Abstract
3336
Introduction: There is great interest in better understanding coronary microvascular disease using mouse models. Typical quantification requires dynamic imaging to estimate the rate constant (K1) of the tracer moving from the blood into the myocardium. Measurements made at rest and stress can be used to determine both myocardial blood flow and myocardial blood flow reserve. We previously utilized a dedicated phantom to investigate the accuracy of rate constants using our in-house kinetic modeling software. In this work, we expand our previous model to include the effect of blood volume within the myocardium. To understand the impact of blood volume on K1 uncertainty, we investigated the ability to simultaneously determine the blood volume and K1 with simulated input.
Methods: We updated our fitting software to include a blood volume fraction (V), which adds a fraction of the arterial activity concentration into the tissue concentration. The tissue and blood time-activity curves (TACs) used for fit input were generated using ideal equations with known values in MATLAB. This allowed post-fit results to be compared to known values to determine fit errors. Parameters that were varied in generating the TACs included blood volume fraction (0, 0.05, 0.1, 0.2 and 0.3), K1 (0.5, 1.5, 2.5 min-1), frame length (1, 2, 5, 10, 15, 20 seconds), FWHM of the input Gaussian (10, 20, 40 secs), and time of the injection peak relative to frame duration.
First, we studied the cubic B-splines representing the TACs. We varied the splines’ knot spacing from 1 to 80 seconds, using various combinations of the parameters above to determine which knot spacing provided robust results. Next, we evaluated the accuracy of fit results across different K1 and blood volume fraction combinations. Finally, we investigated the source of potential fit limitations. During the fitting process, V and K1 must be separated via the factor K1*(1-V), therefore we investigated the difference between the resulting factor and the true, known factor.
Results: The error of the resulting blood volume fraction was insensitive to knot spacing when the knot values ranged from approximately 5 to 20 under a variety of input conditions. For inputs having a frame length of 10 seconds, which is typical in mouse imaging, the mean minimum knot spacing was 10 seconds and that parameter was therefore used for all subsequent fits. Blood volume-fraction results have squared error approaching zero for low values of K1 and blood volume using 10 second frames, but error increased as frame length increased, especially when frame length was greater than input FWHM and at higher K1. Finally, comparing the K1*(1-V)factor from the fit to known values demonstrated accurate recovery even when the two factors were inseparable.
Conclusions: We demonstrated how fit accuracy varies across different inputs and the limitations that arise in separating the K1*(1-V)factor. Future work will evaluate which combination of inputs are necessary for accurate results when spill-over effects are incorporated. Additionally, we will apply the blood volume fraction to our phantom imaging protocol and investigate the ability to simultaneously determine K1 and V in a realistic model.