Improved Objective Functions for Implicit Neural Representations in Deformable Image Registration

Johannes Bastian Gebauer, Maximilian Nielsen, Frederic Madesta, René Werner, Thilo Sentker

Keywords: Implicit Neural Representations, Deformable Image Registration, Multi-Scale Optimization, Thoracic CT

TL;DR: We enhance implicit neural representations with improved loss and regularization terms to achieve state-of-the-art, sub-millimeter deformable image registration on DIR-Lab and DIR-Lab COPD datasets.

Abstract: We propose an enhanced multi-scale Implicit Neural Representation (INR) framework for dense deformable image registration, designed to maximize alignment accuracy and deformation regularity. By modeling the transformation as a coordinate-based neural field, we optimize directly on image pairs using a coarse-to-fine dual-branch architecture that splits motion into global and local components. The objective function is driven by mask-guided Normalized Cross-Correlation and curvature regularization to ensure smooth, anatomically plausible motion. Evaluation on the DIR-Lab 4DCT thorax dataset demonstrates state-of-the-art performance with a mean Target Registration Error (TRE) below 1.0 mm. On the more challenging DIR-Lab COPDgene thorax dataset, the model achieves robust alignment with a mean TRE of 1.23 mm, yielding performance comparable to leading classical optimization frameworks. A comprehensive ablation study confirms that the dual-branch design and multi-scale optimization strategy are necessary to achieve these results, enabling precise registration with modest computational overhead. These findings demonstrate that carefully structured INR frameworks can achieve sub-millimeter precision on standard benchmarks while maintaining robustness in large-deformation scenarios. Source code will be made available upon acceptance at https://github.com/IPMI-ICNS-UKE/INR-DIR.

Primary Subject Area: Image Registration

Secondary Subject Area: Unsupervised Learning and Representation Learning

Registration Requirement: Yes

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 283