Article Preview
TopIntroduction
The increasingly growing amount of data in the field of healthcare industry world-wide necessitate the need of efficient data handling and analyzing techniques. Proper handling and analysis of data by intelligent machines can help in improving the disease detection (cardiovascular disease, cancer, etc.) capability as compared to manual techniques (Priyanga and Naveen, 2018). Skin cancer is now-a-days rampant in many parts of the world especially in the United States and Asian countries. There are three major types of skin cancers namely basal cell carcinoma (BBQ), squamous cell carcinoma (SCC) and melanoma, out of which melanoma is the most dangerous form of skin cancer. Melanoma develops when unrepairable damage is caused to the skin cells called melanocytes which produce the pigment melanin to protect us from the UV radiations. Melanoma is generally caused due to exposure to ultra violet (UV) radiations. Melanoma is characterized by the development of skin lesions that may vary in color, size, shape and texture. The color of the skin usually becomes black or brown but sometimes the color might also be red, blue, purple or white. In the United States, approximately 5.4 million new cases of skin cancer are detected every year. Despite the fact that only 5% cases represent melanoma, 75% of the skin cancer related deaths are due to melanoma, which accounts for around 10,000 deaths annually (Esteva et al., 2017).
Early detection of melanoma increases the chances of its curability as well as the life expectancy of the patient. Earlier, detection of skin cancer required a manual procedure which involves skin biopsy that was both time consuming and expensive. To overcome this problem, alternate means are being used by medical experts to diagnose melanoma using images of skin lesions. Medical image diagnosis requires intelligent systems. Many intelligent systems based on image processing, pattern recognition and machine learning have been proposed by researchers in past where millions of images can be processed at a time and can be used to predict or classify the image as melanomic or non melanomic. Skin lesions are examined and analyzed using a non-invasive procedure known as dermoscopy. Although dermoscopy improved the accuracy of detection of melanoma and reduced the number of biopsies as compared to naked eye examination, however it requires extensive training and expertise. Moreover, dermoscopic images are not available readily. Another method adopted by researchers nowadays is the use of conventional camera to capture the image of the pigmented region of the skin. Images collected from various clinics across the world are also used by researchers to classify them as melanomic or non-melanomic. The images collected from conventional camera are called as non-dermoscopic images. As compared to dermoscopic image, non-dermoscopic image contains fewer details and may have illumination and noise problem but these are readily available (Jafari et al., 2017a). An example of dermoscopic and non-dermoscopic images of melanoma taken from ISIC website and Med-node dataset (Giotis et al., 2015) respectively are shown in Figure 1.
Figure 1. a) Dermoscopic images of melanoma b) non-dermoscopic images of melanoma