On the possibility of application of optical properties of human skin as a diagnostics criterion of nonmelanoma skin cancer

Date

2010-05

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

The current project investigates the possibility of the usage of the absorption co- e cient and the scattering coe cient in order to di erentiate between nonmelanoma cancerous skin and healthy skin. It also investigates the possibility of implementation of arti cial neural networks in order to resolve the healthy/cancerous skin classi cation problem. The therapeutic window, 600-1200 nm, where the di erence between scattering and absorption is much more pronounced, is used. Results, based on Analysis of Variance (ANOVA) and multiple mean comparisons, indicate that Nodular Basal Cell Carcinoma (NBCC) can be distinguished from other nonmelanoma skin cancer and healthy skin on the base of the absorption coe cient. NBCC and Squamous Cell Carcinoma (SCC) can be distinguished from In ltrative Basal Cell Carcinoma (IBCC) and healthy skin on the base of the scattering coe cient. NBCC is characterized by the lowest level of the absorption coe cient, while SCC is characterized by the lowest level of the scattering coe cient. The multilayer perceptron is used in order to solve a complex classi cation problem between healthy types of skin (epidermis, dermis, subcutaneous fat) and cancerous skin types(NBCC, BCC, IBCC) .A neural network with 2 input (scattering and absorption coe cient) and 6 hidden layers is developed. The mean square error (MSE) of classi ca- tion and probability of misclassi cation are estimated. Results indicate that absorption and scattering coe cients provide a basis for supplementary diagnostics of skin cancer and its automated detection. However, additional information, such as age, color of skin and gender is needed in order to increase the accuracy of classi cation.

Description

Keywords

Citation