A Theoretical Comparison of Multi-Delay Arterial Spin Labeling Methods

Joseph G Woods¹, Michael A Chappell², and Thomas W Okell¹

¹FMRIB Centre, NDCN, University of Oxford, Oxford, United Kingdom, ²Institute of Biomedical Engineering, University of Oxford, Oxford, United Kingdom

Synopsis

This study aims to investigate the ideal labeling and delay parameters for single-delay, multi-delay and time-encoded pCASL and compare their theoretical performance in cerebral blood flow (CBF) and bolus arrival time (BAT) estimation under realistic noise levels over two BAT ranges: healthy (500ms-1500ms) and disease (500ms-3000ms).

Introduction

Pseudo-continuous arterial spin labeling (pCASL) is a non-invasive technique that magnetically inverts blood for use as an endogenous tracer for perfusion quantification. In conventional pCASL, only a single post label delay (PLD) is used. Multiple PLDs can also be used and a suitable kinetic model fit to the data to estimate both the CBF and the bolus arrival time (BAT), with the caveat that there will be less images at each PLD to average over in a given scan time.

A recent development, time-encoded (TE) pCASL¹, splits the labeling period into blocks using a Hadamard-type encoding scheme, resulting in greater noise averaging compared to single and multiple delay methods, but the effective labeling duration of each block is shorter, resulting in a smaller signal. To counteract the effect of greater T1 decay for earlier TE blocks, Teeuwissee et al.² proposed adjusting the block durations such that the signal from each block was equal at image excitation, assuming the signal only decays with the T1 of blood (TE-T1adj).

The aim of this study is to investigate ideal labeling and delay parameters for these methods and compare their theoretical performance in cerebral blood flow (CBF) and bolus arrival time (BAT) estimation under realistic noise levels over two BAT ranges: healthy (500ms-1500ms) and disease (500ms-3000ms).

Methods

Simulations were performed using the general kinetic model³; T1 of tissue (T1t) was assumed equal to T1 of blood (T1b) for single-delay as a best-case scenario. Total scan time was kept to 5 minutes. Readout time was assumed to be 1000ms. The number of repeats available was calculated for each implementation. Noise standard deviation (SD) was calculated from multi-delay pCASL data⁴ and added to all control and tag images before subtraction or decoding. True CBF was kept constant at 60ml/100g/min. A range of labeling durations (LDs) and PLDs were tested for all methods. The TR for multi-delay varied with the PLD. A PLD of 49ms was used for both TE methods.

Perfusion was calculated directly⁵ for single delay and by least-square minimization for multi-timepoint data with a Metropolis-Hastings algorithm (500 iterations) for stable estimation of the maximum likelihood distribution; the mode of a 50-bin histogram was taken for the parameter estimates. The bounds on the estimations were CBF: 0-240 (ml/100g/min), BAT: 0-last timepoint (s).

Results and Discussion

We found that a LD of 1800ms was more accurate for multi-delay pCASL. Figure 1 shows the larger PLD range (250-2750ms) for multi-delay results in more accurate perfusion estimates for the majority of the disease BAT range while not significantly affecting the healthy range. The BAT estimates were comparable between the two methods. The drop in late BAT error for 250-1500ms is because the last time-point is earlier than for the longer range, resulting in the upper bound of the fitting algorithm reducing the overestimation of BAT.

For both TE methods, we found that a LD of 4500ms was most accurate over the disease range. Figure 2 demonstrates that 7-block TE pCASL and 11-block TE-T1adj pCASL are best for CBF and BAT estimates over the entire BAT range. The last time-points in 7-block TE-T1adj are relatively sparse so do not sample the inflow well when arrival is delayed, which results in the spike in CBF and BAT estimations.

For single-delay we found that a PLD of 2600ms and a LD of 4000ms gave the most robust results for the disease range tested, in agreement with Zun et al.⁶. This single-delay protocol was compared with the optimal multi-delay, TE and TE-T1adj protocols as described above. Figure 3 and table 1 present the results of our methods comparison. For the healthy BAT range, the CBF estimates for multi-delay had a smaller SD, but TE had more accurate BAT estimates. Single-delay performed well in the disease range because of the long PLD used, while TE-T1adj produced the greatest CBF accuracy for the multi-timepoint methods. Multi-delay had the highest BAT estimate accuracy averaged across the disease range. The SD of TE-T1adj BAT estimates increased much more than the other methods at long BATs due to the sparseness of samples at these times.

Conclusions

We have demonstrated that, theoretically, TE methods can produce CBF and BAT estimates comparable, or in certain cases better, than multi-delay pCASL. This is in contrast to a previous study7, though it should be noted that different simulated parameters have been used. We hope to validate these findings in-vivo.

Acknowledgements

Funding source: EPSRC

References

1. GuentherM. Highly efficient accelerated acquisition of perfusion inflow series by cycled arterial spin labeling. ISMRM Abstract, 2007.

2. TeeuwisseWM, Schmid S, Ghariq E, Veer IM, and van Osch MJP. Time-encoded pseudocon- tinuous arterial spin labeling: Basic properties and timing strategies for human applications. Magn Reson Med, 00:00–00, 2014.

3. Buxton RB, Frank LR, Wong EC, Siewert B, Warach S, and Edelman RR. A general kinetic model for quantitative perhsion imaging with arterial spin labeling. Magn Reson Med, 40:383– 396, 1998.

4. Okell TW, Chappell MA, Kelly ME, and Jezzard P. Cerebral blood flow quantification using vessel-encoded arterial spin labeling. J Cereb Blood Flow Metab, 33:1716–1724, 2013.

5. Alsop DC, Detre JA, Golay X, G¨ unther M, Hendrikse J, Hernandez-Garcia L, Lu H, Mac-Intosh BJ, Parkes LM, Smits M, van Osch MJP, Wang DJJ, Wong EC, and Zaharchuk G. Recommended implementation of arterial spin-labeled perfusion mri for clinical applications: A consensus of the ismrm perfusion study group and the european consortium for asl in dementia. Magn Reson Med, 000:000–000, 2014.

6. Zun, Z., Lebel, R. M., Shankaranarayanan, A. & Zaharchuk, G. What Is the Ideal Labeling Duration for Pseudocontinuous Arterial Spin Labeling? 22, 66506 (2014).

7. Johnston, M., Lu, K., Maldjian, J. & Jung, Y. Multi-TI Arterial Spin Labeling MRI with Variable TR and Bolus Duration for Cerebral Blood Flow and Arterial Transit Time Mapping. IEEE Trans. Med. Imaging 0062, 1–1 (2015).

Figures

Figure 1: Mean CBF error and standard deviation in multi-delay pCASL Two cases are shown, both with six evenly distributed PLDs: the first in the range 250ms-1500ms and the second in the range 250ms-2750ms (see legend). Extending the PLD range improved CBF estimations in the disease range without affecting the healthy range.

Figure 2: Mean CBF error and standard deviation in TE and TE-T1adj pCASL. Two cases are shown, 7 and 11 block variations (see legend). 7 block TE pCASL was more accurate than the 11 block variant for CBF and BAT estimates. The opposite was true for TE-T1adj pCASL.

Figure 3: Simulated CBF and BAT errors and standard deviations for single-delay, multi-delay, TE and TE-T1adj pCASL.

Table 1: Mean absolute errors (MAE) and mean of the standard deviations (SD) of CBF and BAT errors across BAT ranges (healthy and disease) for the ASL methods.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)

2882