Deep learning neural network estimation of tissue oxygenation based on diffuse optical spectroscopy

Diffuse optical spectroscopy (DOS) is an effective technique for diagnosing the component composition of biological tissues. It is based on the analysis of optical radiation transmitted through the tissue at different wavelengths. Tissues are optically inhomogeneous media, so real spectroscopic measurements contain random and systematic errors, and the DOS inverse problem is ill-conditioned, therefore traditional optimization techniques for reconstructing the optical characteristics of the investigated object have a rather high error rate. In this paper we have proposed a predictive method for calculating the tissue oxygenation, based on a deep direct neural network for solving the DOS error problem. The proposed neural network consists of 6 layers and has 141987 parameters. 214220 simulated spectroscopic measurements for various component compositions of tissue were used to train the neural network. Statistical analysis of the calculating accuracy for tissue oxygenation has showed a significant advantage of the developed technique over the traditional optimization approach to solving the DOS problem.