Skin disorders are one of the most common types of disorders that are primarily diagnosed visually with scientific screening observed through dermoscopic evaluation, histopathological evaluation, and a biopsy. Diagnostic accuracy has a strong relevance to physician skill. Painful effects of skin disease hamper the mental condition of a patient. The authors propose an approach to detect the skin diseases based upon image processing as well as machine learning techniques i.e., convolutional neural networks (CNN). CNN is a specific type of neural network model that allows us to extract higher depictions for the image content. It is a deep learning algorithm to perform generative and descriptive tasks. Machine learning generates two types of prediction-batches and real time.
TopIntroduction
Any abnormalities on the skin, whether benign or malignant, are referred to as skin lesions. Cancer is present in malignant lesions. Cancer risk is reduced by early detection and treatment of precancerous skin lesions. According to WHO (World Health Organization) statistics, 132,000 cases of melanoma skin cancer and 2 to 3 million cases of non-melanoma skin cancer occur annually worldwide. A 10% reduction in ozone levels will lead to an increase in skin cancer cases of 3,00,000 non-melanoma and 4,500 melanoma. Although more prevalent, non-melanoma skin cancers like basal cell carcinomas as well as squamous cell carcinomas are less likely to spread and only partially cause disability or even death. Early detection and accurate or precise diagnosis of the skin cancer can aid in the recovery process, ensure proper medical care, and help prevent the worst side effects. Therefore, a system for early detection can help and raise public awareness purpose in identifying different types of skin cancer or other skin disorders like a benign tumour on the skin that look a lot like skin cancer is needed. The automatic classification of skin disorders can assist individuals in recognizing skin conditions as they develop and prompting immediate consultation with medical professionals to receive the necessary medical care. Medical professionals and other biomedical researchers can use a number of related studies are based on digital image processing for the detection along with classification of skin cancers as a means to diagnose skin conditions more precisely and quickly.
Visual inspection of dermoscopic images, in general, necessitates the expertise of a dermatology specialist. Skin lesion diagnosis is difficult, even for an expert physician. As a result, invasive biopsy of the affected lesion is required for doctors to make an accurate diagnosis. A dermoscopic imaging technique captures and visualises deep skin structures by using a non-polarized light source and also magnifying optics.
Without additional technical support, dermatologists have a 65%-80% accuracy rate in diagnosing melanoma. Andre Esteva et al. (2017) proposed method uses convolutional neural networks that classify skin lesions into benign or malignant eosin based on a novel regularized technique. A method for classification of melanoma skin cancer using convolutional neural network is proposed by Nachbar et al. (1994). The application uses the Convolutional Neural Network method and the LeNet 5 architecture for classification, and the percentage of accuracy achieved was 93% in training and 100% in testing. The variety of educational information used from 176 snapshots and 100 epochs.
On the exact diagnosis of the disease that helps aiding in decision-making clinically, early classification of the skin lesions can increase the likelihood of curing cancer before it spreads. Most skin disease images that are used for training are unbalanced and rare, making automatic skin cancer classification difficult to achieve. In addition, the cross-domain adaptability and robustness of the model are major obstacles. To address the above problems and obtain satisfactory results, many deep learning-based skin cancer classification methods have recently become popular (Estava, 2015). Reviews addressing the above borderline problems in relevance with skin cancer classification are even hard to come by. Therefore, in this article, we provide a comprehensive overview of the latest deep learning-based skin cancer classification algorithms. First, an overview of three different types of dermatological images is provided, and then a list of publicly available skin cancer datasets is provided. Subsequently, the successful uses of standard convolutional neural networks for skin cancer classification are reviewed. Highlights of this paper include a summary of several frontier issues such as data imbalance, data confinement, domain fitting, model robustness, and model efficiency and their corresponding solutions in the skin cancer classification task (Khan, 2018). In conclusion, by summarizing different deep learning-based approaches to address current problems in skin cancer classification, we can draw the general conclusion that the development of these methods is generally going in a structured, lightweight, and multimodal direction. Additionally, we have presented our results in figures and tables for reader-friendliness. Deep learning is becoming increasingly popular, but there are still many problems to be solved and opportunities to be pursued for the future (Kim, 2017; Manne, 2020).