Perfusion quantification is difficult with arterial spin labeling (ASL) because of low SNR, transit-delay heterogeneity of labeled blood, and signal decay from T₁-relaxation of labeled spins. By acquiring ASL data at multiple inversion-times (TIs), however, we can better model ASL signal formation and achieve more accurate perfusion measurements. In this study, we present a tracer-kinetic model of the ASL difference signal and correct for inversion-efficiency artifacts to accurately measure renal perfusion from multi-TI ASL data. Although demonstrated with flow-sensitive alternating inversion recovery (FAIR), our approach is generally applicable to other ASL tagging schemes. We validated our proposed method against dynamic contrast-enhanced (DCE) MRI in 15 human subjects.

FAIR ASL images are acquired after alternating slice-selective (SS) and nonselective (NS) inversion pulses. Image subtraction (SS–NS) leaves only the perfusion-weighted signal difference resulting from the inflow of uninverted blood during the time period between inversion and imaging (TI).

Unequal SS and NS inversion efficiencies lead to artificial ASL signal difference that can dominate the perfusion effect. To correct for this, we fit the inversion-recovery formula to SS and NS data acquired at multiple TIs and regenerate the NS signal-vs-TI curve at the same initial magnetization as the SS curve using the fitting parameters. Free of inversion-efficiency artifacts, the ASL difference signal (dS) can be expressed as follows:$$dS(t)=\exp(-t/T_1)\int_{0}^{t}F\cdot{}AIF(\tau)\cdot{}IRF(t-\tau)\cdot{}d\tau\quad(1)$$

Here, F is perfusion, AIF is the arterial input of uninverted blood, and IRF is the impulse retention function describing the passage of blood through tissue. The exponential term accounts for signal decay from T₁-relaxation. Assuming TI is smaller than the renal minimum transit time, IRF remains unity and (1) simplifies to:$$\frac{dS(t)}{\exp(-t/T_{1})}=\begin{cases}0\quad{}&t<t_0\\M_{b0}(1+f_{ss})\cdot{}F\cdot{}(t-t_0)\quad{}&t\ge{}t_0\end{cases}\quad(2)$$

Here, M_b0 is the M₀ of blood, f_ss is the SS inversion efficiency, and t₀ is the transit delay of uninverted blood. By fitting (2) to an ASL difference signal sampled at multiple TIs, F can be determined if M_b0 and f_ss are known. Since they are relatively constant for specific data-acquisition settings, M_b0 and f_ss do not need to be evaluated for every subject.

In this IRB-approved study, renal plasma flow (RPF) was measured from fifteen subjects (9 male, 6 female; ages 24-73) using both ASL and DCE-MRI. All scans were performed at 3T (TimTrio; Siemens) after obtaining informed consent. Agreement between ASL and DCE-MRI was determined from the difference in RPF between the two techniques (ΔRPF=RPF_ASL–RPF_DCE). Correlation between the two methods was also determined.

FAIR ASL protocol: TR 3.68ms, TE 1.84ms, flip angle 180°, FOV 380x380mm, matrix 256x128, slice thickness 8mm, and GRAPPA factor of 2. Seven subjects were imaged at 16 TIs: 150ms, then 200ms–1600ms with 100ms intervals. Eight subjects were imaged at 5 TIs: 150, 500, 800, 1000, 1500ms. At each TI, SS/NS image-pairs were acquired at end-inspiration from a coronal slice through the kidneys (Figure 1a-b). After semi-automatic registration of the kidneys¹ to eliminate respiratory motion, NS and SS signal-vs-TI curves were obtained from the cortex and medulla of each kidney by averaging the signal intensity within cortical and medullary ROIs at each TI. Cortical and medullary RPF was determined from these curves according to the theory outlined above (Figure 1c-d). To investigate the impact of TI-sampling density on the accuracy of perfusion estimation, SS and NS data from the 16-TI subject group were downsampled to different TI-sampling densities by removing data points at uniform intervals along the original curves. The $$$M_{b0}(1+f_{ss})$$$ factor in (2) was set to the average ratio of $$$M_{b0}(1+f_{ss})\cdot{}F$$$ to RPF from DCE-MRI among the 16-TI subjects.

DCE-MRI protocol²: Briefly, 2D dynamic frames were acquired from the kidneys and abdominal aorta after a 4mL gadoteridol bolus-injection using a saturation-recovery 2D-turboFLASH sequence. Contrast enhancement in the aorta and the renal cortex and medulla was analyzed with a 3-compartment tracer-kinetic model³ to extract cortical and medullary RPF.

In the 16-TI subject group, RPF estimates from ASL were in good agreement with those from DCE-MRI (Table 1) when more than 2 TIs were used. ΔRPF was relatively constant, indicating that reducing TI-sampling density did not introduce bias into the perfusion estimation. The standard deviation of ΔRPF, however, increased with fewer TIs. Correlation between the two techniques was R=0.92 with the full set of 16 TIs (Figure 2a), and decreased along with the number of TIs (Figure 2b). Good agreement (ΔRPF = -9±57 mL/min) and correlation (R=0.81) was also found in the 5-TI subject group (Figure 3). In conclusion, the proposed method enables reliable correction of inversion-efficiency artifacts and robust tracer-kinetic modeling of the ASL difference signal, provided that more than two TIs are used.