Multi-scale Lung Image Registration Based on Implicit Neural Representation

Deformable image registration (DIR) is crucial in medical image analysis, particularly in lung image analysis. DIR faces significant challenges when handling large to small-scale deformations, owing to the unique deformation capacity and complex motion patterns of the lung. Although deep learning-based registration methods have been proposed and adopted recently, they still face challenges due to the scarcity of datasets and suboptimal performance in managing large-scale deformations. This paper presents a novel multi-scale image registration approach based on implicit neural representations (INR), designed to overcome the lack of datasets using only moving and fixed images as inputs. The experimental results confirm its superior performance in registration tasks compared to current methods. This research offers a new perspective on the improvement and application of lung image registration.