Learning a Probabilistic Model for Diffeomorphic Registration

Julian Krebs, Hervé Delingette, Boris Mailhé, Nicholas Ayache, Tommaso Mansi

Abstract: We propose to learn a low-dimensional probabilistic deformation model from data which can be used for registration and the analysis of deformations. The latent variable model maps similar deformations close to each other in an encoding space. It enables to compare deformations, generate normal or pathological deformations for any new image or to transport deformations from one image pair to any other image. Our unsupervised method is based on variational inference. In particular, we use a conditional variational autoencoder (CVAE) network and constrain transformations to be symmetric and diffeomorphic by applying a differentiable exponentiation layer with a symmetric loss function. We also present a formulation that includes spatial regularization such as diffusion-based filters. Additionally, our framework provides multi-scale velocity field estimations. We evaluated our method on 3-D intra-subject registration using 334 cardiac cine-MRIs. On this dataset, our method showed state-of-the-art performance with a mean DICE score of 81.2% and a mean Hausdorff distance of 7.3mm using 32 latent dimensions compared to three state-of-the-art methods while also demonstrating more regular deformation fields. The average time per registration was 0.32s. Besides, we visualized the learned latent space and show that the encoded deformations can be used to transport deformations and to cluster diseases with a classification accuracy of 83% after applying a linear projection.

Introduction

Medical image registration—the process of finding voxel-to-voxel correspondences between two images—is a cornerstone of clinical analysis. Traditionally, this is solved via slow, iterative numerical optimization. While recent Deep Learning (DL) methods have significantly accelerated this process, they face several hurdles: the lack of "ground truth" voxel mappings for training, the absence of a structured way to compare deformations across populations, and the risk of physically impossible "folding" in the resulting deformation fields.

This paper introduces an unsupervised deep learning framework that learns a low-dimensional, probabilistic "map" of how images deform. By combining a Conditional Variational Autoencoder (CVAE) with diffeomorphic constraints, the model achieves state-of-the-art 3D cardiac registration in a fraction of a second while enabling advanced tasks like disease clustering and "deformation transport" between different patients.

Methods

The proposed architecture utilizes a Conditional Variational Autoencoder (CVAE) to learn deformations in an unsupervised manner:

  • Probabilistic Latent Space: The encoder maps the difference between two images into a low-dimensional vector ($z$). The decoder is "conditioned" on the moving image ($M$), forcing the latent vector to encode only the motion rather than the image's appearance.
  • Diffeomorphic Constraints: To ensure the deformations are smooth and invertible, the model predicts a Stationary Velocity Field (SVF). This field is passed through a differentiable "exponentiation layer" to produce the final transformation.
  • Multi-Scale Architecture: The model estimates deformations at three different resolutions (full, middle, and coarse scales), capturing both large-scale global motion and fine-grained local details.
  • Unsupervised Learning: The network is trained by minimizing a similarity loss (Symmetric Local Cross-Correlation) between the warped moving image and the fixed target image.
The proposed CVAE architecture featuring an encoder-decoder structure with multi-scale velocity field estimation and diffeomorphic constraints.

Results and Clinical Utility

The model was evaluated on 3D intra-subject registration using 334 cardiac cine-MRI sequences, including healthy subjects and patients with four types of cardiac disease (DCM, HCM, MINF, and RV).

Performance Metrics

The method achieved a mean DICE score of 81.2% and a mean Hausdorff distance of 7.3mm, outperforming or matching state-of-the-art methods like SyN and VoxelMorph. Registration was remarkably fast, taking only 0.32 seconds on a modern GPU. Furthermore, the deformations produced were significantly smoother and more "regular" than other DL-based approaches, as measured by the gradient of the Jacobian determinant.

Quantitative evaluation comparing the proposed method (Our S1 and S3) against baselines across RMSE, DICE, and Hausdorff Distance.

Latent Space Insights

By compressing motion into a 32-dimensional vector, the model creates a clinical "feature space." A simple linear projection of these vectors classified five cardiac disease states with 83% accuracy, suggesting the model successfully discovered the mechanical signatures of these pathologies. This structured space also allows for deformation transport, where the "contraction signature" of a diseased heart can be applied to a healthy heart's geometry to simulate pathology in a new patient.

Discussion

This framework demonstrates that deformation can be treated as a structured language. By decoupling appearance from motion, the model provides both high-speed registration and a tool for population-level anatomical analysis. Future work may explore more complex latent priors, such as Gaussian mixtures, to better represent distinct disease clusters, and further investigate the probabilistic model's capacity for uncertainty quantification in clinical settings.

Summary updated

Abstract

We propose to learn a low-dimensional probabilistic deformation model from data which can be used for registration and the analysis of deformations. The latent variable model maps similar deformations close to each other in an encoding space. It enables to compare deformations, generate normal or pathological deformations for any new image or to transport deformations from one image pair to any other image. Our unsupervised method is based on variational inference. In particular, we use a conditional variational autoencoder (CVAE) network and constrain transformations to be symmetric and diffeomorphic by applying a differentiable exponentiation layer with a symmetric loss function. We also present a formulation that includes spatial regularization such as diffusion-based filters. Additionally, our framework provides multi-scale velocity field estimations. We evaluated our method on 3-D intra-subject registration using 334 cardiac cine-MRIs. On this dataset, our method showed state-of-the-art performance with a mean DICE score of 81.2% and a mean Hausdorff distance of 7.3mm using 32 latent dimensions compared to three state-of-the-art methods while also demonstrating more regular deformation fields. The average time per registration was 0.32s. Besides, we visualized the learned latent space and show that the encoded deformations can be used to transport deformations and to cluster diseases with a classification accuracy of 83% after applying a linear projection.
BibTeX
@article{krebs2019learning,
  author = {Julian Krebs and Hervé Delingette and Boris Mailhé and Nicholas Ayache and Tommaso Mansi},
  title = {{Learning a Probabilistic Model for Diffeomorphic Registration}},
  journal = {IEEE Trans. Medical Imaging},
  year = {2019},
  url = {https://openreview.net/forum?id=ESW1gpZ7ON},
}
Figure