Article Preview
TopIntroduction
Melanoma is one of the most dangerous skin cancer forms, increasing worldwide. In the last decade, melanoma has increased around the world.As per the American Cancer Society report for the year 2020, total reported cases of cancer are 1806590, out of which 100350 are of melanoma, i.e., 5.56 percent of the total cases. Whereas the mortality rate of melanoma is 1.1 percent of the total. Five years survival rate of melanoma is also quite high concerning all cancer (SEER Cancer Stat Facts).These statistics indicate that if skin cancer is diagnosed early, survival chances are much higher than cancer's overall survival rate. Skin cancer occurs on the surface of the skin. Poor skin lesion detection affects lesion segmentation and increases fault in classification. Skin lesion segmentation and classification (Baig et al., 2020; Jafariv el al., 2016) is the process to differentiate skin spots and background in a dermoscopic image and categorize the extracted spot as malignant (melanoma) or benign (non-melanoma). Examining a dermoscopic image is a complex task and the primary requirement for accurate skin lesions classification. In the diagnosis of skin cancer, dermoscopy has become a widely acceptable imaging device.The dermoscopic images inspected by a skin specialist are the solution in identifying skin cancer (Abuzaghleh et al., 2014). The task of inspecting dermoscopic images is complex and time-consuming; as a result, any incorrect judgment may lead to disaster. Therefore, automatic Computer-Aided Diagnosis (CAD) systems become an important tool for better inspection and clinical decision making and support dermatologists. Artificial intelligence provides various complex algorithms for human cognition to analyze and interpret complex medical and healthcare data and support maintaining healthcare ethics. AI provides computer algorithms that draw approximate conclusions without direct human effort. Machine Learning and Deep Learning, the two branches of AI, offer the capacity to gain information, processing, and actual output to the end-user that distinguishes AI from traditional healthcare technologies. Various AI-enabled systems use different AI algorithms that go through data patterns and generate logic for decision making. Such AI-enabled systems' primary aim is to analyze relationships between prevention or treatment techniques and patient outcomes. These systems are designed for practices and support disease diagnosis, cure protocol, drug discovery, personalized medicine, and care. Hospitals are also looking forward to using these AI enables systems for their routine operations, such as patient satisfaction, staff satisfaction, and cost-saving. AI mimicking humans using computers is a reality in healthcare, and big data with AI has changed medical specialties. Rubegni et al. (2002) presented an automated system for diagnosing skin lesions using a digital dermoscopy analyzer to evaluate skin lesions that use machine learning algorithms, and artificial neural networks to classify skin lesions. Manak et al. (2018) reported that Machine Learning enabled live-primary-cell phenotypic-biomarker for post-surgical adverse pathology. The system provides scores of adverse pathologies at the surgical procedure. Mukherjee (2017)discussed how AI-based systems supported in the domain of clinical care and explored its implementation for diagnostic reasoning when software-based systems are gradually involved in clinical decision-making. Skin lesion segmentation (Hasan et al., 2020) is performed on dermoscopic images to differentiate and extract skin spots and background. The feature extraction process is performed on dermoscopic images (Kavitha et al., 2011). Finally, based on selected features, decisions are being taken to decide whether the skin lesion is malignant or benign.For any CAD system, the major and central part is to perform lesion segmentation, as there are huge differences in features such as color, texture, and shape for a different patient. Hair and other noises such as ruler marks also make segmentation quite difficult. The different resolution of images is also a bottleneck in implementing segmentation algorithms. Some popularly known segmentation methods are thresholding, edge-based method, clustering-based method, watershed-based method, PDE based method, and ANN-based method. The thresholding technique is one of the simplest techniques for segmentation. Thresholding is further categorized as Global thresholding, Variable thresholding, and multiple thresholding. Variable thresholding is subcategorized as local thresholding and Adaptive thresholding. Machine learning is an application of artificial intelligence (AI) that automatically caters various algorithms to learn and improve based on experience (Yu et al., 2018). Machine learning algorithms use training data to build a mathematical model used in prediction and classification. The learning process of algorithms begins with observing the data, studying the data patterns, and drawing results in the future. Machine learning's key objective is to program so that computers learn inevitably without human intervention and regulate actions accordingly.The machine learning algorithms categorized as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning and evolutionary learning (Osisanwo et al., 2017). To classify the skin lesion as melanoma or non-melanoma, authors had taken SVM, k-NN, Logistic Regression, and ANN.