Generalize virtual decimation to multimodal registration

Determine how to generalize the virtual decimation factor used to compensate for misspecified conditional-independence assumptions in probabilistic image-registration likelihoods to multimodal registration settings.

Background

Virtual decimation is used in some probabilistic registration methods to down-weight the log-likelihood and compensate for misspecification of the forward model, particularly the assumption that fixed-image voxel intensities are conditionally independent given the moving image. The paper notes that existing formulations rely on strong assumptions, including monomodal registration and spatial correlations induced by Gaussian blurring. Consequently, extending this correction to multimodal registration remains unresolved.

References

However, this technique rests on strong assumptions itself (in particular, that the registration is monomodal, with spatial correlations arising from blurring with a Gaussian kernel), and it is currently unclear how it could be generalized to multimodal settings.

— BINDER: A Latent Variable Model for Probabilistic Medical Image Registration  (2609.19875 - Cerri et al., 17 Sep 2026) in Section Discussion and Conclusion