Development of fast multi-slice apparent T₁ (T_1app) mapping for improved arterial spin labeling MRI measurement of cerebral blood flow

Abstract

Purpose:

To develop fast multi-slice apparent T₁ (T_1app) mapping for accurate cerebral blood flow (CBF) quantification with arterial spin labeling (ASL).

Methods:

Fast multi-slice T_1app was measured using a modified inversion recovery echo planar imaging (EPI) sequence with simultaneous application of ASL tagging RF and gradient pulses. The fast multi-slice T_1app measurement was compared with the single-slice T_1app imaging approach, repeated per slice. CBF was assessed in healthy adult Wistar rats (N = 5) and acute stroke rats 24 hours after a transient middle cerebral artery occlusion (N=5).

Results:

The fast multi-slice T_1app measurement was in good agreement with that of a single-slice T_1app imaging approach (Lin’s concordance correlation coefficient (CCC) = 0.92). CBF calculated using T_1app reasonably accounts for the finite labeling RF duration, while the routine T₁-normalized ASL MRI underestimated the CBF, particularly at short labeling durations. In acute stroke rats, the labeling time and the CBF difference (ΔCBF) between the contralateral normal area and the ischemic lesion were significantly correlated when using T₁-normalized perfusion calculation (R = 0.844, P = 0.035). In comparison, T_1app-normalized ΔCBF had little labeling time dependence based on the linear regression equation of ΔCBF = −0.0247*τ + 1.579 ml/(g∙min) (R = −0.352, P = 0.494).

Conclusions:

Our study demonstrated fast multi-slice T_1app imaging, which improves the accuracy and reproducibility of CBF measurement.

Introduction

Cerebral blood flow (CBF) provides hemodynamic characterization that is often informative in neurological disorders. The arterial spin labeling (ASL) MRI allows a noninvasive CBF measurement, serving as a valuable research tool that complements dynamic susceptibility contrast perfusion imaging (1–4). Briefly, ASL MRI inverts the arterial blood signal by simultaneous application of tagging radio-frequency (RF) and gradient pulses, and the signal difference between the label and reference scans is proportional to the CBF (5–7). Continuous ASL (CASL) MRI has been commonly used in the preclinical setting, owing to its non-invasiveness that enables repeated measurement (8–10). However, the ASL tagging pulse not only saturates the arterial blood signal at the labeling plane but also induces off-resonance semisolid magnetization transfer (MT) effect at the imaging volume, particularly so when a volume coil is used for ASL tagging (11–13). Under such conditions, the apparent longitudinal relaxation time is reduced from the intrinsic one, the correction of which is necessary to improve the accuracy of CASL measurement of CBF (14,15).

CBF calculation requires either the longitudinal relaxation time without RF saturation (T₁) or that under off-resonance RF saturation (T_1app), which yields equivalent results in the case of complete saturation of semisolid macromolecules (12,16–18). However, T₁-normalized CBF quantification assumes long RF saturation, which is often not fulfilled to reduce the scan time (19). The use of a single T_1app for brain CBF calculation is overly simplistic because of heterogeneous T_1app (20). Eng et al. modified the inversion recovery sequence with off-resonance saturation to measure single-slice T_1app (21). Because the MT effect in the CASL scan is slice dependent, the T_1app mapping has to be repeated per slice and hence time-consuming. Recently, an ultrafast chemical exchange saturation transfer (CEST) magnetic resonance spectroscopy (MRS) technique has been demonstrated in phantoms (22–24), and preliminarily ex vivo and in vivo (25–27). This approach simultaneously applies RF and gradient pulses, encoding the CEST spectrum along the readout dimension. Building on the concept of superfast CEST MRS, we amended the inversion recovery sequence with simultaneous application of RF and gradient pulses that match those in the CASL scan to measure multi-slice T_1app, and demonstrated improved CBF measurement in healthy and acute stroke rats.

Theory

The kinetic model for ASL perfusion measurement has been established (5,28). Under the condition that the arterial transit time (ATT) is equal to or larger than the post-label delay (PLD), the arterial blood magnetization is given by I₀/λ, where I₀ is the tissue magnetization without RF irradiation, and λ is the brain/blood partition coefficient. CBF can be shown to be (4,29),

$$\text{CBF}=\frac{\lambda \cdot \left({\text{I}}_{\text{control}}-{\text{I}}_{\text{label}}\right)\cdot \text{exp}\left(\delta /{\text{T}}_{1\text{b}}\right)}{2\cdot \alpha \cdot {\text{T}}_{1\text{app}}\cdot {\text{I}}_0\cdot \left[1-\text{exp}\left(-\tau /{\text{T}}_{1\text{app}}\right)\right]}$$ (1)

where I_control and I_label are control and label signals, respectively, T_1b is the arterial blood longitudinal relaxation time, α is the labeling efficiency, δ is the transit time, and τ is the ASL tagging duration. Under the condition of long CASL tagging, the term 1−exp(−τ/T_1app) is approximately 1. The CBF is given as (17),

$$\text{CBF}=\frac{\lambda \cdot \left({\text{I}}_{\text{control}}^{\text{ss}}-{\text{I}}_{\text{label}}^{\text{ss}}\right)\cdot \text{exp}\left(\delta /{\text{T}}_{1\text{b}}\right)}{2\cdot \alpha \cdot {\text{T}}_{1\text{app}}\cdot {\text{I}}_0}$$ (2)

in which ${\text{I}}_{\text{control}}^{\text{ss}}$ and ${\text{I}}_{\text{label}}^{\text{ss}}$ are the steady-state control and label signals, respectively. Zhang et al. derived the steady-state effect of cross-relaxation with macromolecules on perfusion measurements (12). Under the assumption of complete saturation of macromolecules, we have,

$$\frac{1}{{\text{T}}_{1\text{app}}}=\frac{1}{{\text{T}}_{1\text{int}}}+{\text{k}}_{\text{for}}+\frac{\text{f}}{\lambda }$$ (3)

$${\text{I}}_{\text{control}}^{\text{ss}}={\text{T}}_{1\text{app}}\cdot \left(\frac{1}{{\text{T}}_{1\text{int}}}+\frac{\text{f}}{\lambda }\right)\cdot {\text{I}}_0$$ (4)

where T_1int is the intrinsic tissue longitudinal relaxation time, k_for is the exchange rate from the tissue water to the macromolecules, and T_1app depends on the RF irradiation (30). We can show that

$$\text{CBF}=\frac{\lambda \cdot \left({\text{I}}_{\text{control}}^{\text{ss}}-{\text{I}}_{\text{label}}^{\text{ss}}\right)\cdot \text{exp}\left(\delta /{\text{T}}_{1\text{b}}\right)}{2\cdot \alpha \cdot {\text{T}}_1\cdot {\text{I}}_{\text{control}}^{\text{ss}}}$$ (5)

where T₁ is tissue longitudinal relaxation time without RF saturation $\left(\text{i}.\text{e}.,\frac{1}{{\text{T}}_1}=\frac{1}{{\text{T}}_{1\text{int}}}+\frac{\text{f}}{\lambda }\right)$. Although Eq. (5) quantifies CBF under the condition of a long tagging pulse, Eq. (1) needs to be used when the tagging pulse is not sufficiently long. Importantly, Eq. (1) requires the measurements of I₀ and T_1app.

Methods

Pulse sequence

Fig. 1 shows the routine single-slice and the proposed fast multi-slice T_1app pulse sequences. The single-slice scheme applies an MT pulse at a chosen frequency offset before and during the inversion time (Fig. 1a). This is because CASL MRI induces different MT effects due to their varying distances from the labeling plane, and the single-slice T_1app sequence has to be repeated for each slice. The proposed multi-slice T_1app sequence applies the same gradient and RF pulses as those of the CASL MRI, resulting in a slice-dependent MT effect identical to that during the CASL scan (Fig. 1b). This modification allows multi-slice T_1app-weighting within a single repetition time (TR). The labeling duration and the inversion time are denoted as τ_lab and τ_inv, respectively.

Figure 1:

Illustration of the routine single-slice (Fig. 1a) and the proposed fast multi-slice T_1app (Fig. 1b) pulse sequences.

Animal preparations

The study has been approved by the local Animal Care and Use Committee. Ten adult male Wistar rats (306 ± 21 g) were examined, including five normal rats and five acute stroke rats 24 hr after 90 min transient middle cerebral artery occlusion (tMCAO). Rats were anesthetized with 1.5–2% isoflurane/air mixture with heart rate, blood oxygen saturation, and rectal temperature continuously monitored. The body temperature was maintained within the physiological range with a circulating warm water jacket. For the tMCAO model, a silicone-coated nylon filament was inserted into the right internal carotid artery to block the origin of the middle cerebral artery under the Laser Doppler Flowmetry monitoring. The filament was withdrawn 90 min post occlusion, after which rats were allowed to recover from the anesthesia. Rats that failed to demonstrate at least 70% CBF reduction from the baseline or suffered from subarachnoid hemorrhage (no CBF recovery after filament withdrawal) were excluded. MRI scans were performed 24 hr after the MCAO.

MRI experiments

All experiments were conducted on a 4.7 T MRI scanner (Bruker Biospec, Billerica, MA) with a cross coil setup (volume transmit coil and surface receiver). We used multi-slice single-shot echo-planar imaging (EPI, 2 mm/slice for 5 slices) with a field of view (FOV) of 20 × 20 mm² (matrix = 48 × 48). T₁-weighted images were acquired using an inversion recovery sequence, with inversion times of 250, 500, 750, 1000, 1500, 2000 and 2750 ms, relaxation delay/echo time (TE) = 6500/28 ms, and the number of averages (NA) = 4 (scan time = 3 min 37 s). The same relaxation delay, inversion time, TE and NA were used for the routine single-slice (scan time =26 min 50 s) and the proposed multi-slice T_1app (scan time = 5 min 22 s) sequences, with the RF labeling duration τ_lab of 5 s. Note that the routine T_1app sequence needs to be repeated for each slice, resulting in prolonged scan time. To determine an appropriate RF tagging pulse duration for fast T_1app MRI, we also repeated T_1app-weighted MRI with an RF tagging pulse of 8 s (scan time = 6 min 46 s). The amplitude modulated (AM)-ASL was performed with the following parameters: repetition time (TR)/TE = 6500/28 ms, B₁ = 4.7 μT, labeling gradient strength= 1.5 Gauss/cm, labeling distance = 16 mm, modulation frequency = 250 Hz, and post-labeling duration = 300 ms, and NA = 32 (10). For AM-ASL in normal rats, we set the RF tagging pulse duration to start from 0.5 s and then varied it from 0.6 s to 4.8 s with increments of 0.2 s to evaluate the effect of labeling time on CBF quantification (scan time from 5 min 22 s to 12 min 15 s). For AM-ASL in stroke rats, AM-ASL scans were repeated with labeling time of 0.8, 1.2, 1.5, 2, 3, and 5 s (scan time from 5 min 51 s to 12 min 34s). Also, a fast single-shot isotropic diffusion-weighted MRI was acquired with two b-values of 250 and 1000 s/mm² (TR/TE = 3250/54 ms, NA = 16, scan time = 1 min 36 s) and T₂-weighted images were obtained with two separate EPI scans (TR = 3250 ms, TE=30/100 ms, NA = 16, scan time= 1 min 40 s).

Data analysis

Images were analyzed in Matlab (Mathworks, Natick, MA). T₁ and T_1app maps were calculated by the least-squares fitting of the signal intensities as functions of the inversion time ($\text{I}={\text{I}}_0\left[1-{\beta \text{e}}^{-{\tau }_{\text{inv}}/{\text{T}}_1}\right]$ and $\text{I}={\text{I}}_{\text{ss}}\left[1-{\beta \text{e}}^{-{\tau }_{\text{inv}}/{\text{T}}_{1\text{app}}}\right]$) per pixel, in which τ_inv is the inversion time. There are three free parameters pseudo-I₀ or I_ss, β, and the parameter of interest, T₁ or T_1app. Such a three-parameter fitting routine accounts for inversion efficiency, insufficient signal recovery, and post label delay (Appendix). Parametric T_1app maps measured from the conventional single-slice and proposed multi-slice methods were denoted as T_1app and ${\text{T}}_{1\text{app}}^{\prime }$ respectively. T₂ and ADC were calculated as ${\text{T}}_2=\frac{{\text{TE}}_2-{\text{TE}}_1}{\text{ln}\left(\text{I}\left({\text{TE}}_1\right)-\text{I}\left({\text{TE}}_2\right)\right)}$ and $\text{ADC}=\frac{\text{ln}\left(\text{I}\left({\text{b}}_1\right)/\text{I}\left({\text{b}}_2\right)\right)}{{\text{b}}_2-{\text{b}}_1}$, in which TE_1,2 and b_1,2 are two echo times and b-values, respectively. We calculated CBF using both Eq. (1) and Eq. (5). The ischemic lesion was defined based on a threshold-based segmentation approach of the T₂ map, and the lesion was mirrored along the midline to define the contralateral normal area. We used Lin’s concordance correlation and Bland-Altman plots to evaluate T_1app measurements, and the Pearson correlation to analyze CBF measurements. All values were reported as mean ± standard deviation (SD), and P values less than 0.05 were considered statistically significant.

Results

We compared T_1app obtained from the routine single-slice method (T_1app), repeated per slice and the proposed fast multi-slice T_1app method $\left({\text{T}}_{1\text{app}}^{\prime }\right)$. Fig. 2a shows T_1app and ${\text{T}}_{1\text{app}}^{\prime }$ maps from a representative normal rat. Quantitative analysis of T_1app and ${\text{T}}_{1\text{app}}^{\prime }$ across five healthy rats given in Fig. 2b and 2c supports the observation that fast ${\text{T}}_{1\text{app}}^{\prime }$ closely matches T_1app (Lin’s concordance correlation coefficient (CCC) = 0.92). This finding is also confirmed by the scatter plot and Bland-Altman analysis showing small bias and 98% of the data points lying within ±2SD of the mean difference. The averaged T_1app, ${\text{T}}_{1\text{app}}^{\prime }$ and T₁ values of five slices from five healthy rats were reported in Table S1 (Supporting Information). The frequency offsets relative to the frequency of the labeling plane were matched with the slice positions. Note that T_1app is shorter than T₁ due to concomitant MT and direct RF saturation effects. The longitudinal relaxations (T_1app, ${\text{T}}_{1\text{app}}^{\prime }$ and T₁) maps also display prominent in-plane and over slice variation.

Figure 2: a).

a) Multi-slice T_1app and $T_{1app}^{\text{'}}$ maps obtained from the conventional single-slice method (repeated for each slice) and the proposed multi-slice method, respectively. b) Lin’s concordance correlation coefficient test between T_1app and $T_{1app}^{\text{'}}$, per pixel, from five slices of five healthy rats. c) The Bland-Altman analysis of T_1app and $T_{1app}^{\text{'}}$, per pixel, from five slices of five healthy rats.

Fig. 3 compares T₁- and T_1app-normalized CBF maps from normal rats. The slice was positioned 2 mm posterior from the bregma. Fig. 3a shows CBF maps from two methods under five representative labeling duration of 1.2, 1.6, 2.0, 3.0, and 4.8 s. It is worth mentioning that in the absence of T_1app, the T₁-normalized CBF calculation does not correct for the effect of insufficient ASL tagging durations. Fig. 3b shows the mean CBF values across the slice as a function of the RF labeling duration (N = 5). CBF derived from the T₁-normalized calculation steadily increased with the labeling time. In comparison, T_1app-normalized CBF values showed little dependence on the labeling time. We also investigated the saturation time dependence of CBF measurement. If we assume that the T₁-normalized CBF measurement approaches its steady-state method following an exponential correction term, we have CBF = 1.46* (1-exp(−τ/1.17)) ml/(g∙min). In comparison, T_1app-normalized CBF calculation remained relatively constant and we have CBF = (0.01*τ+ 1.50) ml/(g∙min) from a linear regression (R=0.17, P=0.051). Although the P-value (0.051) is borderline significant, this dependence is weak. For a typical labeling duration of 2–3 s, this equates to a CBF difference of 0.03 ml/(g∙min), which is no more than 2% of the CBF. Note that there is a substantial difference between T₁ and T_1app-normalized CBF values, even at the longest labeling duration. The CBF, at the longest saturation time of 4.8 s, was 1.35 and 1.58 ml/(g∙min) for T₁- and T_1app-normalized calculations, respectively, with a difference of 0.23 ml/(g∙min) (i.e., 14.6%).

Figure 3: a).

a) T₁- and T_1app-normalized CBF maps with five representative ASL labeling times of 1.2, 1.6, 2.0, 3.0, and 4.8 s. b) The effect of ASL tagging pulse duration on CBF quantification from five healthy rats. The error bars represent the inter-subject standard deviations, which were calculated from the mean values of five healthy rats.

Fig. 4 shows multi-parametric images of a representative stroke rat 24 hours after tMCAO. The ischemic lesion exhibited increased T₂ (Fig. 4a), T₁ (Fig. 4b), T_1app (Fig. 4c), and reduced ADC (Fig. 4l), as expected. Figs. 4d, 4e, 4g, and 4h show T₁- and T_1app-normalized CBF with ASL tagging RF duration of 1.5 s and 5 s, respectively. The ischemic region exhibited increased CBF when compared to the contralateral normal region, indicating post-ischemic hyperemia (31,32). Fig. 4f (CBF(T₁, τ=1.5s)/CBF(T_1app, τ=1.5 s)) and Fig. 4i (CBF(T₁, τ=5s)/CBF(T_1app, τ=5 s)) show the ratio maps of T₁- to T_1app-normalized CBF maps with the ASL tagging duration of 1.5 s and 5 s, respectively. Fig. 4f shows the ratio in the ischemia region is lower than that in the contralateral normal tissue, and more generally, regions with lower T_1app have higher ratios, such as the corpus callosum. This observation is consistent with that T₁-normalized CBF is underestimated in regions of long T_1app if the steady-state is not fulfilled. In comparison, Fig. 4i shows a relatively uniform CBF ratio throughout the slice, indicating that the underestimation of CBF(T₁) was mitigated with a relatively long labeling time (i.e., τ= 5 s). Fig. 4j (CBF(T₁, τ=1.5s)/CBF(T₁, τ=5 s)) and Fig. 4k (CBF(T_1app, τ=1.5s)/CBF(T_1app, τ=5 s)) show the ratio maps of CBF with labeling time of 1.5 s to labeling time of 5 s from the T₁- and T_1app-normalized methods. The results demonstrate that whereas T₁-normalized CBF is underestimated when using a short labeling time, T_1app-normalized CBF remains reasonably stable with respect to labeling time.

Figure 4:

Multi-parametric images from a representative acute stroke rat after 24 hr of a 90 min tMCAO. a) T₂ map, b) T₁ map, c) T_1app map, d) T₁-normalized CBF map with a tagging RF duration of 1.5 s, e) T_1app-normalized CBF map with a tagging RF duration of 1.5 s, f) The ratio image of T₁- to T_1app-normalized CBF maps with a tagging RF duration of 1.5 s, g) T₁-normalized CBF with a tagging RF duration of 5 s, h) T_1app-normalized CBF map with a tagging RF duration of 5 s, i) The ratio map of T₁- to T_1app- normalized CBF maps with a tagging RF duration of 5 s, j) The ratio image of T₁-normalized CBF maps, with that of tagging RF duration of 1.5 s over that of 5 s, k) The ratio image of T_1app-normalized CBF maps, with that of tagging duration of 1.5 s over that of 5 s, and l) ADC map.

Fig. 5 compares the T₁- and T_1app-normalized CBF between the ischemic lesion and the contralateral normal area as a function of ASL tagging RF duration (N = 5). The regional CBF values determined from the T₁-normalized method increased consistently with the ASL tagging duration (Fig. 5a), while T_1app-normalized CBF showed little dependence (Fig. 5b). Specifically, CBF (T₁, τ=5s) was 1.38±0.42 and 2.44±0.62 ml/(g∙min) for the contralateral normal region and ischemic lesion, respectively, while the corresponding CBF(T_1app, τ=5s) was 1.61±0.50 and 3.14±0.74 mg/(g∙min). CBF from the ischemic lesion is greatly increased from the contralateral normal area. Fig. 5c shows a significant positive correlation between the CBF difference (ΔCBF), between the contralateral normal tissue and ischemic lesion, and labeling time using T₁-normalized CBF calculation with corresponding linear regression equation of ΔCBF = 0.0956*τ + 0.641 ml/(g∙min) (R=0.844, P=0.035). Fig. 5d demonstrated that T_1app-normalized ΔCBF had little labeling time dependence based on the linear regression equation of ΔCBF = −0.0247*τ + 1.579 ml/(g∙min) (R=−0.352, P=0.494).

Figure 5: a).

a) T₁-normalized CBF from the contralateral normal tissue and ischemic lesion as a function of the CASL RF tagging duration, b) T_1app-normalized CBF from the contralateral normal tissue and ischemic lesion as a function of the ASL RF tagging duration, c) The T₁-normalized CBF difference (ΔCBF) between the contralateral normal tissue and ischemic lesion as a function of CASL RF tagging duration, and d) The T_1app-normalized ΔCBF as a function of CASL RF tagging duration.

Discussion

Our study developed a fast multi-slice T_1app MRI, in good agreement with that using the routine single-slice T_1app mapping approach (Fig. 2). Measurements in both healthy and stroke rats demonstrated the advantage of fast T_1app-normalized CBF quantification, which is more accurate and robust over T₁-normalized CBF mapping. Because the ischemic hyperemia is often associated with edema and blood-brain barrier disruption, accurate CBF quantification is necessary to characterize hyperemia, potentially critical for guiding thrombectomy therapy (33).

McLaughlin et al. derived a general expression for T_1app as a function of amplitude and frequency offset of the off-resonance RF irradiation based on a four-compartment exchange model, showing that T_1app measurement is important for accurate CBF quantification (18). Our work herein compared T₁– and T_1app-normalized methods for CBF quantification. The T₁–normalized method assumes complete saturation of macromolecular protons, which may not be fulfilled in practice. Under the condition of partial saturation of macromolecular protons, we have the steady-state CBF being $\frac{\lambda \cdot \left({\text{I}}_{\text{control}}^{\text{ss}}-{\text{I}}_{\text{label}}^{\text{ss}}\right)\cdot \text{exp}\left(\delta /{\text{T}}_{1\text{b}}\right)}{2\cdot \alpha \cdot {\text{T}}_1\cdot ({\text{I}}_{\text{control}}^{\text{ss}}-{\text{T}}_{1\text{app}}\cdot {\text{k}}_{\text{rev}}\cdot {\text{I}}_{\text{m}})}$, where k_rev is the reverse MT exchange rate, I_m is the unsaturated semisolid macromolecular magnetization (16). It is helpful to point out that although quantitative MT analysis has been well established, the concentration and exchange rate of semisolid macromolecules are not straightforward to map. Therefore, it is not trivial to estimate T_1app (34,35). Partial saturation of semisolid macromolecules may also introduce a CBF underestimation using a T₁-normalized approach even when the ASL labeling time is sufficiently long. We found T₁-normalized CBF is 14.6% lower than T_1app-normalized CBF, at the longest RF tagging pulse duration, likely attributable to incomplete semisolid macromolecule saturation. Our study directly measured multi-slice T_1app, which allows CBF quantification even in the case of partial MT saturation and finite RF tagging pulse duration.

For the T_1app MRI sequence, we assumed that the longitudinal magnetization reaches the steady-state in the presence of RF irradiation before the inversion pulse. To demonstrate this, we performed fast multi-slice T_1app mapping with two labeling durations of 5 and 8 s, with their shortest τ_prep being 2.25 and 5.25 s, respectively. A τ_prep of 5.25 s should be sufficiently long to achieve steady-state. T_1app maps determined from these two labeling durations were in good agreement (Fig. S1, Supporting Information), suggesting that a moderate labeling time of 5 s is sufficient for T_1app mapping. It is worth mentioning that several novel pulse sequence without off-resonance RF irradiation before inversion pulse have been proposed for magnetization transfer imaging and T_1app determination, and hence the corresponding signal can be modeled using $\text{I}={\eta \text{I}}_0{\text{e}}^{-\frac{{\tau }_{inv}}{{\text{T}}_{1\text{app}}}}+ {\text{I}}_{\text{ss}}\left[1-{\text{e}}^{-\frac{{\tau }_{inv}}{{\text{T}}_{1\text{app}}}}\right]$, in which η is the inversion completeness (36,37). Also, the agreement between routine T_1app and proposed T_1app suggests that the inflow effect introduced by concurrent application of gradient and RF pulses had little impact on fast T_1app measurement. This is likely because the inflow effect is much less than that of the tissue. Our study here acquired serial multi-slice EPI readout, which caused a slightly different delay between the end of RF saturation and excitation RF pulse among slices. When we compared the proposed multi-slice with the conventional single-slice T_1app method, which had the same short post labeling delay, T_1app agreed very well (Fig. 2). This agreement shows that the PLD can be fully accounted for using the three-parameter T₁ and T_1app fitting (Appendix). Nevertheless, the proposed T_1app MRI can be implemented with simultaneous multi-slice (SMS) readout to expedite the acquisition (38,39). Alternative fast imaging readouts, such as the variable flip angle method, could also be used to provide fast multi-slice T_1app imaging (40,41).

It’s helpful to discuss the difference between human and rodent brain ASL perfusion imaging. Human ASL imaging has very different transit time and spatial scales from those of rodents. In human imaging, transit times are on the order of 1.5–2 seconds, comparable to T₁ and the labeling duration in many experiments (42). As such, blood T₁ becomes critical for the quantification of cerebral perfusion (4). Also, the considerable distance between the labeling and imaging planes and frequency offsets in humans makes the MT effect drop rapidly in pseudo-Continuous ASL (pCASL). In comparison, because the transit time is short in the rodent brain, there was little or no post-labeling delay introduced, and also label spends only a short time (about 200 ms) in the blood before entering the tissue (30). Additionally, the short distance between the labeling plane and imaging plane, that is, less than 1 cm, also substantially increased the MT effects. Therefore, CBF quantification strongly depends on tissue T₁ and MT in rodent brain perfusion imaging using CASL (18). To summarize, the proposed T_1app work improves experimental perfusion imaging.

CONCLUSION

Our study developed a fast multi-slice T_1app imaging sequence for improving CBF quantification. The multi-slice T_1app approach agreed very well with the single-slice method, which enables accurate CBF measurement in both healthy and acute stroke rats, promising for preclinical acute stroke imaging research.

Supplementary Material

Supp info

Table S1：Comparison of T_1app, $T_{1app}^{\prime }$ and T₁ from each slice from the routine single-slice (repeated for each slice), the proposed multi-slice T_1app, and multi-slice T₁ MRI. Values were reported as mean ± standard deviation (N = 5). The slice position refers to the distance from the bregma.

Figure S1: a) $T_{1app}^{\prime }$ maps obtained from the proposed fast multi-slice T_1app MRI sequence, with two RF labeling times of 5 and 8 s. b) The Lin’s concordance correlation coefficient test between $T_{1app}^{\prime }$ maps with labeling time of 5 and 8 s, per pixel, from five slices of five healthy rats. c) The Bland-Altman analysis of $T_{1app}^{\prime }$ obtained with RF labeling times of 5 and 8 s.

Appendix

T_1app can be numerically solved using the formula

$$I=I_{ss}\left[1-\beta e^{-\frac{{\tau }_{inv}}{T_{1app}}}\right]$$ (A.1)

with three free parameters to describe the steady-state (I_ss), T_1app, an inversion efficiency (β ≡ 1 – η), in which η is inversion completeness (η =−1 for complete inversion). This formula, although not explicitly, accounts for the PLD effect. Briefly, the magnetization recovers towards its equilibrium state during the PLD, which can be described as

$$I=I_{ss}\left[1-\beta e^{-\frac{{\tau }_{inv}}{T_{1app}}}\right]e^{-\frac{PLD}{T_{1w}}}+I_0\left(1-e^{-\frac{PLD}{T_{1w}}}\right)$$

$$=I_0\left[1-e^{-\frac{PLD}{T_{1w}}}\left(\left(1-\frac{I_{ss}}{I_0}\right)\left(1+\frac{I_{ss}\beta }{I_0-I_{ss}}{e}^{-\frac{{\tau }_{inv}}{T_{1app}}}\right)\right)\right]$$ (A.2)

This can be re-written as

$$I=I_0\left(1-e^{-\frac{PLD}{T_{1w}}}\left(1-\frac{I_{ss}}{I_0}\right)\right)\left[1-\frac{e^{-\frac{PLD}{T_{1w}}}\left(1-\frac{I_{ss}}{I_0}\right)}{1-e^{-\frac{PLD}{T_{1w}}}\left(1-\frac{I_{ss}}{I_0}\right)}\frac{I_{ss}\beta }{I_0-I_{ss}}{e}^{-\frac{{\tau }_{inv}}{T_{1app}}}\right]$$

$$\propto I_0^{\prime }\left[1-{\beta }^{\prime }{e}^{-\frac{{\tau }_{inv}}{T_{1app}}}\right]$$ (A.3)

The same formula also applies to T₁ fitting.