Detection of Melanoma using Deep Learning Techniques: A Review
V. Vipin, Malaya Kumar Nath, V. Sreejith, Nikhil Francis Giji, Adithya Ramesh, M. Meera
Abstract
In this paper, a comprehensive review of skin cancer (malignant melanoma) segmentation and classification using computer vision and deep learning techniques have been presented. Among skin cancer, melanoma is the deadly skin cancer and leads to loss of life, if not treated and detected at its early stage. Melanocytes produces melanin are responsible for melanoma. Melanoma can found in all parts of the body, but prominently visible over the skin. It mainly occurs due to exposure to UV radiation and greatly reduced by limiting the skin exposure to UV radiation. Detection of early warning signs of melanoma can prevent further spreading to other parts of the body. Dermatologist examines skin lesion images based on their expertise and experience to analyze malignant melanoma. This procedure is time-consuming and subject to human error. The error may be overcome by using automatic methods for segmentation or detection by using computer vision and deep learning methods. Deep learning based medical image segmentation or detection is found to be efficient and outperformed human-level accuracy. In dermoscopy, there is a wide range of machine learning and deep learning algorithms applied for better classification, segmentation, and analysis of melanoma. These algorithms are trained using deep neural networks on a large number of melanoma images which are both malignant and benign, annotated by experts. A large number of datasets are used in the literature for melanoma detection. Few popularly used datasets are the ISIC archive, HAM10K, PH2, MED-NODE, and DermIS image library. Authors have used the area under the curve (AUC), accuracy, sensitivity, and specificity as the most common evaluation matrices of the models.