Reg-TTR, Test-Time Refinement for Fast, Robust and Accurate Image Registration

Traditional image registration methods are robust but slow due to their iterative nature. While deep learning has accelerated inference, it often struggles with domain shifts. Emerging registration foundation models offer a balance of speed and robustness, yet typically cannot match the peak accuracy of specialized models trained on specific datasets. To mitigate this limitation, we propose Reg-TTR, a test-time refinement framework that synergizes the complementary strengths of both deep learning and conventional registration techniques. By refining the predictions of pre-trained models at inference, our method delivers significantly improved registration accuracy at a modest computational cost, requiring only 21% additional inference time (0.56s). We evaluate Reg-TTR on two distinct tasks and show that it achieves state-of-the-art (SOTA) performance while maintaining inference speeds close to previous deep learning methods. As foundation models continue to emerge, our framework offers an efficient strategy to narrow the performance gap between registration foundation models and SOTA methods trained on specialized datasets. The source code will be publicly available following the acceptance of this work.

💡 Research Summary

The paper addresses a long‑standing trade‑off in medical image registration: traditional iterative methods are robust and generalizable but computationally expensive, whereas deep learning approaches offer rapid inference but suffer from domain shift and limited generalization. Recent registration foundation models, trained on large heterogeneous datasets, partially bridge this gap by providing fast and reasonably robust performance across tasks, yet they still fall short of the peak accuracy achieved by task‑specific models. To close this performance gap without sacrificing speed, the authors propose Reg‑TTR (Test‑Time Refinement), a framework that refines the output of a pre‑trained registration network at inference time using a lightweight optimization loop.

Reg‑TTR operates in two stages. First, a pre‑trained model—either a general registration foundation model such as uniGradICON or a task‑specific network—produces an initial dense deformation field (u_{\text{init}}) for a fixed‑moving image pair ((I_F, I_M)). This field is then treated as a learnable tensor and fed into a second stage where it is iteratively updated using the Adam optimizer. The optimization objective is a hybrid loss: \