Data-Driven Image-Based Protocol for Brain PET Image Harmonization.

Area of Science:

• Neuroimaging
• Medical Physics
• Radiochemistry

Background:

• Quantitative FDG-PET brain imaging faces challenges due to inter-scanner variability, hindering data comparability across research centers.
• Traditional harmonization methods often rely on phantom scans, which may not always be feasible or available for retrospective studies.

Purpose of the Study:

• To develop and validate a data-driven, image-based harmonization protocol for quantitative FDG-PET brain imaging.
• To address inter-scanner variability without the need for traditional phantom scans.

Main Methods:

• The protocol estimates scanner-specific Gaussian smoothing filters using spatially normalized FDG-PET brain images.
• Optimization of filter parameters was achieved using the structural similarity index (SSIM).
• Validation utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, including cognitively normal individuals and Alzheimer's disease patients.

Main Results:

• Robust harmonization was achieved at moderate resolutions (8 and 10 mm FWHM), with estimated filters closely matching phantom-derived ground truths (within 1.2 mm).
• At higher resolutions (6 mm FWHM), discrepancies increased up to 3 mm, particularly for high-resolution scanners like HRRT, due to noise and smaller sample sizes.
• The method effectively reduces inter-scanner variability in retrospective FDG-PET studies.

Conclusions:

• The proposed image-based harmonization protocol is a valuable tool for reducing inter-scanner variability in retrospective FDG-PET studies, especially when phantom data is absent.
• Methodological refinements are necessary to improve accuracy at finer resolutions and accommodate advanced high-resolution PET imaging technologies.