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

Automated summary

This work introduces an approach for image reconstruction in clinical low-dose tomography by combining principles from sparse signal processing with ideas from deep learning. Firstly, it describes sparse signal representation from a statistical perspective and interprets dictionary learning as aligning the distribution generated by a model with the empirical distribution of true signals. The work also shows that dictionary learning can benefit from computational advancements in deep learning. Furthermore, it demonstrates that regularization with dictionaries achieves competitive performance in computed tomography reconstruction.