Strategies for Robust Deep Learning Based Deformable Registration
This work addresses a critical limitation in medical imaging registration for researchers and practitioners, though it appears incremental by building on existing methods.
The paper tackles the poor generalization of deep learning-based deformable registration beyond training data by proposing a simple method that transforms images into MIND feature space and uses an ensembling strategy, achieving significant robustness improvements as evaluated in the LUMIR brain registration challenge.
Deep learning based deformable registration methods have become popular in recent years. However, their ability to generalize beyond training data distribution can be poor, significantly hindering their usability. LUMIR brain registration challenge for Learn2Reg 2025 aims to advance the field by evaluating the performance of the registration on contrasts and modalities different from those included in the training set. Here we describe our submission to the challenge, which proposes a very simple idea for significantly improving robustness by transforming the images into MIND feature space before feeding them into the model. In addition, a special ensembling strategy is proposed that shows a small but consistent improvement.