In the past decades, there have been developments in computational methods for helping dermatologists diagnose skin cancer early. Computerized analysis of pigmented skin lesions (PSLs) is a growing field of research.
Its main goal is to develop reliable automatic tools to recognize skin cancer from images. Studies have shown that automated systems are capable of diagnosing melanoma under experimental conditions. Moreover, computer-aided diagnosis (CAD) systems have the potential to prove useful as a backup for specialist diagnosis, reducing the risk of missed melanomas in highly selected patient populations.
Machine learning (ML) has evolved considerably over the past decade due to the availability of larger image databases and improvements in computer architecture. Advances in deep neural networks have also been a critical factor in making deep learning slowly supplant customary machine learning models for the detection of skin cancer.
There are various types of imaging devices for skin cancer detection. The most common equipment used to investigate the characteristics of pigmented skin lesions is the dermoscope, which is used with a conventional digital camera. Dermoscopic images display subsurface micro-structures of the epidermis and upper dermis. However, these devices are not widely available for public use. On the other hand, conventional digital cameras with spatial resolution (without the dermoscope) are commonly used by non-dermatologists such as general practitioners. Images taken by these devices are called macroscopic or clinical images.
In the process of design and development of a CAD system, it is crucial to know the previous works carried out in the field. In iToBoS, we are reviewing works done on automatic skin cancer detection approaches using machine learning over the past decade with a special focus on clinical images. For this purpose, we found papers in these criteria that were mainly concentrated on segmenting, pre-processing and classifying clinical images of skin lesions. We reviewed the selected articles in terms of pre-processing, image segmentation, feature extraction, change detection, and classification methods. We will also discuss various datasets used in the state-of-the-art papers. Finally, we will write an article that presents the conclusions of this study.