An open system to generate hyperspectral cubes for skin optical reflectance analysis

International audienceOptical visible light analysis of the skin mainly relies on RGB colour imaging. Multispectral imaging of human skin represents a more accurate approach for in vivo analysis of its optical properties. Few dermatology-oriented multispectral devices have been hitherto developed. Their diagnostic abilities for pigmentary tumors or other clinical applications are still being evaluated [1]. We have developed a multispectral imaging system for in-vivo skin optical analysis. It is based on a charge-coupled device camera, a rotating wheel with a set of seven interference filters and a computer. All interference filters were chosen specifically to cover the visible spectrum range from 400 to 800 nm. Light transmitted by an optical fibre is filtered in each waveband to illuminate the area of skin under investigation. A dedicated software drives the rotating wheel and yields a set of seven monochannel images resulting in one multispectral image. In the next step, our system generates a hyperspectral cube from this multispectral image. The reconstruction process uses a light propagation model and a robust method based on an artificial neural network [2]. The robustness of neural network associative memories against noise allows generation of a hyperspectral cube integrating reflectance spectra for high number of single pixels. The advantages of this reconstructed hyperspectral cube are both its independence towards the illuminating conditions allowed by the artificial neural network learning phase and the large amount of data generated. The hyperspectral cube obtained with our multispectral imaging system is amenable to further processing such as classical 2D segmentation algorithms on each reflectance plane from the cube, reconstruction of reflectance spectral curves in each pixel, using various signal processing methods to extract information. In clinical practice, this system can be used for objective monitoring of skin diseases, independently from the light source, or quantification of skin chromophores. A 2D analysis of pigmented skin tumours in a wide range of spectral bands can also be performed. Hopefully, this approach might improve the accuracy of their diagnosis