NCF: neural correspondence field for medical image registration

2025-03-02

Medical Imaging 2025: Image Processing

AI Summary

Medical image deformable registration suffers from poor generalizability due to the scarcity of annotated training data. To address this, we propose the first training-free, single-shot registration paradigm: for any given image pair, our method performs end-to-end optimization of a lightweight implicit neural network (0.06M parameters) using only that pair, directly modeling a continuous coordinate-to-displacement mapping. The optimization jointly minimizes an unsupervised photometric consistency loss and a deformation regularizer, enabling personalized, training-free, high-accuracy registration. Evaluated on the public Lung CT dataset, our approach achieves state-of-the-art performance; on a clinical head-and-neck dataset, it significantly outperforms conventional iterative optimization methods. These results demonstrate superior generalizability across domains and strong clinical applicability without requiring any pretraining or external supervision.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Machine Learning: Multimodal Learning

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• 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

Abstract

Deformable image registration is a fundamental task in medical image processing. Traditional optimization-based methods often struggle with accuracy in dealing with complex deformation. Recently, learning-based methods have achieved good performance on public datasets, but the scarcity of medical image data makes it challenging to build a generalizable model to handle diverse real-world scenarios. To address this, we propose a training-data-free learning-based method, Neural Correspondence Field (NCF), which can learn from just one data pair. Our approach employs a compact neural network to model the correspondence field and optimize model parameters for each individual image pair. Consequently, each pair has a unique set of network weights. Notably, our model is highly efficient, utilizing only 0.06 million parameters. Evaluation results showed that the proposed method achieved superior performance on a public Lung CT dataset and outperformed a traditional method on a head and neck dataset, demonstrating both its effectiveness and efficiency.

Problem

• Addresses accuracy issues in deformable medical image registration.
• Overcomes data scarcity for generalizable learning-based models.
• Proposes a training-data-free method using minimal parameters.

Innovation

• Training-data-free learning-based method
• Compact neural network for correspondence field
• Unique network weights per image pair