Fast Predictive Image Registration
This addresses the computational bottleneck in medical image registration for researchers and clinicians, offering a fast and accurate method with uncertainty estimation, though it is incremental as it builds on existing LDDMM models.
The paper tackles the problem of slow image registration by predicting deformations using a patch-based deep encoder-decoder network, achieving a 1500x speedup in 2D and 66x in 3D compared to GPU-based optimization while maintaining diffeomorphic transformations and better accuracy than predicting deformation or velocity fields.
We present a method to predict image deformations based on patch-wise image appearance. Specifically, we design a patch-based deep encoder-decoder network which learns the pixel/voxel-wise mapping between image appearance and registration parameters. Our approach can predict general deformation parameterizations, however, we focus on the large deformation diffeomorphic metric mapping (LDDMM) registration model. By predicting the LDDMM momentum-parameterization we retain the desirable theoretical properties of LDDMM, while reducing computation time by orders of magnitude: combined with patch pruning, we achieve a 1500x/66x speed up compared to GPU-based optimization for 2D/3D image registration. Our approach has better prediction accuracy than predicting deformation or velocity fields and results in diffeomorphic transformations. Additionally, we create a Bayesian probabilistic version of our network, which allows evaluation of deformation field uncertainty through Monte Carlo sampling using dropout at test time. We show that deformation uncertainty highlights areas of ambiguous deformations. We test our method on the OASIS brain image dataset in 2D and 3D.