Noise equivalent count rate dependent regularized image reconstruction of dynamic 15O-water brain PET

• This work assesses the noise properties of regularized reconstruction of dynamic 15O-water brain PET. The aim was to develop and evaluate a reconstruction method utilizing the noise equivalent counts (NEC) for time-varying setting of the regularization factor in block-sequential regularized expectation maximization reconstruction (BSREM) algorithm, to achieve constant noise levels throughout dynamic image series.Methods: This study consisted of a proof-of-concept phantom study using a Hoffman 3D brain phantom and a retrospective analysis of thirteen subjects who underwent a 10-min dynamic 15O-water PET/CT examination. A set of relationships between image noise and NEC were measured for a range of regularization factors for the Hoffman phantom and for pooled data from three subjects. These relationships were interpolated, resulting in a model describing regularization factors as a function of NEC rate for a set noise level, BSREM NEC-β. For the remaining 10 subjects, comparisons were made to a clinical standard reconstruction method using ordered subsets expectation maximization (OSEM) with 3 iterations, 16 subsets and a 5-mm filter, and to a regularized reconstruction with a single regularization factor for all frames of the dynamic acquisition. Noise level in dynamic frames and quantitative cerebral blood flow (CBF) images were evaluated as well as CBF in a centrum semiovale and grey matter volume of interest.Results: Applying BSREM NEC-β reconstruction derived from the proof-of-concept phantom study led to a stable noise level through the entire dynamic Hoffman acquisition. The noise level in dynamic clinical images using BSREM NEC-β derived from a set of clinical subject data was found to be less stable than for the Hoffman phantom, although lower and more constant noise levels were achieved than for both the clinical standard reconstruction and BSREM β 300 during the first minutes of the dynamic scans. Significant, although relatively small, differences were found for CBF in CS and grey matter, noise and CNR in CBF images for both BSREM reconstruction methods compared to the clinical standard.Conclusion: Large variations in image noise did not impact quantitative CBF values in CS and grey matter with 15O-water brain PET to any large extent. Although nearly constant noise levels were achieved during the early minutes of the scans using a NEC-dependent regularizing factor with BSREM image reconstruction, noise levels dropped for later time points and further optimization of this reconstruction method is needed.

Subject headings

MEDICIN OCH HÄLSOVETENSKAP  -- Klinisk medicin -- Radiologi och bildbehandling (hsv//swe)

MEDICAL AND HEALTH SCIENCES  -- Clinical Medicine -- Radiology, Nuclear Medicine and Medical Imaging (hsv//eng)

Keyword

Dynamic PET/CT

BSREM

image reconstruction

15O-water

regularized reconstruc-tion

Publication and Content Type

vet (subject category)

ovr (subject category)