Related Experiment Video
Updated: Jun 26, 2026
Creating Dynamic Images of Short-lived Dopamine Fluctuations with lp-ntPET: Dopamine Movies of Cigarette Smoking
Published on: August 6, 2013
Direct parametric reconstruction in dynamic PET using deep image prior and a novel parameter magnification strategy
Xiaotong Hong1, Fanghu Wang2, Hao Sun1
1School of Biomedical Engineering, Southern Medical University, 1023 Shatai Road, Guangzhou, 510515, China; Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical University, 1023 Shatai Road, Guangzhou 510515, China; Guangdong Province Engineering Laboratory for Medical Imaging and Diagnostic Technology, Southern Medical University, 1023 Shatai Road, Guangzhou, 510515, China.
This study introduces DIP-PM, a novel deep learning method for positron emission tomography (PET) parametric imaging. DIP-PM significantly improves the reconstruction quality of challenging micro-parameters, enhancing diagnostic accuracy.
Area of Science:
- Medical Imaging
- Positron Emission Tomography (PET)
- Artificial Intelligence in Medicine
Background:
- Parametric imaging in PET is hindered by noisy data and complex kinetic parameter mapping.
- Existing methods struggle with nonlinear and small-value micro-parameters like k2 and k3.
Purpose of the Study:
- To present a novel unsupervised deep learning approach for reconstructing and enhancing PET parametric images.
- To improve the quality of nonlinear and small-value micro-parameters.
Main Methods:
- A direct parametric image reconstruction model, DIP-PM, integrating deep image prior (DIP) with parameter magnification (PM) was developed.
- A U-Net generator predicted parametric images, with subsequent intensity magnification.
- The model was optimized using log-likelihood loss and evaluated on simulated 82Rb and 18F-FDG data, compared against indirect and DIP-only methods.
Main Results:
- DIP-PM outperformed traditional and DIP-only methods in reconstructing micro-parameters (k2, k3) with better structural preservation.
- Achieved superior quantitative metrics (PSNR, NRMSE, SSIM) for both 1-tissue and 2-tissue compartment models.
- Demonstrated advantages on real 18F-FDG data, preserving myocardial structures.
Conclusions:
- The DIP-based direct parametric imaging approach is effective for generating high-quality PET parametric images.
- The proposed DIP-PM method with parameter magnification enhances the fidelity of nonlinear micro-parameter images.
