An unsupervised image registration method employing chest computed tomography images and deep neural networks

Thao Thi Ho; Woo Jin Kim; Chang Hyun Lee; Gong Yong Jin; Kum Ju Chae; Sanghun Choi

doi:10.1016/j.compbiomed.2023.106612

An unsupervised image registration method employing chest computed tomography images and deep neural networks

Comput Biol Med. 2023 Mar:154:106612. doi: 10.1016/j.compbiomed.2023.106612. Epub 2023 Jan 31.

Authors

Thao Thi Ho¹, Woo Jin Kim², Chang Hyun Lee³, Gong Yong Jin⁴, Kum Ju Chae⁴, Sanghun Choi⁵

Affiliations

¹ School of Mechanical Engineering, Kyungpook National University, Daegu, South Korea.
² Department of Internal Medicine and Environmental Health Center, Kangwon National University Hospital, School of Medicine, Kangwon National University, Chuncheon, South Korea.
³ Department of Radiology, Seoul National University, College of Medicine, Seoul National University Hospital, Seoul, South Korea; Department of Radiology, College of Medicine, The University of Iowa, Iowa City, IA, USA.
⁴ Department of Radiology, Research Institute of Clinical Medicine of Jeonbuk National University, Biomedical Research Institute of Jeonbuk National University Hospital, Jeonju, South Korea.
⁵ School of Mechanical Engineering, Kyungpook National University, Daegu, South Korea. Electronic address: s-choi@knu.ac.kr.

PMID: 36738711
DOI: 10.1016/j.compbiomed.2023.106612

Abstract

Background: Deformable image registration is crucial for multiple radiation therapy applications. Fast registration of computed tomography (CT) lung images is challenging because of the large and nonlinear deformation between inspiration and expiration. With advancements in deep learning techniques, learning-based registration methods are considered efficient alternatives to traditional methods in terms of accuracy and computational cost.

Method: In this study, an unsupervised lung registration network (LRN) with cycle-consistent training is proposed to align two acquired CT-derived lung datasets during breath-holds at inspiratory and expiratory levels without utilizing any ground-truth registration results. Generally, the LRN model uses three loss functions: image similarity, regularization, and Jacobian determinant. Here, LRN was trained on the CT datasets of 705 subjects and tested using 10 pairs of public CT DIR-Lab datasets. Furthermore, to evaluate the effectiveness of the registration technique, target registration errors (TREs) of the LRN model were compared with those of the conventional algorithm (sum of squared tissue volume difference; SSTVD) and a state-of-the-art unsupervised registration method (VoxelMorph).

Results: The results showed that the LRN with an average TRE of 1.78 ± 1.56 mm outperformed VoxelMorph with an average TRE of 2.43 ± 2.43 mm, which is comparable to that of SSTVD with an average TRE of 1.66 ± 1.49 mm. In addition, estimating the displacement vector field without any folding voxel consumed less than 2 s, demonstrating the superiority of the learning-based method with respect to fiducial marker tracking and the overall soft tissue alignment with a nearly real-time speed.

Conclusions: Therefore, this proposed method shows significant potential for use in time-sensitive pulmonary studies, such as lung motion tracking and image-guided surgery.

Keywords: CT lung; Deep learning; Image registration; Unsupervised learning.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Algorithms
Humans
Image Processing, Computer-Assisted* / methods
Lung / diagnostic imaging
Neural Networks, Computer
Tomography
Tomography, X-Ray Computed*