DINOv3 with Test-Time Training for Medical Image Registration

TL;DR

This paper introduces a training-free medical image registration method using a frozen DINOv3 encoder and test-time optimization, achieving high accuracy and regular deformations without additional training data.

Contribution

It presents a novel training-free pipeline that leverages a frozen foundation model and test-time optimization for medical image registration, bypassing the need for large training datasets.

Findings

Achieved the best mean Dice score of 0.790 on Abdomen MR-CT.

Improved mean DSC to 0.769 on ACDC cardiac MRI.

Reduced deformation irregularities with low Hausdorff Distance and Log-Jacobian standard deviation.

Abstract

Prior medical image registration approaches, particularly learning-based methods, often require large amounts of training data, which constrains clinical adoption. To overcome this limitation, we propose a training-free pipeline that relies on a frozen DINOv3 encoder and test-time optimization of the deformation field in feature space. Across two representative benchmarks, the method is accurate and yields regular deformations. On Abdomen MR-CT, it attained the best mean Dice score (DSC) of 0.790 together with the lowest 95th percentile Hausdorff Distance (HD95) of 4.9+-5.0 and the lowest standard deviation of Log-Jacobian (SDLogJ) of 0.08+-0.02. On ACDC cardiac MRI, it improves mean DSC to 0.769 and reduces SDLogJ to 0.11 and HD95 to 4.8, a marked gain over the initial alignment. The results indicate that operating in a compact foundation feature space at test time offers a practical and general solution for clinical registration without additional training.