Related Experiment Video
Updated: Oct 1, 2025
14:21
Creating Dynamic Images of Short-lived Dopamine Fluctuations with lp-ntPET: Dopamine Movies of Cigarette Smoking
Published on: August 6, 2013
18.4K
Simultaneous Denoising of Dynamic PET Images Based on Deep Image Prior
Cheng-Hsun Yang1, Hsuan-Ming Huang2
1Institute of Medical Device and Imaging, College of Medicine, National Taiwan University, No.1, Sec. 1, Jen Ai Rd., Zhongzheng Dist., Taipei City 100, Taiwan.
Journal of Digital Imaging
|March 3, 2022
Summary
We developed a novel deep image prior (DIP) method, called double DIP (DDIP), to denoise dynamic positron emission tomography (PET) images. DDIP significantly improves parametric image quality without requiring training data.
Area of Science:
- Medical Imaging
- Computational Imaging
- Artificial Intelligence in Medicine
Background:
- Parametric imaging from dynamic positron emission tomography (PET) data quantifies tracer kinetics.
- High noise in pixel-wise time-activity curves degrades parametric image quality.
Purpose of the Study:
- To introduce a novel unsupervised deep image prior (DIP) method for denoising dynamic PET images.
- To enhance the quality of parametric PET images by reducing noise.
Main Methods:
- Proposed a deep image prior (DIP) based unsupervised denoising method for dynamic PET data.
- Introduced a modified version, double DIP (DDIP), utilizing two DIP architectures for improved input data generation.
- Evaluated performance using computer simulations.
Main Results:
- The DDIP method demonstrated superior performance compared to the single DIP method.
- DDIP combined with data augmentation yielded higher quality PET parametric images than traditional filtering methods (e.g., non-local means, high constrained backprojection).
Conclusions:
- The proposed DDIP method is an effective unsupervised approach for simultaneously denoising dynamic PET images.
- DDIP offers a promising solution for improving the quality of quantitative PET parametric imaging.
