Statistical Analysis of Hair Detection and Removal Techniques Using Dermoscopic Images

Abstract

Deaths due to various types of cancers have increased to a greater extent in decades. Computer-aided diagnosis is the fast and efficient way used in the medical field all around the globe for early diagnosis and treatment of cancer. The design of such automated systems is a major challenge in the medical field due to various aspects and the availability of data for testing these systems. Skin cancer is one such type of cancer that if treated at an early stage helps to reduce the mortality rate. Many technological solutions have been provided by researchers in the last decade for the early detection and classification of skin cancer. Hair detection and removal is one of the primary pre-processing step in the skin cancer detection process. Dull Razor, Adaptive Principal Curvature, E-shaver, etc. are common techniques used for hair detection and removal. These methods aim to remove the hairs from the lesion image, but some artifacts and background abnormalities are left behind in the resultant images. In this paper, slight functional modification using different color spaces and hybridization of various existing techniques for hair detection and removal has been proposed. The proposed techniques are evaluated on standard dermoscopic datasets using different standard performance metrics like Accuracy, Sensitivity, Specificity, False Positive Rate, Peak Signal to Noise Ratio, and Structural Similarity Index Measure. The proposed pre-processing methods are tested for classification accuracy using VGG-16 model. The evaluation results indicate that Modified E-shaver and Modified Dull Razor methods perform better than existing systems.