Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RSTAR4D: Rotational Streak Artifact Reduction in 4D CBCT using a Separable 4D CNN

Ziheng Deng, Hua Chen, Yongzheng Zhou, Haibo Hu, Zhiyong Xu, Jiayuan Sun, Tianling Lyu, Yan Xi, Yang Chen, Jun Zhao

TL;DR

The paper addresses rotational streak artifacts in 4D CBCT caused by sparse phase-specific projections during respiration. It introduces RSTAR4D-Net, a separable 4D CNN that processes the full 4D CBCT volume to jointly model spatial and temporal patterns, coupled with a two-stage Tetris training strategy to enable learning from limited data. Across simulated and real clinical datasets, the approach outperforms 2D CNN baselines and other artifact-reduction methods in SSIM and RMSE, while maintaining anatomical fidelity in projection-domain evaluations. This work enables higher-quality 4D CBCT within short scan times, enhancing motion management and lesion localization for image-guided radiotherapy.

Abstract

Four-dimensional cone-beam computed tomography (4D CBCT) provides respiration-resolved images and can be used for image-guided radiation therapy. However, the ability to reveal respiratory motion comes at the cost of image artifacts. As raw projection data are sorted into multiple respiratory phases, the cone-beam projections become much sparser and the reconstructed 4D CBCT images will be covered by severe streak artifacts. Although several deep learning-based methods have been proposed to address this issue, most algorithms employ 2D network models as backbones, neglecting the intrinsic structural priors within 4D CBCT images. In this paper, we first explore the origin and appearance of streak artifacts in 4D CBCT images. We find that streak artifacts exhibit a unique rotational motion along with the patient's respiration, distinguishable from diaphragm-driven respiratory motion in the spatiotemporal domain. Therefore, we propose a novel 4D neural network model, RSTAR4D-Net, designed to address Rotational STreak Artifact Reduction by integrating the spatial and temporal information within 4D CBCT images. Specifically, we overcome the computational and training difficulties of a 4D neural network. The specially designed model adopts an efficient implementation of 4D convolutions to reduce computational costs and thus can process the whole 4D image in one pass. Additionally, a Tetris training strategy pertinent to the separable 4D convolutions is proposed to effectively train the model using limited 4D training samples. Extensive experiments substantiate the effectiveness of our proposed method, and the RSTAR4D-Net shows superior performance compared to other methods. The source code and dynamic demos are available at https://github.com/ivy9092111111/RSTAR.

RSTAR4D: Rotational Streak Artifact Reduction in 4D CBCT using a Separable 4D CNN

TL;DR

The paper addresses rotational streak artifacts in 4D CBCT caused by sparse phase-specific projections during respiration. It introduces RSTAR4D-Net, a separable 4D CNN that processes the full 4D CBCT volume to jointly model spatial and temporal patterns, coupled with a two-stage Tetris training strategy to enable learning from limited data. Across simulated and real clinical datasets, the approach outperforms 2D CNN baselines and other artifact-reduction methods in SSIM and RMSE, while maintaining anatomical fidelity in projection-domain evaluations. This work enables higher-quality 4D CBCT within short scan times, enhancing motion management and lesion localization for image-guided radiotherapy.

Abstract

Four-dimensional cone-beam computed tomography (4D CBCT) provides respiration-resolved images and can be used for image-guided radiation therapy. However, the ability to reveal respiratory motion comes at the cost of image artifacts. As raw projection data are sorted into multiple respiratory phases, the cone-beam projections become much sparser and the reconstructed 4D CBCT images will be covered by severe streak artifacts. Although several deep learning-based methods have been proposed to address this issue, most algorithms employ 2D network models as backbones, neglecting the intrinsic structural priors within 4D CBCT images. In this paper, we first explore the origin and appearance of streak artifacts in 4D CBCT images. We find that streak artifacts exhibit a unique rotational motion along with the patient's respiration, distinguishable from diaphragm-driven respiratory motion in the spatiotemporal domain. Therefore, we propose a novel 4D neural network model, RSTAR4D-Net, designed to address Rotational STreak Artifact Reduction by integrating the spatial and temporal information within 4D CBCT images. Specifically, we overcome the computational and training difficulties of a 4D neural network. The specially designed model adopts an efficient implementation of 4D convolutions to reduce computational costs and thus can process the whole 4D image in one pass. Additionally, a Tetris training strategy pertinent to the separable 4D convolutions is proposed to effectively train the model using limited 4D training samples. Extensive experiments substantiate the effectiveness of our proposed method, and the RSTAR4D-Net shows superior performance compared to other methods. The source code and dynamic demos are available at https://github.com/ivy9092111111/RSTAR.
Paper Structure (16 sections, 2 equations, 9 figures, 5 tables)

This paper contains 16 sections, 2 equations, 9 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Material and methods
  3. Rotational Streak Artifacts (RSA) in 4D CBCT Images
  4. RSTAR4D-Net
  5. Model Architecture
  6. Tetris training strategy
  7. Experiments
  8. Simulated training and validation datasets
  9. Simulated test dataset
  10. Real clinical test dataset
  11. Comparison Methods
  12. Implementation Details
  13. Results
  14. Experiments on the Simulated Dataset
  15. Experiments on the Real Clinical Dataset
  16. ...and 1 more sections

Figures (9)

  • Figure 1: 4D CBCT imaging with breathing signals. (a) The average image with blurry structure edges, (b) the breathing signal used to sort the projection data into different phases, (c) the phase sorting map, and (d) a phase-resolved image reconstructed from single-phase projection data, which reveals the edge of the diaphragm but suffers severe streak artifacts.
  • Figure 2: Illustration of rotational streak artifacts (RSA). (a1)-(a5) The projection sampling patterns at five respiratory phases, (b1)-(b5) phase-resolved images with RSA.
  • Figure 3: Motion trajectories of densely sampled feature points projected to axial and coronal planes. (a) Respiration-induced motion in the axial plane, (b) RSA-induced motion in the axial plane, (c) Hybrid motion in the axial plane, (d) t-SNE plot of the three types of motion trajectory; (e) Respiration-induced motion in the coronal plane, (f) RSA-induced motion in the coronal plane, (g) Hybrid motion in the coronal plane, (h) The average optical flow component along different axes.
  • Figure 4: The design of the RSTAR4D-Net. (a) The RSTAR4D-Net takes a whole 4D CBCT image as the input. (b1-b2) 4D convolutional modules: (b1) Vanilla isotropic 4D convolutions, (b2) lightweight separable 4D convolutions (proposed).
  • Figure 5: The Tetris training strategy. (a) The RSTAR4D-Net processes an entire 4D CBCT image at the inference stage. (b1) 2D+T images used in Tetris training stage I, (b2) 4D image blocks of various sizes used in Tetris training stage II.
  • ...and 4 more figures