HyperMorph: Amortized Hyperparameter Learning for Image Registration

Hoopes, Andrew; Hoffmann, M.; Fischl, Bruce; Guttag, John V.; Dalca, Adrian V.

doi:10.1007/978-3-030-78191-0_1

Cited by 85 publications

(45 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this framework, the parameters of the hypernetwork, and not the main network, are learned. While originally introduced for achieving weightsharing and model compression [25], this idea has found numerous applications including neural architecture search [26], [27], Bayesian neural networks [28], [29], multi-task learning [30]- [34], and hyperparameter optimization [35], [36].…”

Section: Hypernetworkmentioning

confidence: 99%

“…al. use this method to enable test-time rapid tunability in image registration for medical imaging [36]. Our previous conference work applied this idea to the AO setting for CS-MRI [9].…”

Section: Related Workmentioning

confidence: 99%

“…When Eq. ( 2) is derived from the maximum a posteriori (MAP) estimate, the scaling factors can be computed with respect to the parameters of the likelihood and/or prior distributions [36], [38]. However, in cases where the regularization term cannot be expressed as a valid probability or when multiple terms are used, such an approximation cannot be obtained.…”

Section: Matched Loss Output Scalesmentioning

confidence: 99%

See 2 more Smart Citations

Computing Multiple Image Reconstructions with a Single Hypernetwork

Wang¹,

Dalca²,

Sabuncu³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Deep learning based techniques achieve state-of-the-art results in a wide range of image reconstruction tasks like compressed sensing. These methods almost always have hyperparameters, such as the weight coefficients that balance the different terms in the optimized loss function. The typical approach is to train the model for a hyperparameter setting determined with some empirical or theoretical justification. Thus, at inference time, the model can only compute reconstructions corresponding to the pre-determined hyperparameter values.In this work, we present a hypernetwork based approach, called HyperRecon, to train reconstruction models that are agnostic to hyperparameter settings. At inference time, Hy-perRecon can efficiently produce diverse reconstructions, which would each correspond to different hyperparameter values. In this framework, the user is empowered to select the most useful output(s) based on their own judgement. We demonstrate our method in compressed sensing, super-resolution and denoising tasks, using two largescale and publicly-available MRI datasets. Our code is available at https://github.com/alanqrwang/hyperrecon.

show abstract

Section: Hypernetworkmentioning

confidence: 99%

“…al. use this method to enable test-time rapid tunability in image registration for medical imaging [36]. Our previous conference work applied this idea to the AO setting for CS-MRI [9].…”

Section: Related Workmentioning

confidence: 99%

Section: Matched Loss Output Scalesmentioning

confidence: 99%

See 1 more Smart Citation

Computing Multiple Image Reconstructions with a Single Hypernetwork

Wang¹,

Dalca²,

Sabuncu³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…We obtain the preprocessed dataset from [38]. The MRI scans were preprocessed [39] using Freesurfer [40] by standard steps like resampling, bias correction, skull stripping, affine normalization and center cropping into volumes of 160 × 192 × 224. For our experiments we split the dataset into training, validation and test set of sizes 255, 15 and 144 respectively.…”

Section: A Datamentioning

confidence: 99%

Towards Positive Jacobian: Learn to Postprocess Diffeomorphic Image Registration with Matrix Exponential

Pal¹,

Tennant²,

Ray³

2022

Preprint

View full text Add to dashboard Cite

“…Parallel to our work, Hoopes et al [13] propose to learn the effects of registration hyperparameters on deformation field with Hypernetworks [10], which leverage a secondary network to generate the conditioned weights for the entire network layers. While the Hypernetworks-based method offers immense modulation potential, it adds an enormous number of parameters to the original image registration method.…”

Section: Introductionmentioning

confidence: 99%

Conditional Deformable Image Registration with Convolutional Neural Network

Mok¹,

Chung²

2021

Preprint

View full text Add to dashboard Cite

Recent deep learning-based methods have shown promising results and runtime advantages in deformable image registration. However, analyzing the effects of hyperparameters and searching for optimal regularization parameters prove to be too prohibitive in deep learningbased methods. This is because it involves training a substantial number of separate models with distinct hyperparameter values. In this paper, we propose a conditional image registration method and a new selfsupervised learning paradigm for deep deformable image registration. By learning the conditional features that are correlated with the regularization hyperparameter, we demonstrate that optimal solutions with arbitrary hyperparameters can be captured by a single deep convolutional neural network. In addition, the smoothness of the resulting deformation field can be manipulated with arbitrary strength of smoothness regularization during inference. Extensive experiments on a large-scale brain MRI dataset show that our proposed method enables the precise control of the smoothness of the deformation field without sacrificing the runtime advantage or registration accuracy.

show abstract

HyperMorph: Amortized Hyperparameter Learning for Image Registration

Cited by 85 publications

References 29 publications

Computing Multiple Image Reconstructions with a Single Hypernetwork

Computing Multiple Image Reconstructions with a Single Hypernetwork

Towards Positive Jacobian: Learn to Postprocess Diffeomorphic Image Registration with Matrix Exponential

Conditional Deformable Image Registration with Convolutional Neural Network

Contact Info

Product

Resources

About