Determination of Oxy and Deoxyhemoglobin Concentrations in Skin Tissue Using Spatial Frequency Domain Imaging and Artificial Neural Network

Spatial frequency domain imaging is an emerging technology that enables rapid, wide-field, and non-invasive chromophore mapping. Essentially, in this technique, a large area of the tissue is illuminated with a spatially modulated light field. The light beam reflected by the tissue depends on its optical properties, so that it can provide information about tissue composition through different chromophore concentrations. In this work, we employed a combination of Principal Component Analysis and Artificial Neural Networks to directly determine oxyhemoglobin and deoxyhemoglobin concentrations in skin tissue from diffuse reflectance values obtained from spatial frequency domain imaging. The database consisted of 850500 samples computed from the Beer’s law and Monte Carlo simulations, and it was divided into training, validation, and testing subsets in a 0.7:0.15:0.15 ratio. To reduce overfitting during the network training, Bayesian regularization, based on the Levenberg-Marquardt optimization, was employed. Results showed that the developed model predict values of oxy and deoxyhemoglobin concentrations with a correlation coefficient of 0.997 and 0.982, respectively. The average errors from the expected values were 0.98% and 0.99%, for oxy and deoxyhemoglobin, respectively, with most of the samples showing absolute errors lesser than 4%. The developed model was applied to an in vivo study to determine hemoglobin concentrations in the hand of a volunteer. Results indicate that the developed model provides good performance in determining the oxyhemoglobin and deoxyhemoglobin concentrations, and it can be easily applied to in vivo measurements, with the potential to aid in the diagnosis of vascular changes in skin tissue.