Parametric Images of Myocardial Viability Using a Single ¹⁵O-H₂O PET/CT Scan

Patient Data

Existing data from 20 patients (mean age, 61 y; age range, 34–83 y; 13 men, 7 women) with known or suspected ischemic cardiomyopathy, who had undergone both ¹⁵O-H₂O and ¹⁵O-CO scans on a stand-alone PET scanner, were used. In addition, 10 patients (mean age, 66 y; age range, 55–80 y; 5 men, 5 women) with ischemic cardiomyopathy (ejection fraction < 35%) underwent ¹⁵O-H₂O PET/CT scans. The study was approved by the institutional Medical Ethics Review Committee, and all participants gave written informed consent.

Image Acquisition

Validation of PTI Based on Fitted Blood Volume Fractions (PTI_Vb)

Arterial and venous time–activity curves (C_A(t) and C_RV(t), respectively) were obtained as described previously (17). Traditional ATF (g·mL⁻¹) images were constructed as described elsewhere (9); these were rotated to obtain short-axis images of the heart. Sixteen myocardial-segment volumes of interest were drawn manually on ATF images according to the 17-segment model of the American Heart Association, excluding the apex. This volume-of-interest template was projected onto both short-axis transmission and emission scans. Segment time–activity curves were extracted, and MBF (mL·g⁻¹·min⁻¹), PTF (g·mL⁻¹), and V_A and V_RV (both dimensionless) were obtained using nonlinear regression (NLR) of the single-tissue-compartment model, with corrections for spillover and partial-volume effects (3,16):CT(t)=PTF × MBF × CA(t)⊗e−MBFVT× t+VA× CA(t) + VRV× CRV(t),Eq. 1where V_A represents arterial blood volume and left-ventricular spillover fraction, V_RV right-ventricular spillover fraction, and V_T the partition coefficient of water (which was fixed to 0.91 mL·g⁻¹). Finally, PTI based on ¹⁵O-CO (PTI_CO) and PTI_Vb was calculated usingPTICO=PTFATF=PTF1.06 × (Txnorm−CO),Eq. 2PTIVb=PTF1.06 × (Txnorm−VA−VRV),Eq. 3in which Tx_norm (dimensionless) is the normalized transmission scan (9), CO is the normalized ¹⁵O-CO concentration, and 1.06 represents the density of blood. Correlation and agreement of PTI_Vb and PTI_CO were assessed using both linear regression with zero intercept and Bland–Altman analysis.

Parametric PET/CT Images

Parametric images were generated using a BFM implementation (11,14,15) of Equation 1, as described previously (17). Attenuation-correction images based on the LD CT scan were normalized, and parametric images of V_A and V_RV were subtracted to obtain parametric ATF_Vb (ATFs based on fitted blood volume fractions) images. PTI_Vb images were then calculated as the ratio of PTF and ATF_Vb images. ATF and PTF of voxels with a total blood volume fraction above 0.75, an ATF below 0.25, or a PTF below 0.1 were set to 0 to avoid noise-induced high PTI levels in blood vessels or outside the heart. Average segmental PTI_Vb was compared with PTI_Vb calculated from segmental time–activity curves using linear regression with zero intercept, intraclass correlation coefficient (ICC), and Bland–Altman analysis.

Simulations

Simulations were performed for both BFM and NLR using C_A(t) and C_RV(t) of a randomly selected patient imaged on the PET/CT scanner. Tissue time–activity curves C_tissue(t) were generated for MBF of 1 mL·g⁻¹·min⁻¹ and PTI_Vb levels of 0.5 and 1.0, which represent (nontransmural) scar and healthy tissue, respectively. Tx_norm was fixed to 1 and considered to be noise-free. Different levels of gaussian noise were added to C_tissue(t) (4% and 20%), representing segmental and voxel noise levels, respectively. Lower noise (1%) was added to C_A(t) and C_RV(t), as these time–activity curves are based on large volumes of interest.

Next, MBF, V_A, V_RV, and PTF were obtained using both NLR and BFM. This process was repeated 1,000 times for each combination of noise on C_A(t), C_RV(t), and C_tissue(t). Average PTI_Vb values, together with corresponding bias and coefficient of variation (COV), were calculated for each combination of noise level and PTI_Vb.