Melanoma diagnosis using extraction proper features from dermatoscopic images

Background/ Diagnosis of melanoma in the early stages can have a significant effect on decreasing the mortality rate caused by this type of skin cancer. Melanoma is hard to be diagnosed in the early stages, even by the best physicians. Therefore, providing a method to easier the early diagnosis of this type of cancer is very valuable and vital. In this research, by extraction of images’ features and classifying them, we proposed an algorithm which enables the melanoma diagnosis. In the first step, before proper features’ extraction, we used polarized filter with proper angle for determining accurate boundaries between the normal and affected tissue. In the next step, with help of dermatologists we extracted the lesion’s features. Moreover, we optimized the different features in order to have an exact and accurate classification of skin lesion and reduction in classifier training time. Features optimization occurred via three following Principal Component analysis, sequential forward selection and consultation with the dermatologist. By classifier help, all the optimized features were classified, enabled us to complete the algorithm. In the condition that we used the SVM classifier and optimized features with the direct ordinal selection method, the algorithm succeeded in diagnosing melanoma with the approximate accuracy of 91%. The proposed method is useful for diagnosing the melanoma with a convenient price and considerable high accuracy. Also, this software package has the connection capability to dermatoscopy and is helpful for diagnosing the melanoma.