Fast Predictive Image Registration

By Xiao Yang, Roland Kwitt, Marc Niethammer. 2016

128 citations recorded by the source corpus.

We present a method to predict image deformations based on patch-wise image appearance. Specifically, we design a patch-based deep encoder-decoder network which learns the pixel/voxel-wise mapping between image appearance and registration parameters. Our approach can predict general deformation parameterizations, however, we focus on the large deformation diffeomorphic metric mapping (LDDMM) registration model. By predicting the LDDMM momentum-parameterization we retain the desirable theoretical properties of LDDMM, while reducing computation time by orders of magnitude: combined with patch pruning, we achieve a \(1500\mathsf {x}\)/\(66\mathsf {x}\) speed-up compared to GPU-based optimization for 2D/3D image registration. Our approach has better prediction accuracy than predicting deformation or velocity fields and results in diffeomorphic transformations. Additionally, we create a Bayesian probabilistic version of our network, which allows evaluation of deformation field uncertainty through Monte Carlo sampling using dropout at test time. We show that deformation uncertainty highlights areas of ambiguous deformations. We test our method on the OASIS brain image dataset in 2D and 3D.

Subfields: Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Dropout Inference in Bayesian Neural Networks with Alpha-divergences (2017, methodology): “These have been used for machine vision and medical applications (Kendall & Cipolla, 2016; Kendall et al., 2015; Angermueller & Stegle, 2015; Yang et al., 2016).”
Slice-to-volume medical image registration: a survey (2017, unknown): “…flow estimation using CNNs (where not only the similarity measure (Simonovsky et al., 2016; Zagoruyko et al., 2015) but also the actual registration process (Fischer et al., 2015; Miao et al., 2016; Yang et al., 2016) is learned from examples) suggest that these problems are not an exception.”
Landmark-based deep multi-instance learning for brain disease diagnosis (2017, methodology): “However, such method requires a non-linear registration process which may generate inaccurate registration results ( Miao et al., 2016; Yang et al., 2016; Cao et al., 2017 ).”
Quantifying total uncertainty in physics-informed neural networks for solving forward and inverse stochastic problems (2018, methodology): “Moreover, dropout does not induce much computation overhead and thus has been used as a practical tool to obtain uncertainty estimation eﬀectively in real applications including language modelling [44], computer vision [45, 46] and medical applications [47, 48].”
Robust Non-rigid Registration Through Agent-Based Action Learning (2017, background): “These approaches either only partially remove the above-mentioned limitations as they stick to an energy minimization framework (cf. [8]) or rely on a large number of training samples derived from existing registration results (cf. [1,10]).”
Unsupervised End-to-end Learning for Deformable Medical Image Registration (2017, methodology): “[18] design a deep encoder-decoder network to initialize the momentum of the large deformation diffeomorphic metric mapping (LDDMM) registration model.”
Quicksilver: Fast Predictive Image Registration – a Deep Learning Approach (2017, methodology): “The registration method described here, which we name Quicksilver, is an extension of preliminary ideas we presented in a recent workshop paper [21].”
SVF-Net: Learning Deformable Image Registration Using Shape Matching (2017, unknown): “In [11], the prediction is trained based on ground truth provided by a registration algorithm previously run on pair of images.”
A review of multimodal image matching: Methods and applications (2021, methodology): “o using existing labeled datasets of real scenarios [95,96], many data”
Nonrigid Image Registration Using Multi-scale 3D Convolutional Neural Networks (2017, background): “[4] predicted the initial momentum of a 3D LDDMM registration.”