DeepReg: a deep learning toolkit for medical image registration
Yunguan Fu, Nina Montaña-Brown, Shaheer U. Saeed, Adrià Casamitjana, Zachary M. C. Baum, Rémi Delaunay, Qianye Yang, Alexander Grimwood, Zhe Min, Stefano B. Blumberg, Juan Eugenio Iglesias, Dean C. Barratt, Ester Bonmati, Daniel C. Alexander, Matthew J. Clarkson, Tom Vercauteren, Yipeng Hu
Abstract
Image fusion is a fundamental task in medical image analysis and computer-assisted intervention. Medical image registration, computational algorithms that align different images \ntogether (Hill et al., 2001), has in recent years turned the research attention towards deep \nlearning. Indeed, the representation ability to learn from population data with deep neural \nnetworks has opened new possibilities for improving registration generalisability by mitigating \ndifficulties in designing hand-engineered image features and similarity measures for many realworld clinical applications (Fu et al., 2020; Haskins et al., 2020). In addition, its fast inference can substantially accelerate registration execution for time-critical tasks. \nDeepReg is a Python package using TensorFlow (Abadi et al., 2015) that implements multiple registration algorithms and a set of predefined dataset loaders, supporting both labelledand unlabelled data. DeepReg also provides command-line tool options that enable basic and \nadvanced functionalities for model training, prediction and image warping. These implementations, together with their documentation, tutorials and demos, aim to simplify workflows for prototyping and developing novel methodology, utilising latest development and accessing quality research advances. DeepReg is unit tested and a set of customised contributor \nguidelines are provided to facilitate community contributions. \nA submission to the MICCAI Educational Challenge has utilised the DeepReg code and demos \nto explore the link between classical algorithms and deep-learning-based methods (Montana \nBrown et al., 2020), while a recently published research work investigated temporal changes \nin prostate cancer imaging, by using a longitudinal registration adapted from the DeepReg \ncode (Yang et al., 2020).