Emulating properties of digital color images of human skin and nevi using tissue-simulating phantoms

Our research group aimed to create a set of tissue-simulating phantoms that were low-cost, easy to produce, and representative of the properties of color images of human skin and nevi. The main ingredients include gelatin as a tissue base, and different combinations of coffee and creamer to vary the absorption and scattering of light by the phantoms. Inclusions (mimicking nevi) were constructed using higher concentrations of coffee (to simulate increased melanin content). Cancerous nevi were emulated by changing the concentration of creamer in the inclusion (mimicking cancer-related alterations in tissue scattering). The phantoms were then used for validation and characterization of machine-learning models developed by our group for detecting cancerous nevi in human skin using smartphone images. We compared the output of our algorithms for the phantom data with the results that our algorithms obtained for digital color images of human nevi in the publicly available International Skin Imaging Collaboration (ISIC) database. Our results show initial promise, with a subset of phantoms exhibiting color image data very similar to that of in vivo human skin and nevi tissue.