Table 4.

A list of recent papers related to medical image registration

Author	Year	Application	Model	Dataset	Contributions highlights
Supervised registration
Haskins et al., 2019	2019	3D MR–TRUS prostate image registration	CNN-based network with a skip connection	Private dataset	(1) Using the designed CNN to learn a similarity metric for rigid registration; (2) Proposing a new strategy to perform the optimization.
Cheng et al., 2018	2018	2D CT-MR patches registration	FCN pre-trained with stacked denoising AE	Private dataset	Learning a metric via FCN to evaluate the similarity between 2D CT-MR image patches for deformable registration.
Simonovsky et al., 2016	2016	Registration of T1 and T2-weighted MRI scans	5-layer CNN	ALBERTs	Learning a metric via CNN to evaluate the similarity between aligned 3D brain MRI T1–T2 image pairs for deformable registration.
Yang et al., 2017	2017	Atlas-to-image and image-to-image registration	A deep encoder-decoder network	OASIS, IBIS 3D Autism Brain dataset	(1) Using deep nets to predict the momentum-parameterization of LDDMM; (2) A probabilistic version of the prediction network was developed to calculate uncertainties in the predicted deformations.
Fan et al., 2019a	2019	Brain MR image registration	BIRNet: hierarchical dual-supervised FCN	LPBA40, IBSR18, CUMC12, IXI30	Providing coarse guidance (pre-registered ground-truth deformation field) and fine guidance (similarity metric) to refine the registration results.
Sokooti et al., 2017	2017	3D chest CT image registration	RegNet: a new CNN-based architecture	Private dataset	(1) Training the model using artificially generated DVFs without defining a similarity metric; (2) Incorporating contextual information into the network by processing input 3D image patches at at multiple scales.
Unsupervised registration
Zhao et al., 2019b	2019	3D liver CT image registration	VTN: several cascaded subnetworks	Private data, LITS, MICCAI’07 challenge	(1) Cascading the registration subnetworks to achieve better performance in registering largely displaced images; (2) Proposing invertibility loss for better accuracy.
Kim et al., 2019	2019	3D multiphase liver CT image registration	Based on VoxelMorph (Balakrishnan et al., 2018)	Private dataset	Performing unsupervised registration with cycle-consistency (Zhu et al., 2017).
Balakrishnan et al., 2018	2018	3D brain MRI registration	VoxelMorph: UNet-based network and STN	8 public datasets (e.g. ADNI)	Formulating 3D image registration as a parametric function solving it without requiring supervised information.
Balakrishnan et al., 2019	2019	3D brain MRI registration	An extension of VoxelMorph	8 public datasets (e.g. ADNI)	Extending VoxelMorph by leveraging auxiliary segmentation information (anatomical segmentation maps).
de Vos et al., 2017	2017	2D cardiac cine MR image registration	DIRNet: ConvNet and STN	Sunnybrook Cardiac Data	The first deep learning-based framework for end-to-end unsupervised deformable image registration.
de Vos et al., 2019	2019	3D cardiac cine MRI and chest CT registration	DLIR: stack of multiple CNNs	Sunnybrook Cardiac Data, NLST, etc.	(1) Extending DIRNet to 3D scenarios; (2) Introducing a multi-stage registration architecture by stacking multiple CNNs.
Fan et al., 2019b	2019	3D brain MRI and multi-modal CT-MR pelvic image registration	GAN-based registration framework	LPBA40, IBSR18, CUMC12, MGH10, and private data	(1) Using the discriminator of GAN to implicitly learn an adversarial similarity to determine the voxel-to-voxel correspondence; (2) The proposed framework applies to both mono-modal and multi-modal registration.