Quicksilver: Fast Predictive Image Registration – a Deep Learning Approach

By Xiao Yang, Roland Kwitt, Marc Niethammer. 2017

643 citations recorded by the source corpus.

This paper introduces Quicksilver, a fast deformable image registration method. Quicksilver registration for image-pairs works by patch-wise prediction of a deformation model based directly on image appearance. A deep encoder-decoder network is used as the prediction model. While the prediction strategy is general, we focus on predictions for the Large Deformation Diffeomorphic Metric Mapping (LDDMM) model. Specifically, we predict the momentum-parameterization of LDDMM, which facilitates a patch-wise prediction strategy while maintaining the theoretical properties of LDDMM, such as guaranteed diffeomorphic mappings for sufficiently strong regularization. We also provide a probabilistic version of our prediction network which can be sampled during the testing time to calculate uncertainties in the predicted deformations. Finally, we introduce a new correction network which greatly increases the prediction accuracy of an already existing prediction network. We show experimental results for uni-modal atlas-to-image as well as uni-/multimodal image-to-image registrations. These experiments demonstrate that our method accurately predicts registrations obtained by numerical optimization, is very fast, achieves state-of-the-art registration results on four standard validation datasets, and can jointly learn an image similarity measure. Quicksilver is freely available as an open-source software.

Subfields: Biomedical Engineering, Materials Chemistry, Nuclear and High Energy Physics

How researchers see this paper

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

In Biomedical Engineering

In Materials Chemistry

In Nuclear and High Energy Physics

Citations in the corpus

An overview of deep learning in medical imaging focusing on MRI (2018, methodology): “For example: deformable image registration [97,239]; model-to-image registration [240,241]; MRI-based attenuation correction for PET [242,243]; PET/MRI dose calculation [244]; unsupervised end-to-end learning for deformable registration of 2D CT/MR images [245]; an unsupervised learning model for de”
Unsupervised Learning for Fast Probabilistic Diffeomorphic Registration (2018, background): “which can introduce a bias and necessitate significant preprocessing [23,26,28].”
Deep Learning Applications in Medical Image Analysis (2018, background): “[83] stacked convolution layers in an encoder-decoder fashion, to predict how an input pixel would morph into its final configuration.”
Deep learning in medical image registration: a review (2019, neutral): “[175] Brain 3D-3D Yes MR Deformable [10] Brain 3D-3D Yes MR Deformable”
VoxelMorph: A Learning Framework for Deformable Medical Image Registration (2018, background): “Most of these rely on ground truth warp fields [35]–[39], which are either obtained by simulating deformations and deformed images, or running classical registration methods on pairs of scans.”
A review of deep learning in medical imaging: Imaging traits, technology trends, case studies with progress highlights, and future promises (2020, background): “Still, some unique and varied approaches have achieved state-of-the-art results with relatively fast run times [132]–[135].”
A Gentle Introduction to Deep Learning in Medical Image Processing (2018, methodology): “The quicksilver algorithm is able to model a deformable registration and uses a patch-wise prediction directly from the image appearance [83].”
TransMorph: Transformer for unsupervised medical image registration (2021, neutral): “Cui et al. (2021) and Yang et al. (2017b) incorporated MC dropout layers in their registration network designs, allowing for the estimation of transformation uncertainty by sampling multiple deformation field predictions from the network.”
Recursive Cascaded Networks for Unsupervised Medical Image Registration (2019, methodology): “Some initial works rely on the dense ground-truth flows obtained by either traditional algorithms [14, 56] or simulating intrasubject deformations [36, 53], but their performance is restricted due to the limited quality of training data.”
Deformable Image Registration Using Cue-aware Deep Regression Network (2018, methodology): “Speciﬁ-cally, fully convolutional network (FCN) has been employed in [10] and [51] to learn the deformation momenta and stationary velocity ﬁeld (SVF), respectively.”
Artificial intelligence in radiology (2018, positive): “Studies have shown that deep learning leads to generally more consistent registrations and is an order of magnitude faster than more conventional methods 79 .”
Deep learning in medical imaging and radiation therapy. (2018, methodology): “322 Yang et al. 240 created a synthetic CT dataset through the use of the Radon transform for a known object and modeled different exposure conditions through adding noise to the data, for training a CNN to estimate high-dose projections from low-dose ones.”
An Unsupervised Learning Model for Deformable Medical Image Registration (2018, methodology): “Most of these rely on ground truth warp fields or segmentations [26, 35, 39, 45], a significant drawback compared to our method, which does not require either.”
Deep learning in medical image registration: a survey (2019, methodology): “ make the inference process computationally inefﬁcient. Another method is to use a non-parametric registration model for transformation prediction. One such example is LDDMM model used in QuickSilver [175]. Instead of directly predicting ﬁnal spatial transformation, QuickSilver predict the momentum ”
Secure and Robust Machine Learning for Healthcare: A Survey (2020, background): “In [58], a framework for deformable image registration named”
Real-time Deep Pose Estimation with Geodesic Loss for Image-to-Template Rigid Registration (2018, methodology): “[13] developed a deep autoencoder-decoder convolutional neural network (CNN) that learned to predict the Large Deformation Diffeomorphic Metric Mapping (LDDMM) model, and achieved state-of-the-art performance with an order of magnitude faster optimization in inter-subject and subject-to-atlas deform”
Image Synthesis in Multi-Contrast MRI With Conditional Generative Adversarial Networks (2018, neutral): “first be estimated via multi-modal registration [75].”