MR scientists, Clinical researchers interested in accelerated flow MRIFlow MRI of the carotid arteries has emerged as an important imaging field during the last decade and has been shown valuable for the assessment of hemodynamics in the context of atherosclerosis¹. Applications of volumetric flow measurements in the carotids are nowadays still hampered by excessive scan durations and limited temporal resolutions. We hypothesize that flow measurements can be accelerated by using random undersampling of retrospectively triggered acquisitions in combination with a compressed sensing reconstruction and that validations of accelerated flow measurements can be performed using a pulsatile flow loop setup. Two different experiments were conducted on a 3T MRI scanner (Ingenia Philips Healthcare, NL, software release 5.1.8). First a phantom validation was performed using a pulsatile flow loop setup (LifeTec, Eindhoven, NL). The phantom mimics a human carotid artery and consists of two tubes in a rigid 3D printed block surrounded by agar (Figure 1). An air pressure driven pump created a pulsating water flow between 0.3-6.0 ml/s. The tubes were connected so that flow is occurring in opposite directions in the imaging plane. A conventional head coil was used for signal acquisition. In a second experiment flow measurements were performed in a healthy volunteer using a 8 channel surface receive neck coil. A transversal slice in the neck through left and right common carotid artery was chosen as a measurement plane. For both setups a 2D phase contrast (PC) sequence with a foot-head flow encoding direction was used (VENC 100 cm/s). Two different spatial resolutions were chosen for phantom and volunteer scans (Table 1). 24 cardiac phases were acquired using retrospective PPU gating for the volunteer scan and a peripheral pulse signal for the phantom, leading to a temporal resolution of 40 ms. In all experiments undersampling was performed by acquiring a subset of k-lines in phase encoding direction. The scanner software was adapted to acquire a predefined number of k-lines. K-space was fully sampled at the center within 20% of the matrix size. All remaining k-lines were undersampled randomly (Figure 2). The resulting acceleration factors were 1.5, 1.7, 2.0, 2.2, 2.5, 3.0 and 3.3. Raw data was imported to MATLAB using ReconFrame (Gyrotools, Zurich, CH). Compressed sensing (CS) reconstruction was performed using the BART toolbox². Sensitivity maps were estimated from the data³. An l1 wavelet regularization was performed⁴ using a regularization parameter of 0.01. After reconstruction a linear phase offset correction was performed on the phase difference maps using static regions as a reference. To quantify flow in the regions of interest masking was done by thresholding magnitude images. The segmentation was performed on every dataset and was used for all cardiac phases. Figure 1 shows representative velocity images as obtained in the phantom and in the volunteer. Flow curves for accelerated acquisitions did not show substantial deviations from the fully sampled measurement (Figure 2). The deviation from the fully sampled dataset was on average -0.12+-0.13 ml/s for a higher resolution (0.64x0.64 mm²) and 0.07+-0.23 ml/s for a lower resolution (0.8x0.8 mm²). No dependency between acceleration factor and the accuracy of the flow measurement with respect to the fully sampled dataset could be observed. The in vivo experiment shows successful application of CS acceleration for 2D flow in the carotid arteries at two different resolutions (Figure 3). As shown in Figure 4 no clear trend could be observed between the amount of acceleration and the deviation of the fully sampled flow curve. The deviation from the fully sampled dataset in vivo was -0.20+-0.50 ml/s for a higher resolution (0.8x0.8 mm²) and 0.03+-0.50 ml/s for a lower resolution (1x1 mm²). The variation between the different acceleration factors in vivo can only partially be explained by the undersampled acquisition. Here also physiological variations play a role in combination with the limited temporal resolution of the measurement. The phantom validation convincingly illustrates the reliability of the proposed CS undersampling strategy for flow measurements. Flow measurements can be accelerated by using random undersampling and compressed sensing reconstruction. 2D flow can be accelerated 3-fold using compressed sensing without losing accuracy in flow estimation. Future application of this strategy will focus on 4D flow measurements using the same flow loop setup for validation purposes.