MICDIR: Multi-scale Inverse-consistent Deformable Image Registration using UNetMSS with Self-Constructing Graph Latent

Soumick Chatterjee,Himanshi Bajaj,Istiyak H. Siddiquee,Nandish Bandi Subbarayappa,Steve Simon,Suraj Bangalore Shashidhar,Oliver Speck,Andreas Nürnberge

DOI: https://doi.org/10.1016/j.compmedimag.2023.102267

2023-07-26

Abstract:Image registration is the process of bringing different images into a common coordinate system - a technique widely used in various applications of computer vision, such as remote sensing, image retrieval, and, most commonly, medical imaging. Deep learning based techniques have been applied successfully to tackle various complex medical image processing problems, including medical image registration. Over the years, several image registration techniques have been proposed using deep learning. Deformable image registration techniques such as Voxelmorph have been successful in capturing finer changes and providing smoother deformations. However, Voxelmorph, as well as ICNet and FIRE, do not explicitly encode global dependencies (i.e. the overall anatomical view of the supplied image) and, therefore, cannot track large deformations. In order to tackle the aforementioned problems, this paper extends the Voxelmorph approach in three different ways. To improve the performance in case of small as well as large deformations, supervision of the model at different resolutions has been integrated using a multi-scale UNet. To support the network to learn and encode the minute structural co-relations of the given image-pairs, a self-constructing graph network (SCGNet) has been used as the latent of the multi-scale UNet - which can improve the learning process of the model and help the model to generalise better. And finally, to make the deformations inverse-consistent, cycle consistency loss has been employed. On the task of registration of brain MRIs, the proposed method achieved significant improvements over ANTs and VoxelMorph, obtaining a Dice score of 0.8013 \pm 0.0243 for intramodal and 0.6211 \pm 0.0309 for intermodal, while VoxelMorph achieved 0.7747 \pm 0.0260 and 0.6071 \pm 0.0510, respectively

Image and Video Processing,Artificial Intelligence,Computer Vision and Pattern Recognition,Machine Learning,Medical Physics

What problem does this paper attempt to address?

The paper aims to address several key issues in medical image registration, particularly the limitations of existing methods in handling deformations ranging from small to large and the failure to adequately encode global dependencies (such as the overall anatomical view of the image).

Specifically, the main objectives of the study include:

1. **Developing an image registration method capable of handling various scales of deformation**: Existing methods like Voxelmorph, ICNet, and FIRE have limitations when dealing with larger deformations. Therefore, the proposed method in this study aims to improve performance for both small and large deformations.

2. **Integrating structural information to improve registration quality**: Existing studies often overlook or fail to effectively utilize the structural connectivity information within images. This study introduces a Self-Constructing Graph Network (SCGNet) to better integrate structural information within the images, thereby improving registration results.

3. **Achieving inverse-consistent deformation**: To ensure that the deformation remains consistent in both directions (from the moving image to the fixed image and vice versa), the study employs a cycle-consistency loss.

To achieve these goals, the paper proposes the MICDIR (Multi-scale Inverse-consistent Deformable Image Registration) method, which combines a multi-scale UNet architecture, a Self-Constructing Graph Network (for capturing semantic information), and cycle-consistency loss (to ensure deformation consistency). This method is capable of working across different scales of deformation and has been experimentally validated to show significant performance improvements in brain MRI image registration tasks, outperforming baseline methods such as ANTs and Voxelmorph.