Domain-Shift Immunity in Deep Deformable Registration via Local Feature Representations

TL;DR

This paper demonstrates that deep deformable registration models inherently possess domain-shift immunity due to their reliance on local feature representations, and introduces UniReg, a framework that achieves cross-domain robustness with minimal training data.

Contribution

The paper reveals the mechanism behind domain-shift immunity in deep registration models and proposes UniReg, a universal framework that maintains robustness across domains using fixed feature extractors.

Findings

UniReg performs well across different domains and modalities.

Local feature representations are key to robustness in deformable registration.

Failures in CNN models under modality shift are due to biases in early layers.

Abstract

Deep learning has advanced deformable image registration, surpassing traditional optimization-based methods in both accuracy and efficiency. However, learning-based models are widely believed to be sensitive to domain shift, with robustness typically pursued through large and diverse training datasets, without explaining the underlying mechanisms. In this work, we show that domain-shift immunity is an inherent property of deep deformable registration models, arising from their reliance on local feature representations rather than global appearance for deformation estimation. To isolate and validate this mechanism, we introduce UniReg, a universal registration framework that decouples feature extraction from deformation estimation using fixed, pre-trained feature extractors and a UNet-based deformation network. Despite training on a single dataset, UniReg exhibits robust cross-domain and multi-modal performance comparable to optimization-based methods. Our analysis further reveals that failures of conventional CNN-based models under modality shift originate from dataset-induced biases in early convolutional layers. These findings identify local feature consistency as the key driver of robustness in learning-based deformable registration and motivate backbone designs that preserve domain-invariant local features.