Deep Learning--Based Dictionary Learning and Tomographic Image Reconstruction

This work presents an approach for image reconstruction in clinical low-dose tomography that combines principles from sparse signal processing with ideas deep learning. First, we describe representation terms of dictionaries a statistical perspective and interpret dictionary learning as process aligning the distribution arises generative model empirical true signals. As result, can see coding learned resembles specific variational autoencoder, where encoder is algorithm decoder linear function. Next, show also benefit computational advancements introduced context learning, such parallelism stochastic optimization. Finally, regularization by achieves competitive performance computed compared to state-of-the-art model-based data-driven approaches, while being unsupervised respect tomographic data.