Traditional LDDMM vs Deep Learning deformation fields on NIREP and OASIS

This paper explores the connections between traditional Large Deformation Diffeomorphic Metric Mapping methods andunsupervised deep-learning approaches for non-rigid registration, particularly emphasizing diffeomorphic registration. Thestudy provides useful insights and establishes connections between the methods, thereby facilitating a profound understand-ing of the methodological landscape. The methods considered in our study are extensively evaluated in T1w MRI imagesusing traditional NIREP and Learn2Reg OASIS evaluation protocols with a focus on fairness, to establish equitable bench-marks and facilitate informed comparisons. Through a comprehensive analysis of the results, we address key questions,including the intricate relationship between accuracy and transformation quality in performance, the disentanglement of theinfluence of registration ingredients on performance, and the determination of benchmark methods and baselines. We offervaluable insights into the strengths and limitations of both traditional and deep-learning methods, shedding light on theircomparative performance and guiding future advancements in the field.

本论文探究了传统大变形微分同胚度量映射（Large Deformation Diffeomorphic Metric Mapping）方法与用于非刚性配准的无监督深度学习方法之间的内在关联，重点聚焦于微分同胚配准方向。本研究阐明了两类方法间的关联逻辑，提供了极具价值的研究见解，进而助力研究者深入理解该领域的方法学全貌。
本研究所涉及的方法，均基于传统NIREP与Learn2Reg OASIS评估协议，在T1加权磁共振成像（T1w MRI）图像中开展了全面的性能评估，全程以公平性为核心准则，旨在构建公允的基准测试体系，支撑严谨的科学对比。通过对实验结果的系统性分析，本研究回应了多项核心议题：包括配准性能中精度与变换质量间的复杂关联机制、解耦各配准要素对性能的影响路径，以及基准方法与基线方案的遴选与确定。本研究针对传统方法与深度学习方法的优势与局限展开了深入剖析，清晰呈现二者的性能对比结果，为该领域的后续研究与技术迭代提供了科学指引。