Title : 1D Fractal signatures computer-aided diagnosis methodologies for skin lesion classification
Abstract:
Today, there is an increasing number of skin diseases, so early detection of skin diseases is one of the top priorities worldwide. Ultraviolet radiation and tanning beds are the main risk factors. Early detection of skin lesions is currently a priority task worldwide. It has been measured that the range of precision in specialized dermatological clinics is between 64% to 80%. Therefore, computer-aided diagnosis applications are highly beneficial for dermatologists and their assistants, enabling more objective and accurate diagnoses. However, there are significant challenges to developing such applications, including skin lesions with irregular borders, low contrast between healthy skin and the lesion, unbalanced databases, and images of different sizes. Grossly, a computer-aided diagnosis model consists of a database, image processing (including segmentation, enhancement, and noise removal), feature extraction, and classification. I will discuss a computer-aided diagnosis model constructed from 1D fractal signatures and techniques for handling unbalanced databases. Fractal signatures are based on the mathematical theory of fractals, so lesions are assumed to have amorphous shapes. In this methodology, the lesioned region is not segmented, as is typically done in most work on the same topic. Hence, information on healthy skin is used, as well as information on lesioned skin. The goal is to reduce the time and errors associated with the segmentation process, which is often incorrect due to the amorphous shape of the lesions. Additionally, an assembly of classification spaces was employed to address the database imbalance effectively.
