Determinação da concentração de oxi e deoxihemoglobina a partir de imagens de luz difusa utilizando redes neurais artificiais e sfdi

Abstract

Spatial frequency domain imaging (SFDI) is a technology that allows obtaining chromophore maps quickly, noninvasively and in a wide field. This method consists of illuminating a large area of the tissue with a spatially modulated light field. The light reflected by the tissue varies according to its optical properties, allowing its composition to be identified. In this work, we used a combination of Principal Component Analysis (PCA) and Artificial Neural Networks (ANN) to compute the concentrations of oxy and deoxyhemoglobin in skin tissues, from diffuse reflectance values obtained by SFDI. The database used was generated according to the parameters of human tissue and Monte Carlo simulations totaling 850,500 samples - distributed in training, test and validation sets in the proportion 70:15:15. To minimize the risk of overfitting during the training of the neural network, Bayesian regularization was applied, based on the Levenberg-Marquardt optimization. The results demonstrated that the developed model was able to predict the concentrations of oxyhemoglobin and deoxyhemoglobin with correlation coefficients of 0.997 and 0.982, respectively. The mean relative errors in relation to the expected values were 0.98% for oxyhemoglobin and 0.99% for deoxyhemoglobin, with 90% of the samples presenting mean relative errors below 4%. The model was also applied to an in vivo study to determine the hemoglobin concentrations in the hand of a volunteer. The results show that the developed model performs well in determining the concentrations of these chromophores, proving to be a promising tool for in vivo measurements and with potential to aid in diagnosis.