High-temporal resolution metabolic connectivity resolved by component-based noise correction

Recent advances in functional PET (fPET) enable modeling of metabolic processes with second-level temporal resolution, opening applications such as imaging molecular connectivity comparable to fMRI. However, high-temporal fPET is more noise-sensitive, making meaningful signal extraction challenging. We developed a component-based preprocessing method adapted from fMRI, which models structured noise with tissue-specific regressors and removes low-frequency uptake trends (CompCor). This approach was applied to 20 high-temporal [¹⁸F]FDG-fPET scans from a long-axial PET/CT system (1 s frames) and 16 scans from a PET/MR scanner (3 s frames). Filtering methods were compared across frequency bands, and their effects on metabolic connectivity (M-MC) assessed. Connectivity was strongly influenced by filter strategy and scanner type. CompCor produced more consistent, structured networks than standard bandpass filters. Intermediate frequency bands (0.01-0.1 Hz) gave the most reliable connectivity across PET/CT and PET/MR (r = 0.89), while high-sensitivity PET/CT also revealed structured patterns at 0.1-0.2 Hz. Compared to fMRI, fPET networks appeared more spatially cohesive but less differentiated. In sum, high-temporal [¹⁸F]FDG-fPET enables high within-scan reliability estimation of resting-state M-MC when paired with appropriate denoising, opening a new avenue in molecular imaging. Scanner characteristics and preprocessing critically affect signal quality, while our physiologically informed pipeline improves comparability across systems and studies.