Probabilistic model improves medical image alignment and uncertainty measure
BINDER: A Latent Variable Model for Probabilistic Medical Image Registration
Computer Vision and Pattern Recognition
Summary
Matching medical images taken in different ways or at different times is challenging because doctors need to know which parts correspond to each other. This paper presents a new method that treats the matching process as a guessing game, where hidden connections between tiny parts of the images are estimated. The authors developed fast techniques to find these connections and to measure how confident the method is about the matches, even in complex three-dimensional cases. Their approach works well on many kinds of images without needing extra tuning and is freely available for anyone to use.
What this means in practice
- •For medical imaging teams: Automatically align different medical scans with robust and quantifiable confidence to support diagnosis and treatment planning across various imaging types.
- •For medical device developers: Incorporate the model's uncertainty estimates into imaging tools to enhance multimodal image fusion and navigation systems in clinical settings.$Commercial implications: Enables development of advanced imaging products with reliable registration and uncertainty reporting for hospitals and clinics.
Authors
Stefano Cerri, Amirhossein Hassankhani, Yaël Balbastre, Koen Van Leemput
Abstract
We propose a new probabilistic model for general-purpose medical image registration that builds upon the mutual information registration criterion. It centers around a spatial interpolation technique that assumes latent voxel-wise correspondences between the images being registered. By exploiting these latent variables, we derive dedicated optimization and MCMC sampling techniques that only involve closed-form iterative updates. When applied to nonlinear registration, an efficient demons-like optimization algorithm is obtained that shows robust out-of-the-box performance across a variety of monomodal and multimodal registration tasks. We also demonstrate a corresponding sampler that can quantify, for the first time, uncertainty in multimodal registration scenarios with very high-dimensional 3D deformations. Our code, which we call BINDER (Bayesian INference for DEformable Registration), is freely available at https://github.com/ste93ste/BINDER.