Progressive Retinal Image Registration via Global and Local Deformable Transformations

TL;DR

A hybrid registration framework called HybridRetina is proposed, which progressively registers retinal images with global and local deformable transformations, and uses a keypoint detector and a deformation network called GAMorph to estimate the global and local deformable transformation, respectively.

Abstract

Retinal image registration plays an important role in the ophthalmological diagnosis process. Since there exist variances in viewing angles and anatomical structures across different retinal images, keypoint-based approaches become the mainstream methods for retinal image registration thanks to their robustness and low latency. These methods typically assume the retinal surfaces are planar, and adopt feature matching to obtain the homography matrix that represents the global transformation between images. Yet, such a planar hypothesis inevitably introduces registration errors since retinal surface is approximately curved. This limitation is more prominent when registering image pairs with significant differences in viewing angles. To address this problem, we propose a hybrid registration framework called HybridRetina, which progressively registers retinal images with global and local deformable transformations. For that, we use a keypoint detector and a deformation network called GAMorph to estimate the global transformation and local deformable transformation, respectively. Specifically, we integrate multi-level pixel relation knowledge to guide the training of GAMorph. Additionally, we utilize an edge attention module that includes the geometric priors of the images, ensuring the deformation field focuses more on the vascular regions of clinical interest. Experiments on two widely-used datasets, FIRE and FLoRI21, show that our proposed HybridRetina significantly outperforms some state-of-the-art methods. The code is available at https://github.com/lyp-deeplearning/awesome-retinal-registration.

Figures (5)

• Figure 1: An example demonstrates the registration results of the global transformation and our progressive transformation. We use mosaic images to compare the registration performance, and the continuity of blood vessels across tiles reflects the registration quality. SuperRetina liu2022semi exhibits excellent feature matching performance in global transformation, but the registration results still contain obvious errors.
• Figure 2: Illustration of Our HybridRetina Framework. (a) The steps of global registration and local deformable registration. (b) The process of generating paired data for our proposed GAMorph. (c) The training process of GAMorph.
• Figure 3: A toy illustration of computing affinity map $A_g$. We use matching prior and Manhattan distance to calculate the affinity value $a$ between all pixels.
• Figure 4: Comparison of pixel-wise differences between the moving image and the fixed image using different similarity measurement methods. Results of normalized cross-correlation (NCC) and our proposed edge-guided normalized cross-correlation (ENCC) are shown.
• Figure 5: Mosaic visualization of registration results. We show the alignment in the region enclosed by the red box. We observe misalignment in the results by GeoFormer liu2023geometrized, highlighted by green circles. Best viewed electronically.