Regional impact of time-to-equilibrium on indirect Patlak whole-body parametric imaging: a multi-tissue class analysis at the entire body level

Abstract

Purpose: Patlak parametric imaging is widely employed for kinetic modeling due to its simplicity and robustness. The time-to-equilibrium (t*), which must be defined to estimate kinetic parameters, is currently set empirically and uniformly across the entire body. In this study, we evaluate the regional impact of varying t* values on kinetic parameter estimates using a multi-tissue segmentation approach at the whole-body level.

Methods: Data from 53 patients who underwent one-hour dynamic 18 F-FDG PET/CT scans were retrospectively analyzed. Parametric maps of the net influx rate (K_i) and blood distribution volume (dv) were calculated for four t* values (10, 20, 30, and 45 min) using in-house software (PET KinetiX). Voxel-wise K_i and dv values were extracted from 10 predefined tissue structures through automated segmentation. Using t* = 30 min as the widely accepted reference, relative mean errors and relative absolute mean errors of K_i and dv estimated at t*shifts = 10, 20 and 45 min were calculated for each tissue. Pearson correlation coefficients between K_i or dv reference values and those estimated at t* shifts = 10, 20, and 45 min were also computed.

Results: Compared to the reference t*30, K_i estimates ranged from - 21.4% (liver) to 7.3% (SAT) at t*10, and from - 13.8% (lungs) to 2.4% (brain) at t*20. Median absolute bias was 12.8% at t*10 (6.5% brain to > 25% liver) and 8.6% at t*20 (3.2% brain to > 15% lungs and liver). At t*45, K_i was consistently overestimated, with a median bias of 19.4% (2.7% brain to > 33% lungs and liver) and median absolute bias of 19.8% (5.5% brain to > 33% lungs and liver). For dv, biases ranged from - 25.2% (brain) to 8.6% (spleen) at t*10; - 13.7% (brain) to 5.7% (lungs) at t*20; - 15.5% (liver) to 8.8% (brain) at t*45. Median absolute biases were 14.0% at t*10 (9.8% heart to 25.2% brain), 9.4% at t*20 (7.7% heart to 14.1% brain), and 15% at t*45 (12.4% skeletal muscle to 18.5% brain). Regardless of t*, K_i values exhibited strong linear correlations (r > 0.7) across all organs, whereas dv correlations showed greater variability, falling below 0.7 in 80% of organs at t*45.

Conclusion: Kinetic parameter sensitivity to time-to-equilibrium (t*) varies across organs in Patlak whole-body parametric imaging, underscoring the necessity of adopting flexible or adaptive t* values at the whole-body level.

Keywords: K0inetic modeling; PET; Parametric imaging; Patlak; Quantification; Voxel-wise.

Fig. 1

Boxplots comparing rME of Ki (A) and dv (B) estimated at equilibrium times of 10 min, 20 min, and 45 min post-injection (t*10, t*20 and t*45) versus the reference t*30. The x-axis represents the organs, while the y-axis shows the percentage bias relative to the reference t* = 30 min

Fig. 2

Boxplots comparing rMAE of K_i (A) and dv (B) estimated at equilibrium times of 10 min, 20 min, and 45 min post-injection (t*10, t*20 and t*45) versus the reference t*30. The x-axis represents the organs, while the y-axis shows the percentage bias relative to the reference t* = 30 min

Fig. 3

Voxel-wise difference maps to the reference t*30 of K_i (A) and dv (B) at the entire body level for a representative subject. Images were smoothed with a 3D gaussian filter of 3 mm