Arterial spin labeling (ASL) is a noninvasive method for quantifying regional cerebral blood flow (CBF)^1,2. To achieve high SNR and allow background-suppression, RARE-refocused stack-of-2D-spirals (RARE-SOS) has been increasingly used in ASL imaging^3,4. As an echo-train imaging method, RARE-SOS suffers from signal dropout and T₂ blurring due to prolonged spiral trajectories and/or echo trains⁵ for high-resolution imaging, especially in the single- or two-shot mode to improve temporal-resolution. Parallel imaging provides an alterative to increase spatial resolution without reducing temporal resolution. In this work we attempted to measure whole-brain CBF at favorable spatial and temporal resolutions using accelerated acquisitions both in the partition direction and within each 2D partition. 5 healthy subjects (30±5 yo, 2M+3F) were imaged in a 3T whole-body (Siemens) scanner with a 32-channel coil. Accelerated 1-shot, 2-shot, and fully sampled 4-shot data sets all at 3 mm isotropic resolution along with the corresponding M0 maps were acquired, as listed in the Table. Each 2D partition (i.e., a “stack” in k-space) consisted of one or more spiral interleaves, each of which in turn consisted of 3 regions of different densities – fully sampled, 3X and 5X under-sampled from inner to outer k-space, as shown in Fig. 1. Acquisition in partition direction was 3X under-sampled with 13 auto-calibration partitions (centrically ordered), resulting in a total effective acceleration factor of 5.5. FOV was 216X216X144 mm³. Other acquisition details include: 90% background suppression⁶, 2-s pCASL labeling⁷, 1.5-s post-labeling delay, TE = 12.7 ms. 3D non-Cartesian SPIRiT⁸ and gridding⁹ were used for image reconstruction with the SPIRiT kernel size of 5X5X3.Figure 2 compares the single-shot magnitude images of a representative subject with and without SPIRiT reconstruction and the fully sampled 4-shot magnitude images at all three views. The SPIRiT reconstructed images (middle column) showed slightly more blurring than the 4-shot fully sampled images (right) because of the longer spiral readouts (Table). Compared to the raw images (left) the aliasing artifacts were significantly reduced. Figure 3 shows the reconstructed magnitude images, control-label different maps, and the CBF maps of the accelerated 1-shot, 2-shot, and fully sampled 4-shot images. Arrows indicate signal dropout at the orbitofrontal region of the brain. As expected from the spiral lengths listed in the Table, the 1-shot showed the largest drop-out and the 2-shot presented the least signal loss. The signal dropout in the magnitude images consequently affected the perfusion sensitivity in the orbitofrontal region (2nd and 3rd rows). In the sagittal view, the 2-shot image again showed the least through-slice blurring owing to the reduction in both spiral and echo-train lengths. In addition, the temporal resolution of the 2-shot sequence was twice as high as the fully sampled one. Results of other subjects were similar to the one shown in the Table.Although the total acceleration factor may not be as high as other recent parallel imaging reconstruction works, our method is self-calibrated, i.e., each measurement is completely self-contained and independent of additional scans and assumptions. This prevents any motion contamination of acquired data relative to pre-scan calibrations. Several aspects of the current method can potentially be improved: 1) computation of the image in each channel is time consuming; direct computation of the final combined image¹⁰ may save time of reconstruction; 2) since SPIRiT does not require regular sampling, variable-density sampling in the partition direction may further improve SNR and de-blurring; 3) since in ASL only magnitude images are used, a phase-constraint reconstruction method¹¹ may help improve image quality.A self-calibrated parallel imaging approach was developed to acquire and reconstruct highly accelerated non-Cartesian data for ASL with the whole-brain coverage at decent spatial and temporal resolutions.