Characterization and Reduction of Noise in Dynamic PET Data Using Masked Volumewise Principal Component Analysis

Schematic of higher-order PC noise prenormalization.

Removal of masked volumewise (MVW) PCs from X.

Illustration of masked volumewise PCA procedure.

Arithmetic mean (A–C) and sample SD (D–F) for FBP (dashed line) and OSEM (solid line) over time in the 3 ROIs: high-activity ROI (A and D), low-activity ROI (B and E), and no-activity ROI (C and F).

Arithmetic mean (A–C) and sample SD (D–F) for FBP (dashed line) and OSEM (solid line) for different axial offsets in the 3 ROIs: high-activity ROI (A and D), low-activity ROI (B and E), and no-activity ROI (C and F).

Scaled plots of SD used by higher-order PC, from full-body study using ¹⁸F-FDG (A) and brain study using ¹¹C-Pittsburgh compound B (B). Higher-order PC 1 (□), higher-order PC 2 (⋄), and higher-order PC 3 (▵Δ) are compared with background noise (○). Index of higher-order PC corresponds to number of removed masked volumewise PCs in prenormalization step.

MSE between SD of background noise and scaled SD used by higher-order PC noise prenormalization: result from full-body study (A) and result from brain study (B). Logarithmic scale is used.

MSE in different reconstructions for different prenormalizations, compared with optimal signal . FBP is used in A and B whereas OSEM is used in C and D. Measurements are from adrenal gland VOI (A and C) and stomach VOI (B and D). Index of higher-order PC corresponds to number of removed masked volumewise PCs in prenormalization step. Logarithmic scale is used.

Slices from first 3 masked volumewise PCs from datasets reconstructed with FBP (A–C) and OSEM (D–F). Most adrenal gland and general tracer behavior is described by masked volumewise PC 1 (A and D). Masked volumewise PC 2 (B and E) describes early tracer accumulation in kidneys, whereas tracer concentration in stomach can be separated with masked volumewise PC 3 (C and F).