A Survey on Automated Skin Lesion Segmentation and Classification of Granular Parakeratosis

Abstract

Analyzing skin lesion images through manual and visual inspection is bit difficult in diagnosing skin diseases. This visual analysis of lesion images is very time consuming and cumbersome. Earlier, most of the research concentrates only on melanoma but there are yet other skin diseases to be considered. One of the diseases is Granular parakeratosis. It is a type of skin disease that is characterized by brown or red keratotic papules that combine to form plaques. The important steps in Computer-Aided Diagnosis (CAD) systems are lesion segmentation and classification. The role of lesion segmentation and classification in medical image processing is to enhance the quality of features extracted from a skin lesion. The advancements in machine learning have reduced the rate of misclassifying the lesions. Hence, it is necessary to have an imaging diagnosis system that can detect skin diseases early. In this paper, a review of some of the techniques used in detecting melanoma is presented to determine the technique best suitable for segmentation and classification of granular parakeratosis. Here, the discussion about how Convolutional Neural Networks have been used in identifying the lesion, the algorithms implemented and efficiency of the methods are presented. The other related issues such as data collection and evaluation metrics have been discussed. After a detailed review, it is found that U-net with binary cross entropy can be chosen as the best method for segmentation and SVM for classification of granular parakeratosis.