To assess the possibility of acquiring vessel-selective time-resolved angiograms in the brain in less than one minute using accelerated Cartesian and radial readouts.

Vessel-encoded pseudo-continuous arterial spin labeling (VEPCASL)¹ can visualize the delivery of blood from individual arteries in the brain. This technique has previously been used to obtain 2D vessel-selective dynamic angiograms in ~10 minutes². Here we compare two accelerated readout schemes – one Cartesian and one radial – to demonstrate the feasibility of acquisition in approximately one minute, enabling the method to be deployed in a busy clinical protocol.

Five healthy subjects were scanned on a 3T Verio system (Siemens, Germany), with a 32-channel head coil. Two standard 3D multislab time-of-flight angiography sequences were performed at the start of each scan to localize the four main brain-feeding vessels and position the imaging slice over the circle of Willis. A unipolar VEPCASL³ pulse train was applied using the Optimized Encoding Scheme⁴. Other labeling parameters were as in Okell et al². This resulted in eight labeling cycles followed by one of three 2D bSSFP acquisitions: an accelerated Cartesian readout and two evenly undersampled radial readouts, i.e. constant azimuthal angle spacing (see parameters in Figure 1). The Cartesian readout was applied with GRAPPA factor 4 and acquired in ~63 s. The first radial readout was undersampled (80 spokes, undersampling factor versus fully sampled radial = 4.4) to match the acquisition time of the Cartesian readout. The second radial readout was further undersampled (60 spokes, undersampling factor = 5.9) to assess angiogram acquisition in less than one minute. All readouts maintained a temporal resolution of ~100 ms. All raw images were reconstructed online with Siemens software.

Image analysis was performed using a Bayesian maximum a posteriori (MAP) method⁵ to separate out vessel-specific information. The SNR in the vessels for each subject, using a somewhat heuristic metric, was then calculated as follows:

1. Mask signal from outside the brain and any large artifacts at the center (Figure 2a)

2. Create vessel-specific signal masks according to the dominant feeding artery of each voxel meeting the signal threshold in data averaged across acquisitions. The threshold was set at two-thirds of the 99th percentile of the averaged data signal, to reduce inclusion of voxels away from the main brain-feeding arteries

3. Select a noise region-of-interest from the center of the brain of each subject, avoiding any arteries, to assess noise close to vessels of interest

4. Calculate the SNR for each vessel, considering only signal within the vessel masks.

Across subjects scan times of approximately one minute and below were achieved. Figure 3a shows an example vessel-selective angiogram. Figure 3b illustrates the poorer SNR of the Cartesian acquisition in both peripheral (orange arrow) and more proximal arteries (blue arrow) versus the time-matched radial acquisition (Figure 3c). In Figure 3d increased streak artifacts towards the periphery (orange arrow) can be seen versus the more densely sampled radial acquisition (Figure 3c). However, there is still visibly better SNR (blue arrow) around more proximal vessels versus the Cartesian acquisition.

The mean SNR, averaged across all vessels and subjects, for the Cartesian acquisition (SNR = 3.07 +/- 1.76) was significantly lower than both the 80 spoke (SNR = 6.04 +/- 3.05, p~10^-5) and 60 spoke (SNR = 5.87 +/- 3.02, p~10^-4) radial acquisitions according to paired t-tests (Figure 4). There was no significant difference between the radial acquisitions.

It is possible to acquire vessel-selective angiograms in one minute with accelerated Cartesian and radial readouts. The Cartesian readout with high GRAPPA factor reduced SNR in the center of the brain around vessels of interest. Radial undersampling, however, has reduced SNR at the edges of the FOV, away from these vessels. Additionally, vessel-selective angiograms were acquired in less than one minute by further undersampling the radial acquisition. The result was angiograms with increased streak artifacts, but the peripheral arteries were better delineated than in accelerated Cartesian acquisitions (Figure 3). Cartesian and radial images suffered from sizeable artifacts around the basilar artery if it was included in the imaging slice (Figure 2), perhaps due to pulsatile blood or cerebrospinal fluid flow, which can affect the ability of the MAP analysis to correctly separate vessel information. Future work will include assessing cardiac gating for the method, to reduce artifacts around the basilar artery, and applying the accelerated radial protocols in a patient group to assess their viability in a clinical setting.