Related Experiment Video
Updated: May 6, 2026
Four-Dimensional Computed Tomography-Guided Valve Sizing for Transcatheter Pulmonary Valve Replacement
Published on: January 20, 2022
An end-to-end neural network for 4D cardiac CT reconstruction using single-beat scans
Zhenyao Yan1, Zhennong Chen2, Liang Li1
1Department of Engineering Physics, Tsinghua University, Beijing 100084, People's Republic of China.
Insights
This study introduces a deep learning framework for cardiac CT imaging, significantly reducing motion artifacts from single-beat scans. The method improves image quality and diagnostic accuracy for cardiac diseases, even with irregular heart rhythms.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Disease
Background:
- Motion artifacts are a major challenge in cardiac CT, hindering accurate diagnosis.
- Traditional methods like rescans increase radiation and are ineffective for irregular heart rhythms.
Purpose of the Study:
- To develop a deep learning-based framework for motion artifact reduction in cardiac CT using single-beat rapid scanning.
- To improve the accuracy of cardiac disease detection and diagnosis.
Main Methods:
- A deep learning end-to-end reconstruction framework was proposed.
- A sliding-window approach divided projection data into cardiac phase-centered intervals.
- Denoising and registration networks computed deformation vector fields for motion-compensated reconstruction.
Main Results:
- The method effectively reduced motion artifacts and restored anatomical structures.
- Key metrics showed significant improvements: SSIM increased to 0.7795, PSNR to 37.58, and RMSE (HU) decreased to 49.28.
- Segmentation accuracy improved, with Dice scores for the left ventricle, coronary arteries, and calcified plaques reaching 0.9614, 0.8811, and 0.7774, respectively.
Conclusions:
- The proposed deep learning framework shows strong potential for generalizability in clinical applications.
- It enables accurate cardiac CT imaging from single-cycle scans without heart rate restrictions.
- This approach offers a promising solution for overcoming motion artifacts in dynamic cardiac imaging.
Abstract:
Objective.Motion artifacts remain a significant challenge in cardiac CT imaging, often impairing the accurate detection and diagnosis of cardiac diseases. These artifacts result from involuntary cardiac motion, and traditional mitigation methods typically rely on retrospective rescans, which increase radiation exposure and are less effective for patients with irregular heart rhythms. In this study, we proposed a deep learning-based end-to-end reconstruction framework for dynamic cardiac imaging using single-beat rapid CT scanning.Approach.The method used common cardiac CT projections and applied a sliding-window approach to divide the projection data into overlapping short-scan intervals centered on specific cardiac phases. Each short-scan interval was first reconstructed and then processed through a denoising network, before being fed into the registration module to compute deformation vector fields (DVFs). These DVFs were then used to perform motion-compensated reconstruction. The denoising and registration networks were trained end-to-end by minimizing the difference between the reconstructed images and ground truth in a supervised manner. The model was trained using simulated projection data from 30 real patients and validated on simulated datasets from different institutions and XCAT-generated continuous phantoms.Main results.Experimental results showed that the proposed method effectively reduced motion artifacts and restored key anatomical structures such as coronary arteries. On high-resolution test cases, SSIM improved from 0.7234 to 0.7795, peak signal-to-noise ratio from 35.40 to 37.58, and root mean square error (HU) decreased from 63.98 to 49.28. Additional evaluations showed consistent improvements in segmentation accuracy, with Dice similarity coefficient scores for the left ventricle, coronary arteries, and calcified plaques increasing from 0.8025, 0.7347, and 0.5966 to 0.9614, 0.8811, and 0.7774, respectively.Significance.By relying solely on single-cycle scan data and placing no explicit restrictions on heart rate, the method demonstrated strong potential for generalizability and wider clinical applications.
