4D flow is a MRI technique characterized by long scanning times, which leads to that the acquisition of these data sets is usually confined to a small volume with restricted spatial and temporal resolution. Recently, there has been a consensus regarding the minimum requirements of the MR parameters for the acquisition of 4D flow data (1). However, due to the long scanning times, it is difficult to study the variability of the results obtained when subjected to changes of the MR parameters without having other variables influencing the results. The purpose of this work is to study the variability of different flow parameters due to changes of spatial and temporal resolutions in 4D flow acquisitions through controlled experiments using a realistic normal adult thoracic aortic phantom.

The experiments were performed using a whole-body 1.5T MR scanner (Philips Achieva, Best, Netherlands) with a 4-element phased-array body coil. The phantom model (T-S-N-005, Elastrat Sarl, Geneva, Switzerland) is used with a pulsatile MR-compatible flow pump (Simutec, London, Ontario, Canada), with the same features as explained by Urbina et al (2). We are able to control the pump in terms of flow waveform, heart rate and cardiac output, which allows us to study six different hemodynamic conditions: heart rates of 68 and 88 bpm, each one with different maximum flow rate of 200, 230 and 260 mL/s. For each hemodynamic condition, two 2D phase contrast (PC) sequences were acquired in the ascending and descending aorta (Figure 1). Also, nine 4D flow data were acquired in the entire phantom with different combinations of spatial and temporal resolutions (Table 1). Furthermore, two acquisitions with medium (2.0x2.0x2.0mm3) and low spatial resolution (1.5x1.5x1.5mm3) and with 40ms of temporal resolution were re-acquired for reproducibility analysis. Each phantom condition was scanned at different days and the data acquisitions lasted about 5 hours each session.

We compared the peak flow, mean velocity, maximum velocity between all 4D flow data with the 2D PC sequence. The analyses were performed using a commercial software GT Flow 2.2.15 (Gyrotools LLC, Zurich, Switzerland).

Figure 2 summarizes the results of peak flow, mean velocity and maximum velocity in each hemodynamic condition comparing all 4D flow data with 2D PC data at the ascending (Ao1) and descending aorta (Ao5). We observed a high variability of the results obtained when subjected to changes of the studied MR parameters, particularly sensitive to changes of the temporal resolution. Considering the 2D PC sequence as gold standard, the mean, minimum and maximum peak flow errors are showed in Table 2. We observed minimum errors when the data is acquired with high temporal resolutions and greater errors when the data was acquired with low temporal resolutions. Result of the reproducibility experiment, including the mean difference (bias) and the standard deviations of difference between the first and second measurement are summarized in Table 3.We observed a high variability of the results obtained and a greater accuracy of the flow parameters when using high temporal resolution compared to 2D PC values. From Figure 2 we observed that the temporal resolutions greatly affected the results obtained. However, for a given temporal resolution the results showed small variations for the different spatial resolutions. These findings can be appreciated for any hemodynamic condition studied. Low spatial resolution images obtained result more similar than high spatial resolutions. Result of the reproducibility studied showed an excellent agreement between the two measurements for the medium and low resolutions in Ao1 and in Ao5, showing a maximum bias of 3.2% for both spatial resolution studied.Changing the spatial and temporal resolutions in 4D flow imaging greatly affects different flow parameters that induced errors of up to 23.9%, being the 4D flow sequence particularly sensitive to changes of the temporal resolution.