Reconstruction hyperspectral reflectance cube based on artificial neural networks for multispectral imaging system applied to dermatology

The multispectral imaging (MSI) technique has been used for skin analysis, especially for distant mapping of invivo skin chromophores. We have successfully developed an MSI system with a new approach. Our MSI system captures 11 mono-spectral images of human skin which is too little for providing an accurate diagnostic information. We need something to reconstruct the 11 monoband data sets to the wider range hyperspectral data sets. In this paper, we proposed a method to build a hyperspectral reflectance cube based on artificial neural network (ANN) algorithm. ANN is trained using the 32 natural color from X-Rite Color Checker Passport. The learning procedure the involves acquisition, by a spectrometer. This neural network is then used to retrieve a hyperspectral reflectance cube between 380 and 880 nm with a 5 nm resolution. To evaluate the performance of reconstruction, we used the Goodness of Fit Coefficient (GFC) and Root Mean Squared Error (RMSE). The reconstruction results are very good. The average GFC was 0,9988 and the average RMSE was 0.023. We also tested the quality of reconstruction with healthy skin data sets and the results are good enough. For skin data sets, the average GFC was 0.9855 and the average RMSE was 0.0608.