Reconstructions in 3D widefield Diffuse Optical Tomography (DOT) suffer from poor spatial resolution. Therefore, widefield DOT techniques benefit from incorporating structural priors from a complementary modality, such as the micro-CT. Unfortunately, traditional Laplacian-based methods to integrate the priors in the inverse problem are highly time-consuming. Therefore, we propose a Deep Neural Network based end-to-end inverse solver that combines features from AUTOMAP and Z-net and utilizes the micro-CT priors in the training stage. Initial in silico and experimental phantom results demonstrate that the proposed network accurately reconstructs, in 3D, the absorption contrast with a high resolution.
|