A novel and efficient skin modeling method using monte carlo simulation to extend the database for skin health diagnosis

Abstract

Currently, there are many methods to diagnose skin conditions such as dermatoscopy, biopsy, etc. In this study, we investigated alternative non-invasive skin diagnostic methods. Among them, we found that spectroscopy-based diagnostic approaches combined with artificial intelligence and computer vision showed promising results. However, the lack of a robust, extended database of the optical of human skin is a challenge to the development of these approaches. In this study, we proposed a method to rapidly build and extend a large database of human skin diffuse reflectance and skin parameters for Near Infrared (NIR) wavelengths, specifically in the range 400-1700 nm, based on skin modeling and Monte Carlo simulation using Graphics Processing Unit (GPU)-accelerated Monte Carlo for Multi-layered Media (MCML) simulation program CUDAMCML. We focused on modeling and simulating normal skin, dry skin and aging skin. Additionally, we extended the database on real human skin and evaluated our modeling and simulation method by measuring the diffuse reflectance of real human skin using the diffuse reflective illumination module NIR-M-R2 made by InnoSpectra. The results obtained were evaluated by comparing with existing data on real human skin using a similarity score. The process is accelerated and executed on a GPU using the Facebook AI Similarity Search (FAISS) algorithm. The overall results indicated that our simulation approach can reliably produce the optical properties of certain skin tones and conditions that closely match real human skin. Further development is needed to account for more diversity.