Convolutional deep network for light propagation in heterogeneous bio-tissues

摘要

To calculate the light propagation through heterogeneous bio-tissues, we propose a convolutional deep network withspecified convolutional kernels and calculation rules. The form of convolution kernels is chosen to be capable to imitatethe absorption and scattering event by convolution operation. In the meanwhile, the multi kernel and maskingmechanism we set provide the capability to calculate propagation in voxelized heterogeneous bio-tissue structures withpre-set tissue types and specific optical properties. The two-dimensional convolution operations are carried out multipletimes until all the photons leaves the structure or absorbed by the tissues. Application of our network with kernels in theform of semi-infinite homogeneous radiative transfer equation (RTE) solution, and semi-infinite homogeneous diffusionequation (DE) solution are implemented to three artificial manipulated structures, including homogeneous phantom,two-layer structure and two-layer structure with a third tissue type inside layer two. The result comparing to Monte-Carlosimulation reveals the potential to form a new forward calculation model for diffuse optical tomography.