Quantification of myocardial perfusion using dynamic contrast enhanced imaging provides an objective and sensitive assessment of coronary artery disease in the clinic. While many factors were found to affect the accuracy of perfusion estimates, there has been little evidence on the impact of temporal resolution. This study sheds light on this using simulated, phantom and clinical data. Simulations and phantom experiments demonstrate large deviations between measurements when the same data are sampled with different temporal resolutions, while interpolation of clinical data to uniform temporal resolution leads to significantly different perfusion rates.
Methods
Dynamic perfusion data with physiological transit times were simulated as described before.1,2 Specifically, an arterial input function (AIF) was generated by convolving a delta function with exponential decay functions and myocardial signal intensity (SI) curves were produced using the Fermi function to provide ground truth perfusion rates 1, 2, 3, 4 and 5 mL/g/min for a heart rate (HR) of 60 bpm (1 dynamic/sec). The same continuous functions were resampled at different time points to simulate data for 4 other HR: 30, 90, 120 and 150 bpm. Random Rician noise was added to simulate realistic contrast-to-noise ratios of 30 and 10 for the AIF and myocardial signals respectively, producing 1000 curves for each combination of perfusion rate and HR. Dual-bolus DCE-MRI experiments were performed with a calibrated cardiac phantom.3,4 The phantom was scanned on a 3T Siemens Biograph mMR scanner using an ECG-triggered saturation recovery TurboFLASH sequence (TR 2.1 ms, TE 1.1 ms, TI 100 ms, 10° flip angle, 1 average, 192×144 matrix, 1.46×1.46 mm2 resolution, 8 mm slice thickness). Imaging was repeated 3 times for a fixed cardiac output (3 L/min) and myocardial perfusion rate (3 mL/g/min) for 3 different simulated HR: 60, 90, 120 bpm. Additionally, dual-bolus rest scans from 5 patients with established coronary artery disease were used. Patients gave written consent for their data to be analyzed in this study (ethics approval 15/NS/0030). Perfusion rate was quantified in all data without prior treatment for temporal resolution using optimized Fermi model-based deconvolution.5,6 Data were re-analyzed following piecewise cubic Hermite interpolation7, to a fixed HR of 60 bpm. Perfusion rate was also measured on simulated and phantom data by scaling the values obtained from non-interpolated data with the ratio of true and reference heart rate of 60 bpm. A paired two-sample t-test was used to compare the mean perfusion rate (in all 3 slices) in non-interpolated and interpolated patient data. Statistical significance was set at p < 0.05.Results
Simulations show that perfusion rate is overestimated for lower HR (lower temporal resolution) and underestimated for higher temporal resolution when data are quantified without prior interpolation to the reference rate (Figure 1). For example, percentage errors range between -32.9% (5 mL/g/min) and -34.5% (1 mL/g/min) for a realistic 90 bpm. These are reduced to -11.2% and -6.2% respectively following data interpolation, or 0.7% to -1.7% respectively following scaling of estimated perfusion rates. A similar pattern was seen on phantom measurements (Figure 2). However, scaling of perfusion rates proved to be an equivalent correction method to data interpolation (for 90 bpm, the mean error of -44.7% was reduced to -17.8% and -17.1% for interpolation and scaling respectively). The mean and standard deviation of perfusion rate in clinical non-interpolated and interpolated data was 0.56 ± 0.16 mL/g/min and 0.67 ± 0.20 mL/g/min respectively, estimates that were significantly different (p = 0.036). Figure 3 shows the effect of data interpolation on an example case.Conclusion
The temporal resolution of DCE-MRI has a significant impact on estimated perfusion rates. Scaling of perfusion estimates based on HR alone does not provide perfect correction as the effect is non-linear. It is believed that all common quantification models requiring data discretization assuming fixed time intervals would be similarly affected. However, further studies are needed to prove this assumption and determine the optimal temporal resolution ensuring accurate quantification for each model. Future work will focus on the evaluation of other data interpolation methods, as well as examining the impact on stress imaging, during which HR is typically higher and thus larger perfusion differences are expected.