Kinetic parameter estimation of hepatocellular carcinoma on ¹⁸ F-FDG PET/CT based on Bayesian method

Dynamic PET/CT combined with a dual-input three-compartment model can be applied to assess the kinetic parameters of hepatocellular carcinoma (HCC). The nonlinear least squares (NLLS) method is the most common method for fitting model parameters; however, some limitations remain.

A novel Bayesian-based method was compared with the NLLS method to estimate the kinetic parameters for differentiating HCCs from background liver tissue.

The proposed Bayesian method combined a priori knowledge of the physiological range and likelihood functions of HCC lesions to obtain HCC PET/CT measurements. Metropolis-Hastings sampling was used to numerically estimate the posterior distribution. This study used 5-minute dynamic PET imaging and 1-minute static PET imaging acquired 60 min post-injection from 19 HCC lesions and 17 background liver regions.

The NLLS method indicated that k₃ (p = 0.001) and fa (p < 0.001) were higher in HCCs than in background liver tissue, while K₁ (p = 0.603), k₂ (p = 0.405), k₄ (p = 0.492), V_b (p = 0.112), and K_i (p = 0.091) were not significantly different. The Bayesian method showed that k₃ (p < 0.001), fa (p < 0.001), and K_i (p = 0.002) were higher in HCCs than in background liver tissue, while K₁ (p = 0.195), k₂ (p = 0.028), k₄ (p = 0.723), and V_b (p = 0.018) were not significantly different. For k₃ and fa, the Bayesian method showed a higher AUC value for diagnostic performance in differentiating HCCs from background liver tissue than the NLLS method (0.853 vs. 0.745 and 0.928 vs. 0.886). Additionally, the Bayesian method had smaller Akaike information criteria and residual sum of squares values, as well as fewer parameter estimation outliers, than the NLLS method.

The proposed Bayesian method can accurately and robustly estimate liver kinetic parameters, effectively distinguish between lesions and background liver tissue, and provide accurate information about the uncertainty in parameter estimation.