Robust Rigid and Non-Rigid Medical Image Registration Using Learnable Edge Kernels

2025-12-01

AI Summary

Medical image registration faces challenges including inter-modal intensity discrepancies, spatial distortions, and modality-specific anatomical variations. To address these, we propose a learnable edge-enhanced registration framework: it integrates learnable edge convolutional kernels—optimized under noise perturbation—into a deep learning registration network to explicitly guide feature learning toward anatomical boundaries and enhance structural awareness; supports both rigid and non-rigid deformation modeling via multi-stage feature matching and deformation field estimation. Through systematic ablation studies involving eight variants, we quantitatively validate the contribution of each component. Evaluated on two public benchmarks under three distinct experimental settings, our method consistently outperforms state-of-the-art approaches, achieving significant improvements in both registration accuracy (e.g., TRE, DICE) and anatomical consistency (e.g., boundary alignment, Jacobian determinant smoothness).

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Multi-modal Vision
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Medical image registration is crucial for various clinical and research applications including disease diagnosis or treatment planning which require alignment of images from different modalities, time points, or subjects. Traditional registration techniques often struggle with challenges such as contrast differences, spatial distortions, and modality-specific variations. To address these limitations, we propose a method that integrates learnable edge kernels with learning-based rigid and non-rigid registration techniques. Unlike conventional layers that learn all features without specific bias, our approach begins with a predefined edge detection kernel, which is then perturbed with random noise. These kernels are learned during training to extract optimal edge features tailored to the task. This adaptive edge detection enhances the registration process by capturing diverse structural features critical in medical imaging. To provide clearer insight into the contribution of each component in our design, we introduce four variant models for rigid registration and four variant models for non-rigid registration. We evaluated our approach using a dataset provided by the Medical University across three setups: rigid registration without skull removal, with skull removal, and non-rigid registration. Additionally, we assessed performance on two publicly available datasets. Across all experiments, our method consistently outperformed state-of-the-art techniques, demonstrating its potential to improve multi-modal image alignment and anatomical structure analysis.

Problem

• Improves medical image alignment across modalities and time points
• Addresses contrast differences and spatial distortions in registration
• Enhances structural feature extraction for better anatomical analysis

Innovation

• Learnable edge kernels enhance registration accuracy
• Adaptive edge detection captures diverse structural features
• Method outperforms state-of-the-art in multi-modal alignment