Study on Skin Cancer Classification Based on Image Fractal Dimension

Skin disease is one of the common diseases in human beings. A large proportion of skin diseases eventually develop into skin cancer because of lacking of timely detection and corresponding early treatment. One of the most deadly cancers is malignant melanoma, which kills many people every year. Fractal dimensions can be used to describe the properties of graphic surface roughness. For dermoscopic lesion images, fractal dimension is widely used to study the changes of internal structure of biological tissues. In this study, we used data set of ISIC 2016 to obtain a larger sample of data, excluding lots of hair, skin ulceration et al., the number of remaining pictures is 852, including 676 pictures of benign nevus and 176 pictures of melanoma. In the case of a relatively big sample, the fractal dimension of melanoma is still higher than that of benign nevus, even if the equipment, magnification, exposure time and other conditions are different. At the same time, in order to avoid the influence of the lesion area, we tried to unify the lesion area, and found that the fractal dimension of melanoma and benign nevus are still statistically different, and the p value is lower, which verified that the method of unifying the lesion area is effective. Specifically, changes in fractal dimension may reflect the pathological metabolic changes of melanoma.