Multi-Class Skin Cancer Detection and Classification Using Hybrid Features Extraction Techniques

Abstract

Skin cancer is measured as one of the fatal types of cancer diseases in humans, among numerous kinds of malignancy. Current diagnostic classifications are lacking in finding an effective treatment. The effective and early stage treatment of skin disease can increase the survival rate of patients. Substantial investigative work has been developed to improve computer aided diagnosis system to detect cancer at early stage. However, early detection of skin cancer still requires better accuracy through experiment on digital skin lesion images as a multiclass classification, rather than using biopsy methods. This paper presents an intelligent framework to detect and classify four types of skin cancer. Before classification, noise removal from skin lesion is performed by gaussian filter. Textural and colour features are extracted from skin lesion to detect and classify cancer into four types. Support vector Machine is trained to classify Melanoma, Nevus, Basal and Squamous skin cancer types. Extensive experiments are performed on standard benchmark skin cancer images dataset with an improvement in accuracy of 92.41% after comparison with the well-known methods.