Related Experiment Videos
Denoising of 4D dynamic PET images using spatiotemporal regularization with integrated temporal restoration
Hamed Yousefi1, Mahdjoub Hamdi2, Yujie Wang2
1School of Medicine, Mallinckrodt Institute of Radiology, Washington University in St. Louis, St. Louis, MO, USA. hamed.yousefi@wustl.edu.
EJNMMI Physics
|June 18, 2026
Summary
This study introduces SPRINTER, a novel self-supervised spatiotemporal denoising method for 4D dynamic PET imaging. SPRINTER significantly enhances image quality and preserves temporal fidelity, outperforming existing methods without needing ground-truth data.
Area of Science:
- Medical Imaging
- Neuroscience
- Radiochemistry
Background:
- 4D dynamic PET imaging is crucial for quantitative analysis but often suffers from noise, limiting its clinical utility.
- Noise in low-count frames compromises image quality and accuracy in dynamic PET scans.
- Existing denoising methods may introduce blurring or fail to preserve temporal dynamics.
Purpose of the Study:
- To develop and evaluate SPRINTER, a novel spatiotemporal denoising method for 4D dynamic PET imaging.
- To enhance image quality and quantitative accuracy in noisy PET data without ground-truth information.
- To preserve the fidelity of time-activity curves for improved clinical utility.
Main Methods:
- SPRINTER integrates anatomical priors, self-supervised adaptive principal component analysis (aPCA), and deep learning (3D ResUNet).
- It employs anatomically guided aPCA for temporal smoothing and a 3D ResUNet for spatial denoising, followed by temporal regularization.
- The model was trained on diverse PET/MRI brain scan datasets (220 participants, multiple radiotracers) and validated on simulations and real data.
Main Results:
- SPRINTER significantly outperformed conventional reconstruction (OSEM), spatial-only, temporal-only, NLM, and Noise2Void methods (p < 0.05).
- Marked SNR improvements were observed, especially for noisy tracers (e.g., 15O-HO, 15O-OO), with superior performance to NLM and Noise2Void.
- The method preserved and enhanced contrast (e.g., 18F-FDG, 11C-PiB) and structural fidelity (higher SSIM), indicating effective denoising without signal loss.
Conclusions:
- SPRINTER is a robust, self-supervised spatiotemporal denoising method for 4D dynamic brain PET imaging.
- It effectively enhances spatial clarity and temporal consistency using anatomical guidance and temporal dynamics.
- Its radiotracer-agnostic design and self-supervision make it suitable for clinical applications lacking high-quality reference data.
