Sensitivity-enhanced chemical exchange saturation transfer (CEST) MRI with least squares optimization of Carr Purcell Meiboom Gill multi-echo echo planar imaging

Abstract

Chemical exchange saturation transfer (CEST) imaging is a novel MRI technique that is sensitive to biomolecules, local pH and temperature, and offers considerable advantages for in vivo applications. However, the magnitude of CEST effect for dilute CEST agents undergoing slow or intermediate chemical exchange is typically small, requiring the use of signal averaging to enhance its sensitivity. Given that T₂-induced signal loss can be normalized by asymmetry analysis, the magnitude of CEST effect is independent of echo time. Therefore, CEST imaging with multi-echo echo planar imaging (EPI) readout should yield the same CEST effect as conventional single echo acquisition. Importantly, CEST multi-echo (CESTme) EPI images can be averaged to enhance CEST MRI sensitivity. The goal of this study was to validate CESTme EPI using a creatine-agarose gel CEST phantom with similar T₂ as biological tissue. Using least-squares optimization, we found that the sensitivity of CESTme sequence is significantly higher than that obtained by conventional single echo CEST-EPI acquisition. Specifically, signal to noise ratio (SNR) and contrast to noise ratio (CNR) from the proposed CESTme EPI were approximately equivalent to that obtained by doubling the number of signal averages of the standard single echo CEST MRI sequence. In summary, our results demonstrate CESTme EPI for sensitivity-enhanced CEST imaging.

1. INTRODUCTION

Chemical exchange saturation transfer (CEST) MRI is capable of measuring dilute labile protons, pH and temperature, and holds great promise for a host of biomedical applications (1–4). To date, CEST MRI has been applied in studies to image biomolecules, gene expression and enzyme activity, yielding novel results that have greatly enhanced our understanding of the underlying biological systems (5–9). In addition, endogenous amide proton CEST (APT) imaging is sensitive to pH and mobile protein or peptide levels, and may provide novel insights in disorders such as acute stroke, multiple sclerosis, and tumor (10–18). However, the magnitude of CEST effect is typically small, requiring that the CEST image sensitivity be enhanced for routine use (19–23).

For diamagnetic CEST (DIACEST) agents undergoing slow and intermediate chemical exchange, the CEST effect approaches its steady state exponentially (24,25). Hence, the conventional CEST MRI sequence includes long RF irradiation followed by image readout, including gradient echo, rapid acquisition with relaxation enhancement (RARE), fast low angle shot (FLASH), steady-state free precession (FISP) and echo planar imaging (EPI) (26–31). Given that signal decays following the same T₂ relaxation, asymmetry analysis normalizes the confounding T₂-decay in the control, reference, and label images, and the obtained CEST effect is independent of the echo time and T₂. Because echo planar imaging (EPI) provides reasonable spatiotemporal coverage with readout time shorter than typical in vivo T₂ values, there should be significant residual MR signal following the first echo acquisition, which has not been utilized fully by the routine CEST-EPI sequence. Our study aimed to develop and test whether CEST multi-echo (CESTme) EPI sequence provides significant sensitivity enhancement than the standard CEST single-echo EPI acquisition. To do so, we prepared a tissue-like creatine gel pH phantom, and evaluated the magnitude of the CEST effect, signal to noise (SNR) and contrast-to-noise ratio (CNR) as a function of the number of echoes and signal averages. Our results showed that the CEST effect obtained using CESTme MRI was independent of the echo time, and the use of a least-squares optimization algorithm provided significant sensitivity enhancement compared to the conventional single echo CEST-EPI sequence.

2. RESULTS

T₁ and T₂ maps showed very modest change with pH. For the respective pH 6.5 and 6.0 compartments, T₁ was 2.76 ± 0.05 s and 2.71 ± 0.05 s; T₂ measured 79.9 ± 2.1 ms and 81.7 ± 3.4 ms, respectively. Fig. 1a shows the CESTme MRI control images, whose intensity decayed with the echo number (i.e., TE) due to T₂ relaxation. Fig. 1b shows TE-dependent signals (I₀, I_ref and I_label) of the inner pH compartment. The reference images were consistently lower than the control image due to direct RF saturation. In addition, the CEST label images were lower than the reference images, giving evidence of the pH-sensitive CEST effect. Notably, there appeared to be an oscillation pattern between even and odd Carr Purcell Meiboom Gill (CPMG) echoes, which is attributable to subtle flip angle error and phase dispersion of refocusing pulses (32–34). The decay in signal intensity with respect to echo time can be described by a mono-exponential decay function (i.e., I(TE)=I*exp(-TE/T_2eff)). T_2eff was 99.1, 99.1 and 99.4 ms for the control, reference, and label images, respectively, indicating negligible effects from CEST RF saturation. This also suggested that T₂-relaxation-induced signal decay should not affect the magnitude of CEST effect resulting from normalization by asymmetry analysis. It is important to point out that T_2eff derived from CESTme MRI was slightly longer than the T₂ measured using the single spin-echo MRI sequence. This is because the CPMG sequence can more effectively suppress T₂* and background gradient-induced signal loss than the conventional single spin-echo MRI (35–37).

Fig. 1.

CESTme MRI signal as a function of echo time. a) Image intensity decays with respect to TE. b) The decay of control, reference, and labels images with respect to TE can be described by T₂ relaxation.

Fig. 2a shows CEST images reconstructed from each echo of the CESTme EPI independently (i.e., CESTR(i)=(I_ref(i) − I_label(i))/I₀(i)), where i is the ith echo number. The magnetic field homogeneity was −1.8±1.4 Hz within the slice, and no field correction was necessary (38). Because the control, reference, and label images decay at the same rate, CESTR showed very little change with TE (Fig. 2b). Specifically, CESTR as a function of echo time can be described by linear regression as CESTR(TE)=14.0 − 2.19*10⁻³xTE % and CESTR(TE)=8.19 +1.02*10⁻³x TE %, respectively for the pH 6.5 and 6.0 compartments (Fig. 2b). There was no significant correlation between CESTR and TE (P>0.15) for either pH compartment. However, the standard deviation of the CEST effect increased over TE due to T₂–induced signal decay. Indeed, whereas the contrast between the two pH compartments was independent of TE, and expressed as ΔCESTR=5.85 −3.21*10⁻³*TE % (Fig. 2b), the CNR decayed with TE with an effective time constant of 92.2 ms (Fig. 2c).

Fig. 2.

Evaluation of CEST images obtained from the proposed CESTme MRI. a) CEST-weighted images reconstructed from each of the echoes. b) CEST effect shows little change with TE. c) The CNR between two pH compartments decays with TE due to T₂-induced signal loss.

Fig. 3 shows CEST images obtained from the least-squares optimization of CESTme MRI (number of signal averages (NSA)=2). CESTR maps up to ith echo were averaged based on normalized T₂ decay coefficient (i.e., $\left({\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{N}}}{\mathrm{e}}^{-{\text{TE}}_{\mathrm{i}}/{\mathrm{T}}_2}{\ast \text{CESTR}}_{\mathrm{i}}\right)/{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{N}}}{\mathrm{e}}^{-{\text{TE}}_{\mathrm{i}}/{\mathrm{T}}_2}$). The CNR for the CESTR images was 11.0, 13.9, 15.0, 15.5 and 15.4 from the first echo alone and the least-squares optimized superposition of CEST images from the first two, three, four and five echoes, respectively (Fig. 3a). It is important to note that CNR plateaued when superimposing the first five images, with CNR 41% higher than that obtained using the conventional single echo CEST-EPI. We also compared CNR of CEST images for NSA of 1, 2 and 4. We found that the CNR gain from the proposed CESTme MRI is approximately equivalent to that obtained by doubling the number of signal averages (Fig. 3a). In addition, SNR for each pH compartment was evaluated as a function of spin echo number, with sensitivity gain similar as CNR (Supplemental Materials). We compared CEST maps from the proposed CESTme EPI and the standard single echo CEST EPI (NSA=2). For the single echo CEST-EPI map, CESTR was 14.1 ± 0.5 % and 8.2 ± 0.5 % for the 6.5 and 6.0 pH compartments, respectively (Fig. 3b). In comparison, least squares optimization of the first five echoes obtained using the proposed CESTme MRI yielded CESTR of 13.9 ± 0.3 % and 8.3 ± 0.4 %, respectively (Fig. 3c). The CNR increased from 11.0 to 15.4.

Fig. 3.

Evaluation of sensitivity gain of CESTme MRI. a) CNR from CESTme MRI increased and plateaued with the number of echoes. b) CEST image reconstructed from the first echo. c) CEST image calculated from the least-squares-optimized CESTme MRI of all five echoes.

3. DISCUSSION

Our study demonstrated that for the same scan time CESTme MRI provides significantly improved sensitivity from the conventional single echo CEST-EPI sequence. This is advantageous because the sensitivity gain from the proposed sequence can be harnessed to improve the spatiotemporal resolution, particularly important for time-sensitive studies such as acute stroke MRI. Our study demonstrated CEST multi-echo EPI in vitro using a commonly used 4.7 T preclinical Bruker scanner, and future work will translate and evaluate CESTme EPI in clinic. Notably, because endogenous amide proton CEST MRI contrast is small, it is important to balance the spatial and temporal resolution, which the proposed CESTme EPI may help address. When large image matrix size is necessary, it is necessary to minimize multi-echo acquisition time in order to maximize its sensitivity. This can be somewhat addressed by using fast acquisition strategies including partial k-space sampling, segmented acquisition, spiral EPI and compressed sensing techniques, which are beyond the scope of our current work. Interestingly, because T₂ is longer at lower field strength, T₂-induced signal attenuation is mitigated at typical clinical field strengths, which should aid the sensitivity gain of the proposed CESTme MRI. In addition, the multi-echo EPI acquisition strategy can be implemented in a host of emerging CEST sequences, including frequency-labeled exchange (FLEX), chemical exchange rotation transfer (CERT) and saturation with frequency alternating RF irradiation (SAFARI) (39–41). Moreover, improved sensitivity should facilitate ongoing development of quantitative CEST (qCEST) analysis (42–47).

The proposed CESTme MRI forms an image from every echo independently, and multi-echo images are averaged using the least squares optimization algorithm to enhance its sensitivity. This is different from CEST MRI sequences where multiple echoes are used to fill out k-space for a single image (31,48). Because CEST effect is often calculated using the asymmetry analysis, the effect of refocusing pulse on concomitant magnetization transfer contribution to CEST measurement should be reasonably small. Indeed, Zhu et al. demonstrated similar contrast between cerebellum and cerebrum between the fast low angle shot (FLASH) sequence and gradient echo and spin echo (GRASE) sequence (48). Because our study showed that sensitivity gain of CESTme EPI plateaued after five echoes (Fig. 3), we expect the effect of a relatively small number of refocusing pulses on CEST measurement should be negligible. It is important to note that because SNR of the raw images was relatively high, and the asymmetry analysis calculated the difference between the normalized reference and label images, the Rician noise can be approximated by Gaussian noise to yield a sensitivity gain despite the use of magnitude images (49). This can be improved by taking into account the phase information. However, there may be a subtle yet non-negligible phase shift among echoes, the correction of which requires sophisticated reconstruction algorithms that are beyond the scope of this current work.

4. CONCLUSIONS

Our study demonstrated that CEST images from the proposed CESTme EPI exhibited significantly enhanced sensitivity compared to images acquired using the routine single echo CEST-EPI sequence. These findings hence indicate that CESTme MRI provides a sensitive acquisition strategy that is promising to augment applications of CEST imaging.

5. EXPERIMENTAL

PHANTOM

A CEST phantom was prepared using creatine and low gelling point agarose, as described previously (4). We prepared 1.5% agarose and copper sulfate (0.65 mM)-doped phosphate-buffered saline (PBS) solution (Sigma Aldrich, St Louis, MO). The mixture was microwave heated and immersed in a water bath set at 50°C (Cole-Parmer, Vernon Hills, IL). When the temperature of the gel solution stabilized, we added creatine to reach a concentration of 50 mM, and titrated the pH to 6.5 and 6.0 (EuTech Instrument, Singapore). We then transferred the creatine gel solution into a 2-compartment concentric phantom container; pH was 6.5 and 6.0 for the inner and outer compartments, respectively. The phantom was solidified at room temperature before MRI experiments.

MRI

Images were acquired using a 4.7 Tesla small-bore scanner (Bruker Biospec, Billerica, MA). We used single-shot single-slice EPI (slice thickness = 5 mm, field of view=50×50 mm, image matrix = 64×64, bandwidth=225 kHz). For CEST imaging we probed the creatine amine proton exchange at 6.6 ppm (1.9 ppm from bulk water resonance). Fig. 4 shows the CESTme MRI sequence, which includes a long continuous-wave RF irradiation pulse, followed by CPMG multi-echo EPI acquisition (repetition time (TR)/saturation time (TS)=10/5 s, B₁=0.75 μT). We obtained five echoes with TE multiples of 24 ms. CEST experiments were repeated with the number of signal averages (NSA) varied 1, 2, and 4 times. Their scan time was 30, 60 and 120 s. In addition, T₁ map was measured using an inversion recovery sequence with six different inversion times (TI) of 0.25, 0.75, 1, 2, 3 and 5 s (TR/TE=10 s/23 ms, NSA=2), and T₂ map was derived from five separate spin-echo images with TE of 30, 60, 90, 120 and 150 ms (TR=10 s and NSA=2)(50). Furthermore, we obtained the field map using T₂*-weighted MRI (TR= 10 s and NSA=2), with asymmetric echo time shift of 1, 3, 5 and 7 ms. Images were processed using the Matlab (Mathworks, Natick, MA), and values were reported as mean ± standard deviation (SD). We calculated the CEST effect as CEST ratio (CESTR), following standard asymmetry analysis, using the equation $\text{CESTR}=({\mathrm{I}}_{\text{ref}}-{\mathrm{I}}_{\text{label}})/{\mathrm{I}}_0$, where I_label and I_ref are the label and reference images with RF irradiation applied at the labile proton frequency (Δω_s) and reference frequency (−Δω_s), respectively; I₀ is the control image without RF irradiation. Furthermore, SNR was calculated (i.e. CESTR /σ) for each pH compartment, where σ is the CESTR standard deviation of each pH compartment (i.e. σ_pH=6.5 and σ_pH=6.0). CNR was calculated as (i.e. $\frac{\mathrm{\Delta }\text{CESTR}}{\sqrt{\left({\sigma }_{\text{pH}=6.5}^2+{\sigma }_{\text{pH}=6.0}^2\right)/2}}$), where ΔCESTR is the CESTR difference between the two pH compartments.

Fig. 4.

Illustration of the proposed CESTme MRI sequence, including a long RF irradiation pulse followed by CPMG multi-echo EPI.