Skin lesion images classification using new color pigmented boundary descriptors

Computational methods play an important role in enhancing the diagnosis of the skin cancer. Melanoma is the most fatal type of skin cancers that causes significant number of deaths in recent years. In this paper, novel boundary features are introduced based on the color variation of the skin lesion images, acquired with standard cameras. Furthermore, to reach higher performance in melanoma detection, a set of textural and morphological features are associated with proposed features. Multilayer perceptron neural network is used as classifier in this work. Results analysis indicate that proposed feature set has the highest mean accuracy (87.80%), sensitivity (87.92%), specificity (87.65%) and precision (90.39%) in comparison with the previous works in Dermatology Information System (IS) and DermQuest datasets.

I. Introduction

One of the most usual types of cancer in different countries is skin cancer, of which the incidence rate has increased during recent years [1]. Among all forms of skin cancers, melanoma is the deadliest one [2]. According to estimation of The American Cancer Society, in Unites States about 76380 new cases of melanoma will be diagnosed (46870 in men and 29510 in women) and about 10130 deaths from melanoma will occur (6750 in men and 3380 in women) in 2016 [3], [4].