ANN Estimations of the Absorption Coefficient in Multi-Region Heterogenous Media: MoC solutions as training data

The estimation of the medium absorption coefficient from external measurements is formulated as an inverse problem, with significant applications in optical medicine, including optical tomography. In this work, we propose a framework based on artificial neural networks (ANNs) to estimate the absorption coefficient in multi-region heterogeneous media. The framework integrates a direct transport problem, modeled through the Method of Characteristics (MoC), to simulate the particle intensity distribution, and a multilayer perceptron (MLP) neural network for the inverse estimation of the absorption coefficient. The medium is divided into cells, and the absorption coefficient is estimated as a piecewise constant function. The ANN is trained using a dataset of detector measurements obtained from the direct problem solutions. A preliminary test case is presented, with specific laser pulse parameters and medium characteristics, demonstrating the viability of the approach. The proposed framework offers a promising alternative for accurate and efficient estimation of the absorption coefficient in complex heterogeneous media.