Dermatologic Hyperspectral Imaging System for Skin Cancer Diagnosis Assistance

Abstract

This paper presents the development of a dermatological acquisition system based on hyperspectral (HS) imaging for the assistance in the diagnosis of pigmented skin lesions (PSLs). The developed system is able to capture HS images of 50×50 pixels and 125 spectral bands in the VNIR (Visual and Near Infrared) region between 450 and 950 nm, using a cold light halogen illumination device. The system is able to capture images of a size of 12×12 mm in less than 1 second. Employing this system, a preliminary database of 49 HS images of PSLs from 36 patients was generated. The data was labeled in four different classes and classified using a supervised machine learning method optimized by means of a genetic algorithm. The results obtained in these preliminary experiments demonstrate the potential of the developed system to perform a rapid and accurate assistance in the skin cancer diagnosis task during clinical routine practice