Abstract
In this work, we address the problem of pre-registering images before running any registration model. This step is often considered solved by computing a global affine transformation between the fixed and moving images. However, we claim that accurate local pre-registration of images has a major impact on the quality of the final registration. Our approach introduces BlendNet, a shape-aware network that predicts blending weights to combine multiple locally optimized transformations into a single, dense, and spatially smooth transformation. BlendNet leverages as input fixed and moving images, as well as their corresponding segmentation masks, which provide explicit anatomical shape context and are obtained through manual annotation or a foundational segmentation model. BlendNet is trained with local anatomy and shape-aware supervision and regularization, which enforce a physically plausible and accurate pre-registration. The proposed framework is plug-and-play, as it can be used as a pre-processing step before any other registration method. Through extensive experiments, we demonstrate the benefits of BlendNet. Existing registration methods benefit from a systematic performance boost in alignment accuracy with greater improvements observed for high quality segmentation masks. However, this increase in alignment accuracy may be associated with lower deformation regularity, depending on the used registration method. The learned models and code are made available for research purposes.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_025.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=0dhpQ72Nei
BibTex
@InProceedings{RavGab_BlendNet_MICCAISAT2026,
author = { Ravelomanana, Gabriel AND Pansiot, Julien AND Pujades, Sergi},
title = { { BlendNet: on the Importance of Accurate Local Pre-Registration in the Image Registration Task } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}