A novel method for harmonization of PET image spatial resolution without phantoms.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Analysis

Background:

• Spatial resolution estimation is crucial in medical imaging fields like PET.
• Traditional methods using physical phantoms (e.g., Hoffman phantom) are logistically burdensome and require constant monitoring.
• Phantom data may not always be available, complicating multi-center studies and scanner comparisons.

Purpose of the Study:

• To develop a novel computational approach for estimating spatial resolution directly from human subject PET images.
• To eliminate the reliance on physical phantom data for spatial resolution assessment.
• To provide a generalizable framework applicable to various tracers and imaging modalities.

Main Methods:

• Generalization of 2D logarithmic intensity plots to the 3D Fourier domain.
• Utilizing multiple linear regression with the logarithm of the Fourier transform's squared norm as the dependent variable.
• Employing squared 3D frequencies as predictors to estimate spatial resolution coefficients.

Main Results:

• The method was validated on [18F]florbetapir amyloid PET and ADNI datasets (β-amyloid, FDG, tau PET).
• Resolution estimators ranged from 3.5 mm to 8.5 mm, showing low inter-subject variability for similar scanner configurations.
• High reproducibility (ICC > 0.985), strong cross-tracer correlations, and excellent longitudinal consistency were demonstrated.

Conclusions:

• The proposed computational method obviates the need for surrogate phantom data in spatial resolution estimation.
• This approach offers a versatile framework applicable to diverse PET tracers and other imaging modalities like SPECT.
• It simplifies and standardizes spatial resolution assessment in large-scale clinical studies.