Classification of Skin Lesion Using (Segmentation) Shape Feature Detection

Classification of Skin Lesion Using (Segmentation) Shape Feature Detection

Satheesha T.Y. (Department of Electrical and Computer Engineering, School of Engineering and Technology, CMR University, Bangalore, India)
Copyright: © 2020 |Pages: 8
DOI: 10.4018/978-1-7998-0326-3.ch011

Abstract

Malignant melanoma has caused countless deaths in recent years. Many calculation methods have been created for automatic melanoma detection. In this chapter, based on the traditional concept of shape signature and convex hull, an improved boundary description shape signature is developed. The convex defect-based signature (CDBS) proposed in this paper scans contour irregularities and is applied to skin lesion classification in macroscopic images. Border irregularities of skin lesions are the predominant criteria for ABCD (asymmetry, border, color, and diameter) to distinguish between melanoma and nonmelanoma. The performance of the CDBS is compared with popular shape descriptors: shape signature, indentation depth function, invariant elliptic Fourier descriptor (IEFD), and rotation invariant wavelet descriptor (RIWD), where the proposed descriptor shows better results. Multilayer perceptron neural network is used as a classifier in this work. Experimental results show that the proposed approach achieves significant performance with mean accuracy of 90.49%.
Chapter Preview
Top

Proposed Method

In this section, we represent our extension to the shape sig- nature and show it can be more helpful in describing a concave contour in comparison with the conventional shape signature. The new signature is created based on a known concept: convex hull. Hence, we firstly give the definition of conventional signature and convex hull and then define the new signature. Finally, a dozen of shape and textural feature set will be presented to link with new signature attributes to improve melanoma prognosis.

Conventional Shape Signature

Shape signature is a common method for one-dimensional description of shape contour.

Shape signature is a common method for one-dimensional description of shape contour. For a closed contour 𝐶(𝑡) = (𝑥(𝑡), 𝑦(𝑡)): [1, 𝑇] → ℕ2, (Gonzalez & Woods, 2008).

For example, signature curve of a melanoma lesion is depicted in Figure 1(c). While for an arbitrary 𝜃 melanoma lesion (Figure 1(a)) has three intersection points, its equivalent signature curve only considers the first centroid distance. Also, shaded region in Figure 1(a) shows the region that is not properly de- scribed by shape signature.

Convex Deficiency Based Signature

The convex hull of a region is the smallest convex region containing it. Convex hull is a useful tool for shape description, particularly when the shape boundary includes significant con- cavities (Gonzalez & Woods, 2008).

As an example, Figure 1(a) demonstrates contour of a melanoma lesion and its convex hull. The CDBS curves of melanoma lesion is plotted in Figure 1(d). It is comprehended from CDBS diagram that shapes with abrupt boundaries have more significant peaks on their CDBS curves, and also shape with smooth boundaries tend to have more zero-incidence points. Therefore, concavities of non-convex contours can be exhibited accurately by CDBS curve is plotted in Figure 1(c).

Figure 1.

Illustrations of shape signature and proposed CDBS curves of a melanoma lesion

978-1-7998-0326-3.ch011.f01

Complete Chapter List

Search this Book:
Reset