Conditioned Implicit Neural Representation for Regularized Deformable Image Registration

Sidaty El Hadramy, Philippe C. Cattin

Keywords: Deformable Image Registration, Implicit Neural Representation, Regularization, Hyperparameter optimization

Abstract: Regularization is essential in deformable image registration (DIR) to ensure that the estimated Deformation Vector Field (DVF) remains smooth, physically plausible, and anatomically consistent. However, fine-tuning regularization parameters in learning-based DIR frameworks is computationally expensive, often requiring multiple training iterations. To address this, we propose $\operatorname{cIDIR}$, a novel DIR framework based on Implicit Neural Representations (INRs) that conditions the registration process on regularization hyperparameters. Unlike conventional methods that require retraining for each regularization hyperparameter setting, $\operatorname{cIDIR}$ is trained over a prior distribution of these hyperparameters, allowing real-time tuning during inference. Additionally, it models a continuous and differentiable DVF, enabling seamless integration of advanced regularization techniques. Evaluated on the DIR-LAB dataset, $\operatorname{cIDIR}$ achieves high accuracy and robustness across the dataset by leveraging real-time hyperparameter optimization after training.

Submission Number: 24