An Image Registration Technique for Brain MR Images Using Linear Transform by 3DCNN

In this paper, we focus on brain atrophy in Alzheimer's disease and propose a 3D image registration method of brain MR images for the purpose of calculating the atrophy rate by temporal subtraction technique. The nonlinear 3D image registration model of Stergios et al. is recombined into a linear image registration model, and image registration is performed with isotropic affine and rigid body transformations. In addition, the accuracy is improved by the Instance-specific Optimization, which performs the optimization process for each case. For the training of the model, we used pseudo-data obtained by arbitrarily linear transformation of the ADNI brain MR images. Synthetic data and actual ADNI time series data (AD, MCI) were used for the test. The experimental results show that the addition of the optimization process improves the normalized cross-correlation (NCC) by more than 0.1. The loss function of the model is compared between Grid loss and MSE, and it is confirmed that MSE is effective for NCC. Furthermore, the output images confirm the usefulness of the image registration of the skull in the linear transformation.