Fluorescence Molecular Tomography Reconstruction of Small Targets Using Stacked Auto-Encoder Neural Networks

Abstract

As a noninvasive and quantitative method, fluorescence molecular tomography (FMT) has many potential applications in biomedical field. It has the power to resolve in three-dimension (3D), the molecular processes in small animal in-vivo in both theory and practice. This paper proposes to solve the problem of reconstruction error and speed by using stacked auto-encoders (SAE). A finite element method (FEM) solution to the Laplace transformed time-domain coupled diffusion equations is employed as the forward model. The reconstruction model is formulated under the framework of SAE. Numerical simulation experiments were conducted to compare the reconstruction results of SAE and algebraic reconstruction technique (ART). We demonstrated that the proposed reconstruction algorithm can retrieve the positions and shapes of the targets more accurately than ART. This advantage of SAE is especially reflected in the reconstruction for small targets with a radius of 2 mm and 3 mm.