skip to content

Researcher: Ander Biguri

The field of tomographic reconstruction with data-driven techniques has exploded in the last years. However, many of the research ends up often scattered in repositories around the internet and it is not rare to lack the required information for full reproducibility of experiments and code when trying to replicate other peoples research. We ourselves have done this. To improve that, and with the hopes of easing the translation of these methods into medicine, a big software package called LION (Learned Iterative Optimization Networks) is being developed at CIA. This package includes implementations of all types of data-driven recon methods, including post-processing methods, iterative unrolled methods, learned regularizers, Plug and Play methods and unsupervised/self-supervised methods. The tool also includes useful model functionalities for training and defining models, pre-defined CT experiments, CT operators, datasets (download, pre-processing, loaders) and codes to ensure long term reproducibility of saved models. Learn more at and become a contributor!