Guided block matching and 4-D transform domain filter projection denoising method for dynamic PET image
Lin Xin1, Weihai Zhuo1, Haikuan Liu1
1Institute of Radiation Medicine, Fudan University, 2094 Xietu Road, Shanghai, 200032, China.
EJNMMI Physics
|September 25, 2023
Summary
A new Guided Block Matching and 4-D Transform Domain Filter (GBM4D) projection method significantly improves dynamic PET imaging by reducing noise. This novel approach enhances image quality and accuracy in oncology applications.
Area of Science:
- Medical Imaging
- Nuclear Medicine
- Image Processing
Background:
- Dynamic Positron Emission Tomography (PET) is crucial in oncology for visualizing and quantifying radiotracer uptake.
- Image noise in dynamic PET, stemming from low photon counts, is more pronounced than in static PET, potentially impacting imaging quality.
- Effective denoising is essential for improving the diagnostic accuracy of dynamic PET scans.
Purpose of the Study:
- To develop and evaluate a novel denoising method, Guided Block Matching and 4-D Transform Domain Filter (GBM4D) projection, for dynamic PET image reconstruction.
- To enhance the quality of dynamic PET images by reducing noise while preserving important image features.
- To improve the accuracy of quantitative measurements derived from dynamic PET data.
Main Methods:
- The proposed GBM4D method involves Anscombe transformation of the sinogram, followed by denoising using hard thresholding and Wiener filtering.
- Key steps include guided block matching, collaborative filtering, and weighted averaging, with guided block matching applied to accumulated PET sinograms to mitigate photon count issues.
- Performance was benchmarked against wavelet, total variation, non-local means, and BM3D methods using simulations and real patient data.
Main Results:
- GBM4D demonstrated superior performance over other methods in phantom studies, achieving higher structural similarity and peak signal-to-noise ratio across all time frames.
- The method resulted in the lowest root mean square error in time-activity curves for all tissues.
- Application to real patient data confirmed GBM4D's ability to produce the highest image quality.
Conclusions:
- GBM4D exhibits robust denoising capabilities essential for dynamic PET imaging.
- The method effectively preserves image edges, crucial for accurate anatomical and functional delineation.
- Both qualitative and quantitative assessments validate GBM4D's superior temporal and spatial denoising performance in dynamic PET.
