Published in the Journal of Open Source Software November 2020, DeepReg: A deep learning toolkit for medical image registration is a product of a close collaboration between InstaDeep and world-leading research institutes University College London, King’s College London, and Massachusetts Institute of Technology. The work was led by Yunguan Fu, Research Engineer at InstaDeep.

Deep Learning based image registration

It was Dr Yipeng Hu, Co-author and Lecturer at UCL, who first got the idea behind the new open-source toolkit after identifying a gap in the current open-source toolkit offering. “We saw a real need in the community for a tool like DeepReg as other popular packages today do not support Deep Learning based image registration. Rather, they implement specific published algorithms without automated testing or focus on classical methods, while DeepReg focuses on image registration using Deep Learning. This allows professors to teach registration and TensorFlow coding through a platform that enables users to try new ideas with their own dataset, says Fu. 

Fu et al., (2020). DeepReg: a deep learning toolkit for medical image registration. Journal of Open Source Software, 5(55), 2705,

Advancements in Deep Learning in the field of medical imaging have been going from strength to strength over recent years, specifically due to its ability to push the limits on what is possible and its ability to learn from populated data through Deep Neural Networks. “We are proud to be part of challenging the status quo in algorithmic medical imaging through DeepReg. Our proven expertise in Deep Learning and knowledge of healthcare science was perfectly complemented by the prowess of our research partners resulting in a new toolkit offering which continues to improve training efficiency. This makes the process easier, especially for people who might not be familiar with deep learning or registration”, says InstaDeep’s CEO Karim Beguir. 

Real-life applications

The Python packaged toolkit implements a range of major unsupervised and weakly-supervised learning-based registration algorithms using TensorFlow, alongside a set of predefined dataset loaders, supporting both labelled and unlabelled data. Together with flexible APIs, DeepReg facilitates the application of deep learning-based registration on custom data sets allowing the user to install it locally and deploy training and predictions on their machines. 

Through a variety of examples from diverse clinical domains, DeepReg is highly applicable to real-life environments as it provides tutorials from clinical domains and demonstrates easy and efficient deployment on custom problems. The open-source software tool is now available to use on GitHub.


For more information, get in touch via