FMT-ReconNet: Fluorescence Molecular Tomography Reconstruction using Prior Knowledge and Deformation Neural Network

Fluorescence molecular tomography (FMT) is a powerful imaging technique for 3D reconstruction of internal fluorescent sources. However, its spatial resolution is limited by a simplified forward model and an ill-posed inverse problem. To address this, we introduce FMT-ReconNet, a deep neural network comprising a spatial transformer network (STN) for source transformation and a V-Net for reconstruction. The STN calculates affine transformation parameters based on template and target FMT surface distributions, transforming the template source into prior knowledge for the target. Concatenating this prior with the target's surface distribution, V-Net predicts and reconstructs the source accurately. FMT-ReconNet represents a significant advancement in precise FMT imaging reconstruction.