Abstract:In recent times, melanoma skin cancer is a commonly affecting disease over the globe. Visual analysis at the time of skin lesion examination is not an easier process, as there is a resemblance among lesions. Several imaging techniques have been available in the literature for skin lesion diagnosis and classification. Therefore, computer aided diagnosis (CAD) models are developed for assisting professionals in the diagnostic procedure. However, skin cancer diagnosis using dermoscopic images is highly challenging due to the limitations of artefacts, blurry boundary, poor contrast, and variable lesion shapes and sizes. Numerous research works have been carried out to resolve the challenging issues that exist in the skin lesion classification process. Recently, deep learning (DL) based CAD models have become more familiar in the skin lesion classification process. With this motivation, this paper reviews the existing practices, issues, and projections of image collection, pre-processing, segmentation, feature extraction, and classification of skin lesion images. The survey aimed to report the statistical results from the recently presented models. Besides, the efficiency of different classification models is compared explicitly designed to detect skin lesion diagnosis and examined the equivalent outcomes. Furthermore, the existing methods are reviewed based on the major objective, methodology used, and experimental outcomes. At last, the limitations involved in some of the reviewed methods are pointed out and future improvements are suggested.

A Survey on Deep Learning based Computer-Aided Diagnosis Model for Skin Cancer Detection using Dermoscopic Images