A significant number of people throughout the world are affected with skin cancers. Globally, it is estimated that more than 1 million new cases of nonmelanoma skin cancer were diagnosed in 2018 with 65,000 associated deaths. Thus, it is important that the skin cancer has to be detected as early as possible before the cells start invading and spreading.
In this paper, we examine imaging processes regarding illumination phase, aperture phase, and gray-scale imaging formation for human skin cancer detection (HAM10000). Results show that these physical layers appear to have only a slight improvement on classification. However, what is most interesting to us is to compare between each of these physical parameters and their final trained weights, which are supposed to be highly related to the image configuration we feed into our CNN.
