Long time frames in myocardial [15O]H2O analysis

Usual assumption that the average radiotracer concentration during a time frame represents the instantaneous concentration at frame middle time point ("midframe approximation") is considered appropriate if the time frames are relatively short. Since in myocardial [15O]H2O PET studies the input function is derived from LV cavity in the dynamic image, the time frames affect both tissue and input data.

With longer time frames, the time-activity curves (TACs) look smoother. Sharp TAC peaks are more or less flattened, depending not only on the length of frames but also on whether the peak resides in middle of frame or between two frames. The AUC of TACs is still correct, if frame durations are considered in calculation, but midframe approximation with dot-to-dot integration will lead to some bias in AUC.

A simplistic simulation is performed to demonstrate these effects. The data and script are available in gitlab.utu.fi/vesoik/simulations.git. First, a noise-free arterial blood TAC is computed, based on an average of [15O]H2O PET studies, at short (0.5 s) time intervals. Model developed for the analysis of dynamic [15O]H2O heart studies (Iida et al., 1988, 1991, and 1992) is used to simulate regional TACs of whole myocardium and LV cavity, setting perfusale tissue blood flow (ptMBF) to 1 (mL blood)/(mL tissue * min), perfusable tissue fraction (PTF) to 0.60, and Va (accounting for the contribution of blood in LV cavity to the moving heart muscle) to 0.25. The TACs are shown in Figure 1.

BTAC and simulated regional TACs
Figure 1. Arterial blood TAC (black), and the simulated LV cavity (blue) and muscle (red) TACs. Data is in SUV units.

The simulated TACs are then shifted in time 0-5 s, and two different time frame schedules are applied to the data: either 5 s or 10 s initial time frames are used in this simulation. Figure 2 shows how the appearance of TACs is affected by merely because of the timing of the peak, especially with longer frame times.

Frame set #1 with TACs shifted 0 s Frame set #1 with TACs shifted 1 s Frame set #1 with TACs shifted 2 s Frame set #1 with TACs shifted 3 s Frame set #1 with TACs shifted 4 s Frame set #1 with TACs shifted 5 s Frame set #2 with TACs shifted 0 s Frame set #2 with TACs shifted 1 s Frame set #2 with TACs shifted 2 s Frame set #2 with TACs shifted 3 s Frame set #2 with TACs shifted 4 s Frame set #2 with TACs shifted 5 s
Figure 2. Simulated data after two different time frame schemes are applied, first five plots with initial 5 s frames, and the following five with 10 s frames. For both cases, TACs were shifted 0-5 s to show the effect of position of the peak relative to frames on the appearance of TACs.

When myocardial perfusion is estimated from these simulated TACs using MBF model with midframe approximation (like in Carimas™), the perfusion is slightly overestimated, <1% with shorter frames and 1-2% with longer frames. Time shift has minimal effect on results despite of visual differences in TACs. With higher perfusion the relative biases would be higher.

To study the effect of measurement noise on the perfusion estimates, Gaussian variation was added to the original TACs with frequent sample times, and then the same time shift and time frame schemes were applied to the noisy curves. In this simplistic simulation only 100 noise realisations were made (see Figure 3).

Frame set #1 with noisy TACs shifted 0 s Frame set #1 with noisy TACs shifted 1 s Frame set #1 with noisy TACs shifted 2 s Frame set #1 with noisy TACs shifted 3 s Frame set #1 with noisy TACs shifted 4 s Frame set #1 with noisy TACs shifted 5 s Frame set #2 with noisy TACs shifted 0 s Frame set #2 with noisy TACs shifted 1 s Frame set #2 with noisy TACs shifted 2 s Frame set #2 with noisy TACs shifted 3 s Frame set #2 with noisy TACs shifted 4 s Frame set #2 with noisy TACs shifted 5 s
Figure 3. Similar time frame schemes and time shifts as in Figure 2 were applied to 100 noise realisations of the original data. With longer time frames the noise level is obviously lower.

Myocardial perfusion was estimated from these noisy data sets. With framing scheme #1 the mean ptMBF was 1.06±0.21, 1.06±0.20, 1.06±0.21, 1.07±0.21, 1.07±0.21, and 1.06±0.21 for the time shifts of 0-5 s. With framing scheme #2 the mean ptMBFs were 1.07±0.19, 1.07±0.19, 1.06±0.19, 1.07±0.19, 1.06±0.20, and 1.06±0.20. Note that since the added noise level in this simulation was arbitrary, the resulting levels of biases and variances are arbitrary as well. However, the simulation results indicate that noise leads to overestimation of perfusion, but similar results are obtained whether the initial frame length is 5 or 10 seconds, despite of visual differences in plotted data.


See also:



Tags: , , , , ,


Updated at: 2021-10-26
Created at: 2021-10-24
Written by: Vesa Oikonen