In Arterial Spin Labeling (ASL) the high individual variability of arterial transit-times (ATT) especially in patients often leads to arterial transit-delay (ATD) artefacts¹ due to non-optimal timing of bolus-length (BL) and post-labeling delay (PLD). Timings tailored to the individual patient are highly warranted here, especially in pathologies like ischemic stroke or Moya-Moya. This also affects the free-lunch (FL) approach² in Hadamard pseudo-continuous ASL (H-pCASL), where the first Hadamard-subbolus is used as a long conventional pCASL-bolus and the remaining subboli for time-encoding (fig. 1a). Early detection of non-optimal timing of BL and PLD is possible with the recently proposed Walsh-sorted H-pCASL³ (WH-pCASL), where intermediate perfusion-weighted images (PWI) are obtained during the running measurement. However, with this technique PLD, net BL, and individual sub-bolus lengths are fixed parameters. For optimal results, these measurement parameters have to be modified during runtime. Therefore, we propose an acquisition strategy that allows automatic adaption of sub-bolus lengths during a running scan, with the aim to keep the FL-bolus as long as possible to get optimal signal, but short enough to avoid ATD-artefacts.In order to determine the optimal PLD for the FL-bolus, the timepoint at which all tissue-voxels are filled with labelled blood is identified. At this PLD the ATD-artefacts are greatly reduced without shortening the FL-bolus unnecessarily. Therefore the intermediate decoded images from WH-pCASL are assessed. They correspond to several long subboli. These are composed of short subboli whose lengths are variable, but sum up to the long BL. This degree of freedom is used to adapt the length of the first (FL-) subbolus. To decide whether the PLD of the FL-bolus length should be shortened or prolonged, the number of voxels above noise level (NoV) is determined. If in one image the NoV is higher than in another, the labelled blood already reached more voxels in that image. The NoV at the longest PLD (NoV_late) is compared with the NoV at earlier PLDs (NoV_early).
If NoV_late > NoV_early, some voxels are not yet filled and the border of the FL bolus is shifted to the left (longer FL-PLD, shorter FL-BL) (fig. 1b). If NoV_late ≤ NoV_early, the maximum number of voxels is filled. In this case the last acquisition is repeated with the FL-bolus border shifted to the right (shorter FL-PLD, longer FL-BL) to increase the FL-bolus signal.

Encoding:

For encoding an 8x8 Walsh-sorted Hadamard-matrix is used which is mirrored left to right and where the second row is acquired twice, however once with label and control interchanged. This results in a 9x8 encoding matrix. From the first three acquisitions two PWIs are computed (fig 1b), whose NoVs are subsequently compared by a Python-based algorithm. For full decoding the left-right mirrored Walsh-sorted Hadamard-matrix and the corresponding images are used in analogy to conventional H-pCASL.

Imaging:

Two healthy volunteers (age 25-26, one female) were scanned with a 3T system (Skyra, SIEMENS Healthcare) using a 16-channel head-coil. The initial SBLs were: 650, 650, 650, 650, 300, 300, 300 [ms]. A PLD of 100 ms was chosen. Two hyperbolic secant pulses were used for background suppression. For the readout a segmented 3D-GRASE⁴ readout was used (slices: 24, segments: 4, resolution: 2x2x5 mm3 (interpolated), TR: 5000 ms, TE: 19 ms).

For both subjects the initial FL-PLD was too short and the algorithm consequently reduced the FL-bolus length stepwise to increase its PLD. After two steps to the left, the eight resulting images were used to decode the images for the seven final sub-boli (fig 2). Figure 3a shows images from the initial FL-bolus and figure 3b those from the final (both with identical TIs and LUT).

The PWIs in figure 2 clearly show an increase of the NoV with increasing PLDs. This confirms the correct performance of the applied feedback algorithm. Furthermore, figure 3a presents bright spots resulting from arterial signal which are greatly reduced in figure 3b which emphasizes the possibility of automatic reduction of ATD-artefacts by the presented method. Thus, the proposed dynamic feedback algorithm successfully identified an optimized FL bolus-length.
However, the choice of decision criteria, next to NoV, provides further optimization possibilities. For instance, analysis of image histograms could be used to identify arterial artefacts.

The presented acquisition strategy for Free-Lunch H-pCASL allows an adaptation of ASL-parameters like BL and PLD during a running measurement. This way an ASL-sequence can be automatically tailored to an individual subject. Especially in clinical applications where pathologies demand parameter adaption this can considerably reduce scan-time and increase the quality of ASL-data.