Uncertainty quantification in non-rigid image registration via stochastic gradient Markov chain Monte Carlo

Daniel GrzechImperial College London, UK, Mohammad Farid AzampourImperial College London, UK
Technische Universität München, Germany
Sharif University of Technology, Tehran, Iran
, Huaqi QiuImperial College London, UK, Ben GlockerImperial College London, UK, Bernhard KainzImperial College London, UK
Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany
, Loïc Le FolgocImperial College London, UK
UNSURE2020 special issue
Publication date: 2021/10/27
PDF · arXiv · Code

Abstract

We develop a new Bayesian model for non-rigid registration of three-dimensional medical images, with a focus on uncertainty quantification. Probabilistic registration of large images with calibrated uncertainty estimates is difficult for both computational and modelling reasons. To address the computational issues, we explore connections between the Markov chain Monte Carlo by backpropagation and the variational inference by backpropagation frameworks, in order to efficiently draw samples from the posterior distribution of transformation parameters. To address the modelling issues, we formulate a Bayesian model for image registration that overcomes the existing barriers when using a dense, high-dimensional, and diffeomorphic transformation parametrisation. This results in improved calibration of uncertainty estimates. We compare the model in terms of both image registration accuracy and uncertainty quantification to VoxelMorph, a state-of-the-art image registration model based on deep learning.

Keywords

deformable image registration · uncertainty quantification · SG-MCMC · SGLD

Bibtex @article{melba:2021:016:grzech, title = "Uncertainty quantification in non-rigid image registration via stochastic gradient Markov chain Monte Carlo", authors = "Grzech, Daniel and Azampour, Mohammad Farid and Qiu, Huaqi and Glocker, Ben and Kainz, Bernhard and Le Folgoc, Loïc", journal = "Machine Learning for Biomedical Imaging", volume = "1", issue = "UNSURE2020 special issue", year = "2021" }

2021:016 cover