Gaussian Primitive Optimized Deformable Retinal Image Registration

2025-08-22

AI Summary

Retinal image deformable registration faces two key challenges: gradient vanishing in large homogeneous regions and insufficient stable supervision from sparse vascular structures. To address these, we propose a Gaussian primitive optimization framework that parameterizes learnable Gaussian primitives—each defined by position, displacement, and radius—as structured control points, actively anchoring at high-gradient regions (e.g., vessels) to enhance gradient propagation. Displacement fields are propagated via K-nearest-neighbor Gaussian interpolation, and the framework jointly optimizes keypoint consistency and intensity alignment in an end-to-end multi-task fashion. On the FIRE dataset, our method reduces target registration error (TRE) from 6.2 to 2.4 pixels and improves AUC@25px from 0.770 to 0.938, outperforming state-of-the-art approaches. The core contribution lies in introducing geometrically interpretable, gradient-friendly Gaussian primitives as sparse yet robust deformation modeling units.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Deformable retinal image registration is notoriously difficult due to large homogeneous regions and sparse but critical vascular features, which cause limited gradient signals in standard learning-based frameworks. In this paper, we introduce Gaussian Primitive Optimization (GPO), a novel iterative framework that performs structured message passing to overcome these challenges. After an initial coarse alignment, we extract keypoints at salient anatomical structures (e.g., major vessels) to serve as a minimal set of descriptor-based control nodes (DCN). Each node is modelled as a Gaussian primitive with trainable position, displacement, and radius, thus adapting its spatial influence to local deformation scales. A K-Nearest Neighbors (KNN) Gaussian interpolation then blends and propagates displacement signals from these information-rich nodes to construct a globally coherent displacement field; focusing interpolation on the top (K) neighbors reduces computational overhead while preserving local detail. By strategically anchoring nodes in high-gradient regions, GPO ensures robust gradient flow, mitigating vanishing gradient signal in textureless areas. The framework is optimized end-to-end via a multi-term loss that enforces both keypoint consistency and intensity alignment. Experiments on the FIRE dataset show that GPO reduces the target registration error from 6.2,px to ~2.4,px and increases the AUC at 25,px from 0.770 to 0.938, substantially outperforming existing methods. The source code can be accessed via https://github.com/xintian-99/GPOreg.

Problem

• Overcoming sparse vascular features in retinal image registration
• Addressing vanishing gradients in homogeneous retinal regions
• Improving deformation field accuracy through structured message passing

Innovation

• Gaussian Primitive Optimization for structured message passing
• Keypoints as descriptor-based control nodes with trainable parameters
• KNN Gaussian interpolation for coherent displacement field