Related Experiment Video
Updated: Oct 28, 2025
Retrospective Cardiac Gating with A Prototype Small-Animal X-ray Computed Tomograph
Published on: February 21, 2025
GapFill-Recon Net: A Cascade Network for simultaneously PET Gap Filling and Image Reconstruction
Yanchao Huang1, Huobiao Zhu2, Xiaoman Duan3
1School of Biomedical Engineering and Guangdong Provincial Key Laboratory of Medial Image Processing, Southern Medical University, Guangzhou, Guangdong 510515, China; Guangdong Province Engineering Laboratory for Medical Imaging and Diagnostic Technology, Southern Medical University, Guangzhou, China; Nanfang PET Center, Nanfang Hospital, Southern Medical University, Guangzhou, Guangdong 510515, China.
This study introduces GapFill-Recon Net, a novel convolutional neural network (CNN) for Positron Emission Tomography (PET) image reconstruction. The framework efficiently reconstructs PET images from incomplete data, outperforming traditional methods in quality and speed.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Healthcare
- Computational Science
Background:
- Positron Emission Tomography (PET) image reconstruction faces challenges due to data loss from detector gaps.
- Incomplete projection data significantly impacts the quality of reconstructed PET images.
Purpose of the Study:
- To develop an efficient convolutional neural network (CNN) framework for simultaneous PET image and sinogram reconstruction.
- To address the issue of partial projection data loss caused by gaps between detector blocks in PET imaging.
Main Methods:
- Proposed GapFill-Recon Net, a CNN framework with two blocks: Gap-Filling and Image-Reconstruction.
- Trained and validated the network using 43,660 pairs of synthetic 2D PET sinograms and images from the MOBY phantom.
- Evaluated performance on whole-body mouse Monte Carlo simulated data.
Main Results:
- GapFill-Recon Net demonstrated superior image quality compared to Filtered Back-Projection (FBP) and Maximum Likelihood Expectation Maximization (MLEM), evidenced by higher SSIM and PSNR, and lower rRMSE.
- Achieved reconstruction speeds comparable to FBP and significantly faster (83x) than MLEM.
- Effectively balanced image quality and reconstruction efficiency.
Conclusions:
- GapFill-Recon Net offers a significant advancement over traditional PET reconstruction algorithms.
- The proposed CNN framework provides optimal performance in both image quality and reconstruction speed.
- This method effectively addresses data loss issues in PET imaging, enhancing diagnostic accuracy and efficiency.
