Analysis of magnetization transfer (MT) influence on quantitative mapping of T₂ relaxation time

Abstract

Purpose:

Multi-echo spin-echo (MESE) protocol is the most effective tool for mapping T₂ relaxation in vivo. Still, MESE extensive use of radiofrequency pulses causes magnetization transfer (MT)-related bias of the water signal, instigated by the presence of macromolecules (MMP). Here, we analyze the effects of MT on MESE signal, alongside their impact on quantitative T₂ measurements.

Methods:

Study used 3 models: in vitro urea phantom, ex vivo horse brain, and in vivo human brain. MT ratio (MTR) was measured between single-SE and MESE protocols under different scan settings including varying echo train lengths, number of slices, and inter-slice gap. MTR and T₂ values were extracted for each model and protocol.

Results:

MT interactions biased MESE signals, and in certain settings, the corresponding T₂ values. T₂ underestimation of up to 4.3% was found versus single-SE values in vitro and up to 13.8% ex vivo, correlating with the MMP content. T₂ bias originated from intra-slice saturation of the MMP, rather than from indirect saturation in multi-slice acquisitions. MT-related signal attenuation was caused by slice crosstalk and/or partial T₁ recovery, whereas smaller contribution was caused by MMP interactions. Inter-slice gap had a similar effect on in vivo MTR (21.2%), in comparison to increasing the number of slices (18.9%).

Conclusions:

MT influences MESE protocols either by uniformly attenuating the entire echo train or by cumulatively attenuating the signal along the train. Although both processes depend on scan settings and MMP content, only the latter will cause underestimation of T₂.

1 |. INTRODUCTION

The field of quantitative MRI (qMRI) has been gaining increased attention owing to its capacity to optimally use the dynamic range of MRI contrast and provide accurate, stable, and most importantly, reproducible information. Quantitative T₂ (qT₂) relaxation is a clinically valuable parameter, with proven applicability to a wide range of pathologies including stroke, multiple sclerosis, cardiac disease, cancer detection, musculoskeletal diseases, and dysregulated iron content.^1–7

Biased T₂ assessment can occur for various experimental imperfections such as B₀ and B₁ inhomogeneities, MT interactions, flip-angle errors, water exchange between different compartments, and low SNR.^8–11 Although these imperfections affect any type of acquisition, the mapping of qT₂ in vivo is specifically challenging because of the long scan times associated with full spin-echo (SE) acquisitions (10s of min), and the inherent contamination of fast multi-echo SE (MESE) protocols by stimulated and indirect echoes. These cause MESE signals to strongly deviate from the theoretical exponential model S(t) = S₀exp[−t/T₂] leading to erroneous T₂ estimations.¹² Several qMRI techniques, which successfully overcome this limitation, have been recently developed. These are typically based on improving the MESE signal models to incorporate the effects of stimulated echoes and of the transmit field $({\text{B}}_1^{+})$ inhomogeneities^13–17 or on avoiding the use of spin-echo-based protocols altogether.^18,19 Further to improving the accuracy of measured T₂, these techniques also aim to produce values that are reproducible across scanners and scan settings through the use of full Bloch simulations of the specific MRI pulse sequence being used for data collection.^20,21

Another factor that can bias the measurement of T₂ is the transfer of magnetization (MT) between the macromolecular pool (MMP) and the free water pool. The incidental saturation of the MMP and subsequent MT can be instigated by the train of refocusing RF pulses used in MESE protocols^17,22 or by the large number of RF pulses in steady-state sequences.^23,24 MT saturation (MT_SAT) effects can then occur because of cross-relaxation or chemical exchange between water molecules and the MMP and are particularly apparent in multi-slice versus single-slice acquisitions, owing to the corresponding increase in the number of RF pulses. Because of its low mobility, the MMP can have a very broad spectral line shape, causing it to be indirectly excited and saturated by RF pulses that are targeted at slices outside the slice of interest. The MMP then interacts with the observable water, causing attenuation of its signal.^25,26 Another somewhat complementary effect is the direct attenuation of the water pool associated either with water that does not fully recover between consecutive TRs or with partial excitation of neighboring slices because of imperfect slice profile. Remarkably, these effects are also classified as MT, although they involve direct saturation of the water rather than one that is mediated by the MMP. As such, these effects are considered separately, and denoted as direct-MT (MT_DIR).

Previous studies have investigated the effects of incidental MT on the assessment of transverse relaxometry and biased T₂ weighting, mainly in multi-slice imaging schemes or by applying off-resonance RF pulse to saturate the MMP.²⁷ Uddin et al.²⁸ investigated the effect of reduced echo train lengths (ETL) in multi-slice MESE scans, showing the stability of fast T₂ components in the brain when shortening the ETL. Weigel et al.²⁹ showed that for low flip angles, the contrast of 2D multi-slice TSE sequences is less affected by the number of slices and is more sensitive to T₂ weighting rather than T_1. MacKay et al.¹¹ discussed the advantage of combining MT and T₂ mapping for investigating the myelin and intra-/extracellular water pools and further showed that MT will bias the T₂ nature, depending on the length and amplitude of RF pulses being used.³⁰

The extent of MT-driven signal attenuation in single- and in multi-slice MESE remains an open question. Moreover, it is not always clear whether MT attenuation happens uniformly, or cumulatively, along the echo train. Specifically, a uniform attenuation will decrease the overall signal (and SNR), albeit leave the relative decay-pattern unaffected; cumulative attenuation, on the other hand, gradually builds up along the echo-train, thereby “accelerating” the signal decay and causing underestimation of the measured T₂ value.

Here, we present an analysis of MT effect on qT₂ values acquired using fast MESE protocol, and processed using the echo-modulation-curve (EMC) algorithm. This post-processing technique uses Bloch simulations of the MESE pulse-sequence to produce the true T₂ value of tissues (i.e., closely matching values obtained using single spin-echo [SSE] acquisitions).^15,16,20 EMC-based T₂ values were shown to be highly stable across protocol implementations, scanners, and scan settings, exhibiting, for example, inter-subject coefficient of variation (CV) of 2.5%, inter-scanner CV of 0.7%, and intra-scanner CV of 1.6% for in vivo brain.³¹ Similar stability resulted for in vitro phantoms (unpublished) and for in vivo hip cartilage scans,³² indicating the potential of integrating qT₂ in multi-center and longitudinal studies.³¹ Herein, the effects of MT on the EMC algorithm were investigated on 3 models: an in vitro urea phantom, providing controlled settings with well-defined spectral content; an ex vivo horse brain presenting a physiological MMP content, yet free of flow or motion artifacts; and in vivo human brain. Attention was given to the different influence of MT on the acquired signal versus its influence on the ensuing T₂ values, to the difference between single-SE and multi-SE scans, and to MT dependence on MESE protocol parameters, including the echo train length, the number of slices, and the inter-slice gap.

2 |. METHODS

2.1 |. Samples preparation

An MRI phantom was prepared in vitro using a urea CO(NH₂)₂ crystal powder reagent. The urea molecule is favorable for studying the effects of MT as it contains 2 amide groups with high saturation efficiency,³³ and is free from proton-proton J-coupling modulations.^25,34 We note that urea MT_SAT is expected to be dominated by chemical exchange rather by intermolecular dipole transfer.^35,36 Six caulked 100 mL tubes were filled with distilled water and urea at concentrations of 0, 0.5, 1, 2, 3, and 4 M. The 0 M tube served as a reference solution, devoid of any MT_SAT effects, yet susceptible to diffusion and MT_DIR effects. Tubes were further doped with 0.25 mM MnCl₂ to reduce the water T₁/T₂ values to physiological range. The 6 tubes were inserted into a 3.3 L water container doped with 0.2 mM of MnCl₂. To fix the tubes’ positions, 6 additional open 50 mL tubes were placed around the urea solution tubes, along with an open 100 mL tube placed at their center (Figure 1A).

FIGURE 1.

T₂-weighted images (TE = 50 ms) of the 3 models investigated in this study. (A) In vitro urea phantom, (B) ex vivo horse brain, and (C) in vivo human brain. Quantitative MTR and T₂ values were estimated for the marked ROIs, consisting of 150, 15, and 10 voxels in the urea phantom, horse brain, and human brain, respectively. ROIs in the physiological samples were marked on the T₂-weighted images as to contain as homogeneous area as possible. ROIs 1–6 in (A) correspond to urea concentrations of 0, 0.5, 1, 2, 3, and 4 M, respectively. ROIs 1–6 in (B) correspond to the cortical GM, right CN, thalamus, genu of CC, midbrain, and right putamen, respectively. ROIs 1–6 in (C) correspond to the dura matter, WM fascicles, left CN, thalamus, left GP, and genu of CC, respectively

A post mortem horse brain was used as a second MT model. Brain was initially fixed in formaldehyde and kept at 4°C 24 h before scanning. Then, the brain was extracted from the fixative solution, gently dried, and induced in phosphate-buffered saline (PBS) ×1 to remove residues of formaldehyde and negate the effect of aldehyde fixative with respect to the sample’s T₂ and MT properties.³⁷ The sample was then moved to the scanner room, maintaining a constant temperature of 20°C. Immediately before the MRI scan, the brain was immersed in proton-free FC-770 Fluorinert fluid and positioned so as to fix the ventricles and cerebellum in the transverse plane (Figure 1B).

2.2 |. MRI and MRS scans

All scans were performed on a whole-body 3T Siemens Prisma scanner (Siemens Healthineers, Erlangen, Germany). Comparing MT-related changes in SSE versus MESE protocols can be confounded by the fact that many MRI vendors implement SSE and MESE using 2 separate pulse sequence schemes (e.g., having different RF pulse shapes, crusher and/ or spoiling gradients schemes, and parallel imaging capabilities). To circumvent this difference, we modified the MESE protocol to include single-echo mode and used it for both SSE and MESE acquisitions.

Phantom scans included MRS analysis, T₁ and T₂ measurements, and estimation of MTR. Because the intensity of MT interactions depends, among other things, on the spectral bandwidth of the MMP and its shift from water, we first assessed the spectral separation between the water and urea peak in the urea phantom using STEAM+PRESS (STRESS) short echo time single voxel spectroscopy (SVS).^38–40 An in-house optimized 412 ms 6-pulse water suppression scheme was played out before excitation. STRESS scan parameters were: TR/TE = 1500/15 ms; VOI = (40 mm AP) × (20 mm RL) × (15 mm IS) = 12 cc; spectral bandwidth (BW) = 2 kHz; water suppression BW = 75 Hz; vector size = 1024; T_acq = 512 ms; NEX = 16; N_{dummy−scans} = 4. Because MESE MT_DIR depends on T₁ and T₁/T₂ ratio,^41–43 the T₁ values of the urea phantom were further mapped using a standard inversion recovery (IR) imaging protocol with TR > 5 × T₁ and 20 TIs (see Table 1 for a full list of scan parameters).

TABLE 1.

MRI scan parameters

Parameter	→	TR (s)	TE (ms)	Pixel size (mm2)	Slice thick, (mm)	BW (Hz/Px)	GRAPPA factor60	Inv./ref. angle	ETL	Total scan time (min) × avg
Protocol ↓
IRa	in vitro	13	8	1.25 × 1.25	6	200	2	180°	-	1:44 × 1
SSE	in vitro	3	10,20,...,300	1.25 × 1.25	4	200	2	180°	-	3:49 × 3
	ex vivo			0.89 × 0.89	3		1		-	9:43 × 2
MESE	in vitro	3	10	1.25 × 1.25	4	200	2	180°	30	3:49 × 3
	ex vivo	3		0.77 × 0.77	3				30	5:46 × 2
	in vivo	2.4		1.50 × 1.50	3				20	3:17 × 1

Abbreviations: BW, bandwidth; Inv., inversion; Ref., refocusing; avg, average; ETL, echo train lengths; IR, inversion recovery; MESE, multi‐echo SE; SSE, single SE; thick., thickness.

^aIR sequence used 20 TIs: (50, 125, 200, 275, 350, 390, 425, 450, 470, 500, 530, 550, 575, 600, 650, 800, 950, 1100, 1500, 2000, 5000) ms

All in vivo data were collected for a single 30-year-old healthy male volunteer, after obtaining an informed consent and under the approval of the local Helsinki and IRB committees. Reference SSE T₂ mapping scans, free from MT, were performed on the in vitro and ex vivo models, excluding the in vivo model because of the long duration required by this scan. MT effects were evaluated on the 3 assayed models: urea phantom, ex vivo horse brain, and in vivo human brain using (1) reference (MT-free) single-slice SSE scan (excluding the in vivo model), (2) single-slice MESE, and (3) series of multi-slice MESE with increasing number of slices and with increasing inter-slice gap allowing to investigate MT dependence on scan settings. All scan parameters are delineated in Table 1.

2.3 |. Data post-processing

Single voxel STRESS spectroscopic data were Fourier transformed to measure the spectral distance between the urea and water peaks. T₁ maps were generated using a standard fit of the IR images on a voxel-by-voxel basis.⁴⁴ Quantitative T₂ maps were generated using exponential fitting of the SSE data and using the EMC algorithm for the MESE data.¹⁶ Both fitting procedures were done on a voxel-by-voxel basis after normalizing each decay curve to its first echo. We note that because of the non-exponential decay pattern of MESE protocols, the intensities of the normalized decay curves started at unity, rose in the second echo because of stimulated echo, and progressed in a fluctuating decay pattern. Following normalization, decay curves were truncated below a 0.1 threshold relative to the first echo to avoid fitting noise and the contamination from Rician noise pattern.⁴⁵ Representative T₂ maps of all 3 models can be found in the Supporting Information Figure S1.

MT-related signal attenuation was quantified by comparing the signal between single-slice SSE and single-slice MESE, and between single-slice MESE and multi-slice MESE. Although the SSE signal decays exponentially, the MESE signal is governed by stimulated echoes from the second echo and on, limiting the assessment of unbiased MTR between the 2 protocols to the first echo time only. This was calculated according to MTR = (S_SSE – S_MESE) /S_SSE × 100%, where S_SSE denotes non-saturated magnetization from the SSE protocol, and S_MESE denotes the single-slice MESE signal affected by MT saturation. The second set of comparisons used single- slice MESE as a reference, and compared it to a series of multi-slice MESE scans with increasing number of slices and inter-slice gap. MTR was calculated traditionally according to^11,46: MTR = (S_{Single - slice} – S_{Multi - slice}) /S_{Single - slice} × 100%, where S_{Multi - slice} denotes the signal from the middle slice of the odd-numbered slice series, corresponding to the slice of interest, imaged in the single-slice scan. Because MTR can be measured between MESE scans separately for each echo in the series, a nominal value was calculated by averaging the MTR across the MESE echo-train. In vivo MT contrast (MTC) images were generated based on the MTR between single- and multi-slice MESE. As MTR values showed minimal variation along the echo-train, the MTR of the first-echo was used as a representative MTC value, reflecting the signal dependence on the local macromolecular content.

Mean and SD of T₁, T₂ and MTR were calculated in 6 regions-of-interest (ROIs) in the urea phantom, 6 representative ROIs in the ex vivo horse brain, and 6 representative ROIs in the human data (Figure 1). ROIs were manually segmented by a trained neuroscientist with 11 years of experience and were chosen to cover different types of tissues having a diverse range of microstructural compositions. Horse brain ROIs included the cortical gray matter (GM), right caudate nucleus (CN), thalamus, genu of corpus callosum (CC), midbrain, and right putamen. Human brain ROIs included the dura mater, white matter (WM) fascicles, left CN, thalamus, left globus pallidus (GP), and genu of CC. Aiming to factor out any tissue- related variations, small anatomic ROIs were segmented so as to minimize any heterogeneities or defragmentation of the investigated tissues.

3 |. RESULTS

Table 2 presents the T₁ and T₂ values of the urea phantom calculated from single-slice IR and SSE experiments, respectively. These values were thereafter used as a reference, because these data include no saturation of the MMP and are therefore free of MT_SAT effect. A consistent decrease in T₁ and T₂ values was observed with increasing urea concentrations, a known effect attributed to the more restricted mobility, increased viscosity,⁴⁷ and higher dipolar coupling between water molecules in the presence of macromolecules.

TABLE 2.

Urea phantom concentrations and corresponding T₁ and T₂ relaxation times

Tube no.	Urea concentration (M)	T1 (IR) (ms)	T2 (SSE) (ms)	T1/T2
1	0.0	793 ± 3	62.0 ± 0.3	12.8 ± 0.1
2	0.5	778 ± 3	60.4 ± 0.4	12.9 ± 0.1
3	1.0	763 ± 5	60.2 ± 0.3	12.7 ± 0.1
4	2.0	744 ± 2	60.1 ± 0.3	12.4 ± 0.1
5	3.0	715 ± 4	59.3 ± 0.4	12.1 ± 0.1
6	4.0	673 ± 3	58.0 ± 0.4	11.6 ± 0.1

Values denote mean ± SD, calculated for each tube within the ROIs shown in Figure 1A

The urea phantom spectrum is presented in Figure 2C exhibiting 2 peaks 128 Hz apart, one at 5.7 ppm for urea protons, and a second one at 4.7 ppm from residual water protons. To understand whether a chemical-shift related slice misregistration is expected in the presence of two proton pools, we calculated the spectral width of excited slices based on the protocol’s slice-selective gradient G_ss and slice thickness Δz values according to: SW_slice = γG_ssΔz. Given the scan settings delineated in Table 1, we found this spectral range to be 710 Hz (i.e., ±305 Hz from the slice central frequency). This indicates that the excited urea and water slices colocalize with a ~82% spatial overlap, a factor that is important for interpreting intra- and inter-slice MT interactions.

FIGURE 2.

MT related urea phantom results. (A) T₂ values for SSE and MESE exhibit inverse correlation with urea concentration because of more restricted mobility and higher dipolar coupling between water molecules in the presence of macromolecules. (B) Relative change in T₂ values between MESE and SSE, calculated using (T_2,MESE − T_2,SSE)/T_2,SSE × 100%, where T_2,SSE considered the reference, unbiased by MT_SAT. Although SSE T₂ is lower at 0 M concentration because of diffusion effects, MT dominates at higher concentrations, causing MESE values to be underestimated. (C) Water suppressed single voxel spectroscopy (SVS) for the 4 M urea tube (tube 6 in Figure 1A) showing a urea peak at 5.7 ppm and a residual water peak at 4.7 ppm

Figure 2A and B illustrate the bias in T₂ values as a result of MT when shifting from single-echo to multi-echo acquisition scheme. Similar to the finding in Table 2, a concentration-dependent reduction in T₂ is observed for both the SSE and MESE protocols, indicating the spectral broadening of the water peak in the presence of urea molecules. Although SSE is influenced by diffusion, which causes a lower apparent T₂ values (as can be seen for the MMP-free reference tube), MT produces a more dominant bias leading MESE T₂ values to be significantly lower as urea concentrations increase. Specifically, a decrease of 12.2% was measured for MESE between the reference water tube and 4 M tube versus 6.4% for SSE. The resulting dispersive graph pattern (Figure 2A) suggests that MT_SAT interactions bias the signal already in single-slice MESE acquisition. From spin-physics perspective, this means that MMP saturation gradually builds up along the echo train of MESE protocols, thereby “accelerating” the signal decay. Figure 2B presents the relative change in T₂ values between SSE and MESE, offering a quantitative measure of the effect size and the resulting shift from positive to negative ratio. Values exhibited a gradual decrease with increasing urea concentration, starting with a relative T₂ change of 2.0 ± 0.8% for the reference tube and decreasing down to −4.3 ± 1.2% for the 4 M tube.

The influence of MT on the signal intensity did not systematically translate to a change in the T₂ relaxation time. Figure 3 exemplifies this by examining the correlation between T₂ and MTR. MTR values were calculated for the first echo from single slice SSE and MESE and were adjusted in this case by changing the echo train length (ETL) in the MESE scans (Figure 3A). As can be seen, the more echoes in the train, the higher is the saturation of the MESE signal because of shorter dead time between consecutive TRs. Unpaired two-tailed t-test of the difference in MTR for increasing ETLs was statistically significant (P < 0.001), with mean MTR values of 0.55 ± 0.09%, 0.95 ± 0.15%, and 1.23 ± 0.23% for ETL of 10, 20, and 30, respectively. In contrast to the signal attenuation, the matching T₂ values remained relatively constant across different ETL values, implying that MT attenuation and MTR values remained constant across the echo train.

FIGURE 3.

MTR and T₂ dependence on the ETL in a single-slice MESE scan of the urea phantom. (A) MTR between the 1st echo of SSE and MESE increases with longer ETLs (**P < 0.0001). (B) T₂ values remain stable for increasing ETLs, indicating that uniform MT attenuation occurs across the entire MESE echo train. Error bars represent SD within the ROI marked in Figure 1A

A more endogenous representation of MT effects in MESE protocols was provided by the ex vivo horse brain having a higher macromolecular content and wider spectral width. Figure 4 presents the single slice MESE T₂ values for the 6 ROIs shown in Figure 1B, exhibiting consistent underestimation of T₂ in comparison to SSE values. Groups differences, estimated using unpaired two-tailed t-test, were statistically significant (P < 0.001) across all ROIs. MESE T₂ values were underestimated versus SSE by: cortical GM, 13.8 ± 4.7%; right CN, 5.8 ± 2.2%; thalamus, 10.4 ± 2.5%; genu of CC, 9.3 ± 7.1%; midbrain, 7.4 ± 1.4%; and putamen, 6.6 ± 1.8%. The average T₂ underestimation for all 6 ROIs was 8.9 ± 2.9%, although in the absence of non-MT reference it is impossible to isolate the contribution of MT from other factors such as diffusion. The Cohen d criterion that measures the effect size was also calculated to estimate the mean T₂ differences between MESE and SSE values. According to Cohen et al.,⁴⁸ the criterion is defined by $d=\left(\overline{T_{2,\text{SSE}}}-\overline{T_{2,\text{MESE}}}\right)/\left(\text{pooled}\text{SD}\right),$ where $\overline{T_{2,x}}$ represent the mean spatial T₂ value from protocol x. This yielded a standardized effect-size of “huge” (d > 2.0) in the cortical GM, thalamus, midbrain, and putamen and “very large” (d > 1.2) in the LCN and genu of CC.⁴⁹

FIGURE 4.

Underestimation of T₂ values in a single-slice MESE scan of the ex vivo horse brain. T₂ values were estimated within 6 representative ROIs for single- slice SSE and MESE protocols, showing consistent MT-related underestimation of MESE derived values. *P < 0.001; **P < 0.0001. Cohen’s d effect size was found to be very large (d > 1.2) in the (*) ROIs and huge (d > 2.0) in the (**) ROIs

The shift from single to multi-slice acquisitions is traditionally viewed as a major cause of MT bias. This is exemplified in Figure 5, presenting the MTC in a human brain between single- and multi-slice scans and for different number of slices and inter-slice gaps. An increased signal loss is exhibited in the MTC maps for increased number of slices, generating an image contrast, which correlated with the tissues’ macromolecular content. A reversed effect appeared for increased inter-slice gap, as both MT_DIR from overlapping slice profiles and MT_SAT from off-resonance saturation are decreased. Numeric single-slice versus multi-slice MTR and T₂ values are shown in Figure 6 for the 3 assayed models and for different number of slices and inter-slice gaps. MTR correlated positively with the number of slices, with smallest increase observed in the urea phantom because of its lower macromolecular content in relation to the physiological tissues. The mean MTR values in Figure 6 were averaged across all ROIs in each model, displaying a non-linear growth of total 2.8% in vitro, 18.5% ex vivo, and 18.9% in vivo when increasing the number of slices. We note that the in vivo brain and in vitro urea, phantom MTR compared 3 and 9 slices with zero inter- slice gap, whereas ex vivo horse brain MTR compared 3 and 7 slices with 50% gap, causing the ex vivo MTR to have a relatively lower MTR baseline. However, its similarity to the average in vivo MTR change suggests that both brain models share similar MMPs.

FIGURE 5.

In vivo human brain MTC maps. MT contrast is based on the 1st echo MTR between a single-slice and the middle slice of multi-slice scan series. (A–D) MTC for increasing number of slices (gap 0%). (E–H) MTC for increasing inter-slice gap for a 9-slice series

FIGURE 6.

Multi-slice MESE MTR (A–C) and T₂ values (D–F) as a function of increasing number of slices. Values are presented for all 3 models investigated in this study: (A and D) in vitro urea phantom; (B and E) ex vivo horse brain; and (C and F) in vivo human brain. MTR was measured between single-slice acquisition and the middle slice in a multi-slice scan series. MTR exhibits positive correlation with the number of slices and with the higher amount of macromolecules characterizing the physiological tissues. No trend was seen in the corresponding T₂ values (D–F), suggesting that MT attenuation was constant across the MESE echo train. We note the relatively low MTR baseline in the ex vivo scans (B) because of the fact these scans used a 50% inter-slice gap versus a 0% gap in the other models (A and C)

In contrast to the elevation in MTR, a negligible and inconsistent change was observed in the corresponding T₂ values (Figure 6D–F). This T₂ stability once again implies that, for all 3 models, a constant MT attenuation occurs across all echoes in the train. The level of MTR is, in turn, determined by the amount of MT_DIR and by the extent of off-resonance saturation of the MMP in the slice-of-interest when scanning other slices in the series. It is important to note that the intra-slice MT_SAT (i.e., saturation that develops across the echo train) does not contribute to the single-versus-multi slice MTR, because it occurs similarly in any MESE scan regardless of the number of acquired slices.

Figure 7 shows the dependence of MTR and T₂ on the inter-slice gap in multi-slice MESE scans. As can be seen in Figure 7A–C, increasing the gap lowers off-resonance saturation, thereby reducing the corresponding MTR levels. The MTR attenuation for the ex vivo and in vivo tissues (Figure 7B and C) was moderate and, in most cases, consistent. Contrary to that, the in vitro urea model (Figure 7A) exhibited a sharp drop in the MTR for high inter-slice gap, owing to the narrow spectral range of the MMP in this phantom and the high localization of the urea and water slices (see Figure 2C). All corresponding T₂ values (Figure 7D–F) exhibited no correlation with the inter-slice gap, suggesting that, akin to when increasing number of slices, any signal attenuation occurs similarly across the entire echo train. MTR reduction when shifting from 0% to 200% gap was averaged across all ROIs, displaying a non-linear decline of 21.6%, 11.5%, and 21.2% for the in vitro (9 slices), ex vivo (7 slices), and in vivo (9 slices) models, respectively. Note that for the in vitro phantom, a small MTR of 3.3 ± 0.2% was measured at 200% gap, most probably as a result of partial longitudinal recovery of the water molecules in the slice of interest.

FIGURE 7.

Multi-slice MESE MTR (A–C) and T₂ values (D–F) as a function of increasing inter-slice gap. Values are presented for all 3 models investigated in this study: (A and D) in vitro urea phantom; (B and E) ex vivo horse brain; and (C and F) in vivo human brain. MTR was calculated between single-slice acquisition and the middle slice in a multi-slice scan series (N_slices = 7 for ex vivo [B and E] and 9 for in vitro [A and D] and in vivo [C and F]). MTR exhibited inverse correlation with the inter-slice gap, and positive correlation with the higher macromolecular content characterizing the physiological tissues. No trend, however, was found in the corresponding T₂ values, suggesting that MT levels are constant across the MESE echo train

To gain more insight into the baseline macromolecular content of each model, normalization was applied to the MTR values in Figure 7A–C, with respect to the case of 0% inter-slice gap. This normalization allowed to factor out the dependence on scan setting and isolate the effect of increasing gap values in relation to the spectral range of the MMP. Results from all ROIs were averaged together, normalized, and are presented in Figure 8. Two very clear patterns emerge from this figure: first, the highly similar macromolecular content of the two physiological tissues (horse brain and human brain), and second, their strong disparity from the synthesized urea solution. Whereas the urea phantom exhibits a relatively low macromolecular content, the physiological tissues, although from completely different sources, produce higher MTR with very similar dependence on inter-slice gap, suggesting a similar spectral distribution of their MMPs.

FIGURE 8.

Normalized MTR values calculated for the data shown in Figure 7A–C and averaged across all ROIs. Normalization allowed us to eliminate differences relating to scan settings and isolate the MMP effect on MTR, highlighting the similarity between the two physiological tissues, as opposed to their discrepancy from the urea phantom model

4 |. DISCUSSION

4.1 |. Saturation of the MMP in MESE protocols leads to underestimation of T₂

The results presented in this article demonstrate that MT can bias MESE-based quantification of T₂ values in vivo. The extensive use of high power RF pulses in this acquisition scheme saturates the macromolecules residing in physiological samples, which, in turn, exchange their magnetization with the mobile water molecules through chemical-exchange and magnetization-transfer interactions. These two mechanisms cannot be distinguished from one another and are conjointly labeled as MT_SAT⁵⁰. Both interactions, however, involve incidental saturation of the MMP, leading to attenuation of the water signal, and in certain settings, to underestimation of the measured T₂ relaxation times. A third type of signal attenuation, linked to the extensive use of RF pulses, is the direct saturation of the water pool due to slice-profile interactions or incomplete T₁ recovery between TRs. Also categorized as MT, this latter mechanism is separately denoted as MT_DIR and was found to cause a uniform attenuation of all echoes in the train, thereby having little influence on the normalized signal decay curve and the ensuing T₂ values. From a practical point of view, multi-slice acquisitions are designed to alleviate MT_DIR effects related to slice-profile imperfections by interleaving the slice order. This means that instead of acquiring slices sequentially (i.e., ascending or descending order), the series of odd-numbered slices are acquired first and then the even-numbered slices.^{17,25,26,51–53}

Quantifying the effects of MT during an in vivo scan is highly challenging because of extensive scan-times associated with the need to acquire both MESE and a reference SSE data. The urea molecules, which appear endogenously in the body (e.g., in the urine, kidneys, and blood),⁵⁴ provided a good model for studying MT at controlled levels, having a well-defined and measurable spectral content. The urea peak is located 1 ppm away from water, translating to ~128 Hz at 3T. Because of the relatively large slice bandwidth used in most imaging experiments, this spectral difference meant that the urea–water slice misregistration is much smaller than the slice thickness, leading MT_SAT effects to emerge already in single slice acquisitions as was seen in Figure 2A and B. Specifically, the presence of urea molecules had a 2-fold influence on the water signal. First, a decrease in baseline T₂ was observed, attributed to increased dipole-dipole interactions with the 2 hydrophilic amide groups. The soluble urea molecules slow the diffusion and rapid rotation of the smaller water molecules leading to longer intermolecular coherence time and corresponding faster relaxation.⁵⁵ The second effect of urea was MT with the water, further attenuating the signal and accelerating its relaxation rate. Although the intrinsic decrease in T₂ baseline occurs already in SSE, this protocol is not affected by MT, but rather susceptible to a secondary diffusion-related attenuation because of its use of very long echo times. Juxtaposing SSE’s diffusion effect vis-à-vis the MT attenuation of MESE signals, we found the latter to be more dominant (see Figure 2), causing higher attenuation of the signal and ultimately leading to underestimation of MESE T₂ values. The higher MT saturation efficiency of amide protons in the urea molecules accentuated the MT effects in this model.³³ Although endogenous molecules are characterized by a typically lower exchange rate (e.g., mean residence time of 100s of ms for myelin)⁵⁶ and lower corresponding saturation efficiency, the overall MT effects in the in vivo models did not differ significantly from the urea phantom. We attribute this to the larger spectral range and higher fraction of the MMP in vivo. MTR was therefore analyzed similarly in all models, reflecting the overall transfer of magnetization occurring within MESE protocol timescales.

The RF train in MESE schemes constitutes the driving force for intra-slice MT_SAT, accelerating the signal attenuation, and eventually translating into underestimation of T₂. The RF shape, duration, and amplitude will significantly affect the acquired signal by governing the pattern of both stimulated echoes and MT.¹⁸ Identifying this dependence on the RF pulse properties, we compared SSE and MESE qT₂ values using a modified MESE scheme that can act as a single SE experiment, thereby factoring out any implementation differences and, more importantly, by using the EMC technique that factors out the signal contamination by stimulated echoes that otherwise mask MT effects.

4.2 |. Accumulated versus uniform saturation of MESE echo train

One of the key findings of our study was that MT-related bias of T₂ is caused by pure MT_SAT, occurring already within the confines of a single-slice MESE echo-train. As more RF pulses are applied along the train, intra-slice MMP saturation accumulates and the signal attenuation increases. This accelerates the MESE signal decay and manifests as shorter T₂ value as can be seen in the SSE versus MESE dispersion pattern in Figure 2B.

Direct water saturation (i.e., MT_DIR) also affects the signal in a single-slice acquisitions, albeit, by uniformly decreasing the steady-state magnetization across the entire decay curve. In this case, MT_DIR from on-resonance water saturation will decrease the initial available magnetization in any given TR because of partial T₁ recovery, as the TR is significantly smaller than 5×T₁. This is supported by the stability of the T₂ values for different ETLs, in contrast to the respective increase in MTR (Figure 3). The average SSE versus MESE MTR for ETL=10, 20, and 30 was found to be 0.55 ± 0.09%, 0.95 ± 0.15%, and 1.23 ± 0.23%, respectively, implying that the signal does not undergo full longitudinal recovery between consecutive TRs, thereby lowering the equilibrium magnetization when using longer ETLs.

The accumulated signal attenuation caused by MT_SAT, on the other hand, had a similar effect on all train lengths, leading to very similar T₂ values. Increasing the ETL from 10 to 20 or 30 had a minor effect on ensuing T₂ values, because the MMP is already highly saturated at the 10th echo for physiological T₂ components. Therefore, the additional saturation at later echoes, where the signal has already decayed, will not cause significant change of the decay pattern.

Interestingly, the MTR in the increasing ETL experiment inversely correlated with urea concentrations, presenting higher MTR for the reference water tube, and lower MTR for the 4 M tube. This behavior resulted from the fact that the amount of MT_DIR experienced by the slice of interest is ultimately governed by the T₁/T₂ ratio and by the absolute T₁ value.^42,43 In our case, the presence of urea affected T₁ more than T₂ (see Table 2) leading to lower MT_DIR at higher con- centration. These findings were further corroborated by the fact that MTR indeed diminished to zero when using TR > 5 × T₁ (result not shown), confirming the hypothesis that the MT effects in Figure 3A consist chiefly from water-related MT_DIR between consecutive TRs.

4.3 |. Multi-slice acquisitions attenuate the signal baseline but do not alter the corresponding T₂ values

Traditionally, MT bias is associated with explicit MT preparation or with multi-slice acquisition schemes, where the MMP in the slice-of-interest is being saturated while acquiring neighboring slices. This is generally valid when considering the MT influence on the signal amplitude and will reflect in T₂-weighted images.^26,57 MT_sat-related signal attenuation in multi-slice acquisitions occurs separately in each TR and in two modes, where only one of them biases the measured T₂. One mode develops by the on-resonance RF pulses applied during the excitation and refocusing of the investigated slice, and the second mode results from off-resonance RF pulses applied during the acquisition of other slices. The first on-resonance mode will affect the MR signal gradually as the MT_SAT develops along the RF train, decreasing the measured signal, and translating into underestimation of T₂. The second off-resonance mode, on the other hand, will not bias the measured T₂, but only decrease the initial transverse magnetization in the slice of interest. This will not change the decay pattern, therefore leaving the T₂ values unaffected.

An increase in saturation, and a corresponding elevation in MTR, was observed when shifting from single- to multi-slice acquisitions. This was well correlated with the spectral characteristics of the macromolecules in the phantom and in the representative ROIs that were chosen within the ex vivo and in vivo brains. Owing to their broad MMP spectral content, a more significant multi-slice effect was observed in the physiological tissues. As expected, positive correlation was observed with increased number of slices because of the larger number of RF pulses applied. Negative correlation was observed with increased inter-slice gap, first, because of the fact that less MT_SAT from excitation of other slices occurs in the slice-of-interest as the spectral distance between the slices increases, and second, because of reduced slice crosstalk. This was mostly apparent in the urea phantom having a limited spectral width, and causing MTR to drop almost to zero for high inter-slice gap (Figure 7A). Both the slice crosstalk (MT_DIR) and off-resonance saturation (MT_SAT) is largely circumvented in multi-slice scans by employing the interleaved slice ordering scheme.¹⁷ This reduces the direct saturation from excitation of adjacent slices and partly alleviates off resonance excitation of the MMP in the slice of interest. The MT-related signal variations for different number of slices or inter-slice gap did not translate to a change in the measured T₂ values, because all echoes in the MESE train were attenuated in a uniform manner, resulting in a similar decay pattern. As illustrated in Figures 6 and 7, this finding was stable across all models used in this study.

We note that the ex vivo experiment was executed before the in vivo scans, whereas some of the parameters were optimized when running the subsequent in vivo experiments. One example is the ETL, which was reduced to 20 in the in vivo scans, because this ETL was found to be sufficient for producing stable T₂ value (see “signal truncation” in the EMC method, under data post-processing). From SAR perspective, shortening the echo-train allowed to increase the number of slices to 9 in vivo versus 7 slices ex vivo. This parameter difference did not change the final conclusion, because both the number of slices and the inter-slice gap did not change the measured T₂ values.

4.4 |. Tissue composition and compartmentation

Apart from scan parameters and macromolecular composition, MT interactions, and the corresponding bias in T₂, are also directly influenced by tissue microarchitecture. This results from the existence of several sub-voxel compartments or from partial voluming effects at tissue edges. By using homogeneous and well-defined ROIs, rather the whole brain structures, we avoided the latter effect, while minimizing variations because of different tissue compartmentations.

As can be seen in Figure 8, the physiological samples had similar MT characteristics, with small variability occurring between different types of tissues. One frequently investigated example is white matter compartmentalization into extra-cellular water, intra-cellular water, and water trapped between myelin sheaths. As suggested by MacKay et al.,⁴³ the short and long T₂ components in this type of tissue might respond differently to MT effects, with the overall T₂ bias being dependent on the local relative fraction of the two. Another example is the coexistence of water and fat molecules within the same voxel causing a complex signal bias (e.g., fat in the muscle or hematopoietic bone marrow).^58,59 These tissues present another level of complication, as each sub-compartment can have different T₁ and T₂ relaxation times and therefore experience a different MT bias. As part of our ongoing work, we are investigating the estimation of quantitative T₂ and MT in multi-T₂ environments, aiming to characterize, and subsequently correct for, MT bias in specific tissue types and within different brain structures.

5 |. CONCLUSIONS

MT interactions have a non-negligible influence on in vivo MESE acquisitions. This manifests either as a uniform attenuation of the entire echo train, which leaves the T₂ values unaffected, or as a cumulative decrease of the signal, which translates to an underestimation of T₂ values. Our results revealed an overall MT-related T₂ bias in the range of −4.3–2.0% in vitro, and 5.8–13.8% ex vivo. Between the two mechanisms classified as MT, the cumulative, intra echo-train saturation of the MMP (i.e., MT_SAT) was found to affect T₂, whereas direct water saturation (i.e., MT_DIR) had an impact on the signal amplitude. From a practical point of view, however, it is the integrated effect of protocol settings, the macromolecular spectral-content, and tissue compartmentation that will determine the final amount of MT_SAT and MT_DIR, and whether they bias the measured T₂ values.

In the current study, we adopted the EMC algorithm as a T₂ mapping technique. This was crucial factor for disentangling the influence of stimulated echoes and MT interactions, allowing genuine analysis of the effects of MT in MESE protocols. We believe that the ensuing results can be generalized to any type of T₂ postprocessing technique, as long as it correctly incorporates the influence of stimulated echoes and $({\text{B}}_1^{+})$ inhomogeneities. Improving the overall understanding of MT mechanisms in MESE protocols will facilitate the development of new acquisition or postprocessing schemes that will eventually negate or compensate for these effects (e.g., Ehses et al.²³) and yield more accurate quantification of relaxation times. Expanding the EMC simulations to model 2- or 3- pools is highly challenging from computational point of view and is part of ongoing work. Such models are, moreover, tissue-specific as they depend on the microscopic architecture of the tissue being investigated. Our multi-slice results (Figures 6 and 7) provide some optimism as they imply that multi-slice EMC simulations are not necessarily required for modeling MT when quantifying the T₂ relaxation value, as most of the accumulated signal attenuation occurs within the confines of the echo train and the slice of interest.

Supplementary Material

Supporting Figure S1

FIGURE S1 Representative T2 maps of in vitro urea phantom (A and B), ex vivo horse brain (C and D), and in vivo human brain (E and F). SSE, single spin echo; MESE, multi echo spin echo (ETL 20 or 30). Maps are shown for different scan settings: single slice (A-C and E); multi slice with Nslices 5 and 50 inter slice gap (D); and multi slice with Nslices 9 and 100 inter slice gap (F)