ADMIR–Affine and Deformable Medical Image Registration for Drug-Addicted Brain Images

Abstract

We proposed an unsupervised end-to-end Affine and Deformable Medical Image Registration (ADMIR) method based on convolutional neural network (ConvNet). ADMIR includes three key components: an affine registration module for learning the affine transformation parameters, a deformable registration module for learning the displacement vector field, and a spatial transformer for getting the final warped image from both affine and deformable transformation parameters. To evaluate its performance, the magnetic resonance images of drug-addicted brains were used to train and test the model, and we compared it with two state-of-art methods in terms of Dice score, Hausdorff distance, and average symmetric surface distance. The experimental results demonstrated that our proposed ADMIR model outperforms existing methods even with the images without pre-alignment, which suggests that the ADMIR model can be used to achieve quick medical image registration with high accuracy. INDEX TERMS Deformable image registration, affine registration, convolutional neural network, unsupervised learning, end-to-end registration.

MICDIR: Multi-scale Inverse-consistent Deformable Image Registration using UNetMSS with Self-Constructing Graph Latent

“…The performance of the proposed method was compared quantitatively and qualitatively against three deep learning baseline models: ICNet Zhang (2018), Voxelmorph Balakrishnan et al (2019, and ADMIR Tang et al (2020), direct optimisation of the deformation field using the RMSPROP optimiser, and the gold standard -ANTS SyN registration.…”

“…Voxelmoroph (Balakrishnan et al, 2019) works with concatenated fixed and moving images and predicts the DVF of the given image pair, which is then applied on the moving image using a spatial transformer. ADMIR (Tang et al, 2020) -Affine and Deformable Medical Image Registration is an end-to-end method for affine and deformable image registration utilising CNNs. This method does not require the images to be pre-aligned, which in turn helps to do image registration quickly with good accuracy.…”

An optimal self adaptive deep neural network and spine‐kernelled chirplet transform for image registration

“…Experiments show that the technique is 880× quicker than conventional optimization‐based methods in medical image registration and reaches standard performance. Tang et al 20 presented an Affine and Deformable Medical Image Registration (ADMIR) approach based on a CNN. ADMIR consists of three major components such as a learning registration module that is deformable to the vector field of displacement, from deformable and affine transformation parameters the warped image of the spatial transformer was obtained, and a learning module based on affine registration of the parameters of the affine transformation.…”

Preliminary Feasibility Study of Imaging Registration Between Supine and Prone Breast CT in Breast Cancer Radiotherapy Using Residual Recursive Cascaded Networks

“…Yu et al [28] proposed a 3dimensional (3D) unsupervised network based on a metabolic constraint function and a multi-domain similarity measure for 3D Positron Emission Tomography (PET) and CT image registration. Tang et al [29] proposed an unsupervised end-to-end network which includes an affine registration module and a deformable registration module. Fu et al [30] proposed an unsupervised end-to-end network for four-dimensional (4D) lung images named LungRegNet which also consists of two subnetworks named CoarseNet and FineNet.…”