument used to examine the inside of the colon.

Color Histogram: A graphical representation of a distribution of colors within an image. The data contained in a histogram is obtained by counting the occurrence of each possible color of the respective color model within the image.

Classifier: An algorithm to assign unknown object samples to their respective classes. The decision is made according to the classification feature vectors describing the object in question.

Colonoscopy: A medical procedure during which a physician is examining the colon for polyps using a colonoscope.

Fourier Transform: An algorithm used to decompose a signal (e.g., an image) into its frequency components and to compute the frequency spectrum for a given signal.

Class ification Feature Vector: A collection of classification features describing the properties of an object.

Complete Chapter List

Search this Book:
Reset