Abstract:Deformable Image Registration (DIR) is essential for aligning medical images that exhibit anatomical variations, facilitating applications such as disease tracking and radiotherapy planning. While classical iterative methods and deep learning approaches have achieved success in DIR, they are often hindered by computational inefficiency or poor generalization. In this paper, we introduce GaussianDIR, a novel, case-specific optimization DIR method inspired by 3D Gaussian splatting. In general, GaussianDIR represents image deformations using a sparse set of mobile and flexible Gaussian primitives, each defined by a center position, covariance, and local rigid transformation. This compact and explicit representation reduces noise and computational overhead while improving interpretability. Furthermore, the movement of individual voxel is derived via blending the local rigid transformation of the neighboring Gaussian primitives. By this, GaussianDIR captures both global smoothness and local rigidity as well as reduces the computational burden. To address varying levels of deformation complexity, GaussianDIR also integrates an adaptive density control mechanism that dynamically adjusts the density of Gaussian primitives. Additionally, we employ multi-scale Gaussian primitives to capture both coarse and fine deformations, reducing optimization to local minima. Experimental results on brain MRI, lung CT, and cardiac MRI datasets demonstrate that GaussianDIR outperforms existing DIR methods in both accuracy and efficiency, highlighting its potential for clinical applications. Finally, as a training-free approach, it challenges the stereotype that iterative methods are inherently slow and transcend the limitations of poor generalization.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the Deformable Image Registration (DIR) problem in medical image registration. Specifically, the goal of DIR is to calculate a Displacement Vector Field (DVF) for deforming the moving image to align it with the fixed image. This is crucial for processing medical images with significant anatomical changes and is helpful for applications such as disease tracking and radiotherapy planning. #### Limitations of existing methods: 1. **Classical iterative methods**: Although these methods are robust and generalize well across different modalities, they are usually computationally expensive, especially when dealing with large - sized images. 2. **Deep learning methods**: Although DL methods can automatically extract features and increase processing speed, they rely on large - scale datasets, which are difficult to obtain in the medical field. In addition, these methods often show poor generalization ability when applied to data distributions different from the training set. 3. **Implicit Neural Representation (INR) methods**: Although these methods are flexible and smooth, due to their implicit nature, they lack interpretability and are slow in processing, unable to meet the requirements of real - time applications. #### Innovations of GaussianDIR: - **Compact representation based on Gaussian primitives**: GaussianDIR uses a set of sparsely distributed Gaussian primitives to represent image deformation. Each Gaussian primitive is defined by a central position, a covariance matrix, and a local rigid transformation. This representation not only reduces noise and computational overhead but also improves the interpretability of the model. - **Fusion of global smoothness and local rigidity**: By using the K - Nearest Neighbors (KNN) algorithm to select the nearest Gaussian primitives and adopting the α - blending method to calculate the deformation of each voxel, global smoothness and local rigidity are captured. - **Adaptive density control mechanism**: Dynamically adjust the density of Gaussian primitives according to the deformation complexity to avoid over - fitting or under - fitting. - **Multi - scale Gaussian primitives**: Combine coarse and fine deformations to reduce the risk of the optimization process getting trapped in local minima. - **Efficient method without training**: Compared with traditional iterative methods, GaussianDIR significantly improves processing speed while maintaining accuracy, challenging the inherent inefficiency of iterative methods and overcoming the generalization limitations of deep learning methods. #### Experimental results: The experimental results show that GaussianDIR outperforms existing DIR methods on multiple datasets such as brain MRI, lung CT, and heart MRI, and performs excellently in terms of accuracy and efficiency, demonstrating its great potential in clinical applications. In conclusion, this paper proposes a novel Gaussian - primitive - based deformable image registration method, GaussianDIR, which solves the deficiencies of existing methods in terms of computational efficiency and generalization ability, providing a more efficient and accurate solution for medical image registration.

Gaussian Primitives for Deformable Image Registration

Neural Graphics Primitives-based Deformable Image Registration for On-the-fly Motion Extraction

GraformerDIR: Graph convolution transformer for deformable image registration

A Local Registration Approach Of Medical Images With Niche Genetic Algorithm

4D-CT Deformable Image Registration Using Multiscale Unsupervised Deep Learning.

GiNGR: Generalized Iterative Non-Rigid Point Cloud and Surface Registration Using Gaussian Process Regression

A Hybrid Scheme for Efficient Deformable Image Registration.

LeanGaussian: Breaking Pixel or Point Cloud Correspondence in Modeling 3D Gaussians

An Unsupervised Deep Learning Approach for 4DCT Lung Deformable Image Registration

4D-CT Deformable Image Registration Using an Unsupervised Deep Convolutional Neural Network.

Continuous sPatial-Temporal Deformable Image Registration (CPT-DIR) for motion modelling in radiotherapy: beyond classic voxel-based methods

Multi-Objective Learning for Deformable Image Registration

Technical Note: Deformable Image Registration on Partially Matched Images for Radiotherapy Applications.

Uncertainty estimation and evaluation of deformation image registration based convolutional neural networks

DeepGMR: Learning Latent Gaussian Mixture Models for Registration

Evaluation of various Deformable Image Registrations for Point and Volume Variations

TransDIR: Deformable imaging registration network based on transformer to improve the feature extraction ability

Medical Image Registration via Neural Fields

A Plug-and-Play Image Registration Network

Improving Deformable Image Registration Accuracy through a Hybrid Similarity Metric and CycleGAN Based Auto-Segmentation

Deformable CT Image Registration Via a Dual Feasible Neural Network