Optimization of Iron Oxide Nanoparticle Detection Using Ultrashort TE Pulse Sequences: Comparison of T₁, T₂* and Synergistic T₁-T₂* Contrast Mechanisms

Abstract

Iron oxide nanoparticles (IONPs) are used in various MRI applications as negative contrast agents. A major challenge is to distinguish regions of signal void due to IONPs from those due to low signal tissues or susceptibility artifacts. To overcome this limitation, several positive contrast strategies have been proposed. Relying on IONP T₁ shortening effects to generate positive contrast is a particularly appealing strategy since it should provide additional specificity when associated with the usual negative contrast from T₂* effects. In this paper, Ultrashort TE (UTE) imaging is shown to be a powerful technique which can take full advantage of both contrast mechanisms. Methods of comparing T₁ and T₂* contrast efficiency are described and general rules that allow optimizing IONP detection sensitivity are derived. Contrary to conventional wisdom, optimizing T₁ contrast is often a good strategy for imaging IONPs. Under certain conditions, subtraction of a later echo signal from the UTE signal not only improves IONP specificity by providing long T₂* background suppression, but also increases detection sensitivity, as it enables a synergistic combination of usually antagonist T₁ and T₂* contrasts. In vitro experiments support our theory and a molecular imaging application is demonstrated using tumor-targeted IONPs in vivo.

Introduction

Iron oxide nanoparticles (IONPs) have been used as exogenous labels in a wide variety of MRI applications (1,2). These particles are usually regarded as negative contrast agents because their strong magnetic moment causes protons diffusing through their vicinity to lose signal coherence (3,4), resulting in signal voids with commonly used T₂- and T₂*-weighted pulse sequences. A major challenge is the need to distinguish regions of signal void due to IONPs from those due to low signal tissues or susceptibility artifacts, particularly when the signal-to-noise ratio (SNR) is low (2). While a pre-contrast image can be used for comparison, the slow accumulation of IONPs in tissues limits the ability to acquire post-contrast images during the same session, requiring precise co-registration of the pre- and post-contrast images, acquired hours to days later, to permit meaningful interpretation. Alternative MR acquisition strategies are desirable that are specific to IONPs and simpler to acquire.

Several investigators have recently developed original acquisition strategies to generate positive contrast from IONPs. These techniques rely on different approaches, including: 1) exploiting the frequency shift induced by IONPs (off-resonance techniques (5–8)); 2) exploiting the perturbed magnetic field surrounding IONPs (gradient-compensation techniques (9,10)); or 3) using post-processing methods to identify these magnetic field inhomogeneities (such as susceptibility-gradient mapping (11), image cross-correlation, (12) and even quantitative susceptibility-imaging (13,14)). Interestingly, these approaches share a common feature with the usual T₂- and T₂*-based approaches, as they all exploit the magnetic field perturbations caused by IONPs. It might therefore be expected that these techniques would share similar limitations, the most likely being poor specificity regarding other sources of magnetic field perturbations or resonance frequency shift such as large-scale bulk magnetic susceptibility (BMS) effects (e.g., due to air/tissue interfaces) or imperfect B₀ shimming and/or chemical shift (e.g., from fat/lipid present in the region of interest) as demonstrated by Farrar and coworkers (6).

A method of obtaining a positive contrast from IONPs that does not rely on magnetic field perturbations, is to exploit their T₁ shortening effect using appropriate imaging sequences (15–18). However, because of the short T₂ and T₂* induced by IONPs, even at relatively low concentrations, the positive T₁ contrast signature from IONPs is usually strongly attenuated at the minimum echo time (TE) attainable using typical clinical pulse sequences. Moreover, IONP clustering and compartmentalization often occur in vivo (e.g., in liver Kupffer cells), increasing the T₂* reduction as compared to in vitro homogeneous suspensions (19–21). In addition, T₁ and T₂ effects appear to be reduced (20–23), and may quench at high concentration due to a limited water-exchange regime (24). These phenomena typically limit the ability to obtain a T₁ effect from IONPs (21–23,25). In this situation, image weighting may be uncertain because the dominant contrast mechanism varies with TE and IONP concentration as well as the particle biodistribution within the sample.

Ultrashort echo time (UTE) pulse sequences (26) have been used to probe the T₁ effect generated by IONPs (27–29) including studies at high concentration. This approach has shown great promise for in vivo applications. Indeed, the technique can reduce signal loss due to rapid signal decay of short T₂ or T₂* components. In contrast to usual “long” echo time sequences that require an echo formation to probe the MR signal, UTE sequences acquire signal directly during the free induction decay (FID) following a short RF pulse. The center-out radial k-space trajectory combined with a ramp-sampling strategy allows signal acquisition very soon after the excitation, thus minimizing the effect of rapid signal decay. Specially tailored RF pulses (30) can be used to reduce relaxation effects that occur during the RF pulse and to avoid the need of rephasing gradient lobe. On the hardware side, a fast transmit/receive switch ensures that minimal time is lost after spin excitation, prior to data acquisition. Strong gradient capabilities also allow reduction of the signal observation time to minimize blurring effects. Using this technique, typical nominal TEs of 8–12 μs can be obtained on clinical scanners (31).

Multiecho versions of UTE sequences extend this unique T₁ contrast capability by providing perfectly registered datasets acquired at latter echo times, offering complementary T₂* ¹ contrast and the opportunity to generate an original positive contrast that combine T₁ and T₂* shortening in a synergistic way. These two effects are usually opposed with conventional pulse sequences because T₁ shortening enhances MR signal while T₂* shortening reduces it. However, when the signal from a later TE image is subtracted from the signal obtained on the UTE image, long T₂* components are reduced and the UTE detection takes advantage of the T₁ effect induced by IONPs (see Fig. 1). This subtraction technique (called SubUTE here) provides long-T₂* signal suppression and, more generally, reverses T₂* contrast; it is thus considered to be a positive contrast technique. Note that the introduction of T₂* contrast provided by SubUTE makes it prone to reduced specificity regarding undesired magnetic field perturbations, similar to that of the more usual positive contrast methods described above. SubUTE has been successfully used for IONP detection by several investigators (27,32–34), although the synergistic effect has only received brief consideration (27).

Figure 1.

a) Schematic depiction of T₁ and T₂* contrast. After subtracting the long TE signal from the UTE signal, long T₂* components are reduced whereas short T₂* components are preserved. The positive contrast that results combines both T₁ and T₂* effects in a synergistic way. Corresponding contrast intensities are emphasized in b). The + sign indicates positive contrast whereas the − sign stands for negative contrast.

Expected improvements in detection specificity and sensitivity toward IONPs remain open for further investigation. The present study analyzes the balance between T₁ and T₂* effects obtained with IONPs, and provides criteria to understand and optimize image contrast. UTE and SubUTE detection of IONPs are studied theoretically and demonstrated experimentally. A proof of concept study for in vivo molecular imaging is performed, applying these techniques to tumor-targeted IONP detection.

Theory

In the following study, the MR contrast optimization is limited to the widely used RF-spoiled gradient echo (SPGR) sequence. Assuming the RF pulse duration is short compared to T₂*, the SPGR MR signal is described by the well known steady state equation, which includes T₁ and T₂* as well as user controlled MR parameters (flip angle θ, repetition time TR, and echo time TE):

$$S(\theta ,TR,TE)=S_{0}sin\theta \frac{1-E_1}{1-E_{1}cos\theta }{e}^{-TE/T_2\ast },\text{where}{E}_1=e^{-TR/T_1}$$ [1]

S₀ is a sensitivity factor accounting for proton spin density, voxel size, total scan time, RF coil sensitivity, and reception gain. This factor may be spatially heterogeneous. We assume that it does not depend on other specified parameters (i.e., θ, TR, TE, T₁ and T₂*). Spin density is included in this global sensitivity factor since this paper aims to consider the contrast in tissue compartments in which the spin density is assumed to be fixed.

For reasons of simplicity Eq. 1 utilizes an exponential decay to describe the complex T₂*-loss mechanism. In the general case this is not correct. The magnetic field heterogeneities which lead to T₂* effects can be classified according to their spatial scale (35). Microscopic effects take place at the molecular level and are hence irreversible because of random motion (i.e. they equally affect T₂ and T₂*). They are usually considered to decay in an exponential way.

Macroscopic effects happen at a scale much larger than the voxel size and are typically due to poor shimming or air-tissue interfaces. As shown by Yablonskiy (35) these lead to a non-exponential decay. However, if the corresponding intravoxel dephasing is low (i.e. small voxel size and low bandwidth), deviation from the exponential model is reduced.

Finally mesoscopic heterogeneities have an intermediate length scale, smaller than the voxel size, but much larger than molecular distances. The boundary between microscopic and mesoscopic scales may be formulated as a transition from the motional narrowing regime (i.e. irreversible loss) to the static dephasing regime that is assumed to be satisfied on the mesoscopic scale (35). IONP induced heterogeneities fall into both microscopic and mesoscopic scales depending on the size of the particles. In the mesoscopic contrast regime, the exponential decay model is satisfied for IONPs as shown by Yablonskiy and Haacke (36).

UTE signal is ideally modeled with TE set to zero, to describe the creation of a pure T₁-weighted image with no T₂* contamination. The subtraction of a later echo signal from the UTE signal, SubUTE(TE) = S(UTE) − S(TE), leads to a composite image that provides hybrid T₁ and T₂*contrast (Fig. 1).

The tissue relaxation time dependence on IONP concentration is described with the usual linear relaxivity approximation:

$$\frac{1}{T_1}=\frac{1}{T_{10}}+r_1[\text{IONP}]\text{and}\frac{1}{T_2\ast }=\frac{1}{T_{20}\ast }+r_2\ast [\text{IONP}]$$ [2]

where T₁₀ and T₂₀* are the unenhanced tissue relaxation times, r₁ and r₂* are the corresponding IONP relaxivities expressed in s⁻¹.mM⁻¹, and [IONP] denotes the iron concentration in mM.

Following the approach of Buxton et al. (37), we define β = (ΔT₂*/T₂*)/(ΔT₁/T₁) as an index of T₂* over T₁ relaxation effect efficiency. Such a parameter allows comparison of both contrast mechanisms in the relaxation time domain. From Eq. [2], β varies with IONP concentration:

$$\beta =\frac{r_2\ast T_{20}\ast +[\text{IONP}]r_1{r}_2\ast T_{10}{T}_{20}\ast }{r_1{T}_{10}+[\text{IONP}]r_1{r}_2\ast T_{10}{T}_{20}\ast }$$ [3]

Interestingly for low IONP concentration β → (r₂*T₂₀*)/(r₁T₁₀), whereas β → 1 for high concentration. As a result, special attention must be paid to T₁₀ and T₂₀* values when comparing the relative intensity of T₁ and T₂* contrast mechanisms, especially at low concentration. T₁₀ may be orders of magnitude longer than T₂₀*, whereas r₂* is typically much larger than r₁, although this depends on field strength, particle size (38) and contrast regime (4,19). Figure 2 displays β as a function of [IONP] for different sets of baseline conditions using IONP relaxivities measured in 1.8% agarose gel, at 3T (see “Methods”). Although r₂*/r₁ is about 30, the T₁ variation induced by IONPs dominates the T₂* variation when T₁₀ is much larger than T₂₀* (T₁₀ = 2.7 s and T₂₀* = 25 ms correspond to our in vitro experimental conditions).

Figure 2.

Relative variations of relaxation times as a function of IONP concentration. β is defined as (ΔT₂*/T₂*)/(ΔT₁/T₁). β is given by (r₂*T₂₀*)/(r₁T₁₀) in the low concentration limit and tends to 1 at high concentration. β < 1 indicates that T₁ variations are larger than T₂* variations, and vice-versa when β > 1. r₁ = 4.5 mM⁻¹.s⁻¹ and r₂* = 145 mM⁻¹.s⁻¹ were used here, consistent with in vitro measurements. Four unenhanced relaxation time sets were considered: T₁₀/T₂₀* of 0.8s/80ms as a realistic case of one order of magnitude relaxation time ratio; T₁₀/T₂₀* of 1s/67.2ms as a realistic case leading to β ≈ 2.2 at low concentration (corresponding B_opt = 1, see Fig. 3); T₁₀/T₂₀* of 1s/31ms as a realistic case leading to β ≈ 1 at low concentration; and T₁₀/T₂₀* of 2.7s/25ms corresponding to our in vitro experimental conditions.

Contrast is defined as the signal difference induced by a certain concentration of IONPs within a given tissue (i.e., C=ΔS= S_[IONP] – S_{no IONP}). For small concentrations, it can be calculated as:

$$C\approx \frac{\partial S}{\partial [\mathit{IONP}]}[\mathit{IONP}]\text{or}\text{equivalently}\text{as}C\approx \frac{\partial S}{\partial T_1}\mathrm{\Delta }{T}_1+\frac{\partial S}{\partial T_2\ast }\mathrm{\Delta }{T}_2\ast$$ [4]

where ΔT₁ and ΔT₂* are the relaxation time changes caused by the small quantity [IONP].

Contrast-to-noise ratio (CNR) per unit time (i.e. $\mathit{CNR}\propto C/\sqrt{TR}$) is an index of contrast efficiency. To provide a comparison of SPGR with SubUTE contrast efficiency, a factor of $\sqrt{2}$ noise amplification is applied to SubUTE, corresponding to the addition of uncorrelated noise arising from the two signals. The time normalization is necessary when comparing sequences with different TRs; this accounts for the greater number of signal averages that would be used if TR is made shorter, thus keeping the total acquisition time constant. An interesting consequence is that TR can be arbitrarily set to a low value without significantly affecting the optimization results, as the corresponding optimum obtained for TR→0 is essentially flat. This was first described by Ernst and Anderson (39) regarding signal optimization, briefly mentioned for the T₁ contrast by Edelstein et al. (40) and extended to a general contrast optimization, including T₂, by Buxton et al. (37). The practical result is that as long as TR is short compared to T₁, θ can be adjusted to preserve optimum signal or contrast efficiency.

Methods

Theoretical contrast optimization

We consider three different contrast regimes: 1) the usual positive T₁ contrast (CNR_T1) obtained with SPGR at the shortest possible TE (i.e., UTE); 2) the usual negative T₂* contrast (CNR_T2*), typically obtained with an SPGR sequence at long TE using a low θ to reduce concomitant T₁ effects (37); 3) the positive, combined T₁-T₂* contrast (CNR_T1–T2*) obtained with SubUTE images.

In order to compare the contrast efficiency of these three regimes, we optimize θ, TR, and TE, assuming that the IONP properties r₁ and r₂* and baseline tissue relaxation times T₁₀ and T₂₀* are known.

We further investigate the balance between T₂* and T₁ effects by assessing the ratio of optimal T₂* contrast over optimal T₁ contrast defined as B_opt. Such a parameter represents the translation of β (Eq. 3) from the relaxation time domain into the image domain, under optimal detection conditions. It is simply calculated as B_opt = |CNR_T2*/CNR_T1 | , where CNR_T2* and CNR_T1 are calculated from the SPGR signal (Eqs. 1, 2 and 4) using respective optimal MR parameters. B_opt =1 defines the boundary between dominant T₂* contrast (B_opt>1) and dominant T₁ contrast (B_opt <1). The CNR gain obtained with the SubUTE approach is computed as well. It is defined as the ratio of optimal CNR_T1–T2* over the maximum achievable CNR that can be obtained with an SPGR sequence: this can be either CNR_T2* or CNR_T1 depending on the dominant mechanism, as indicated by B_opt. Theoretical calculations and parameter optimization were performed with Mathematica (Wolfram Research Inc., Champaign, IL, USA).

Experiments

All MR data presented in this paper were acquired at room temperature on a Sigma HDx 3T scanner (GE Healthcare, Milwaukee, WI, USA).

Peptide-Coated IONPs

Nanoworm IONPs (41) (about 70×30 nm² elongated particles) were used in this study because of their superior targeting properties as compared to spherical particles. The tumor-homing peptide CR(NMe)EKA was chemically coupled on the surface of the nanoworm IONPs to produce tumor-targeting particles (42). Synthesis and coating of IONPs were performed as described previously (38, 39).

In vitro study

Nanoworm IONPs were characterized by MRI. Several concentrations were prepared (0, 0.09, 0.18, 0.9, and 1.8 mM Fe) by dilution in 1.8% agarose gel (Type I, Sigma-Aldrich Inc., St. Louis, MO, USA) and were stoked in 1 mL tuberculin syringes. Longitudinal and transverse relaxivities were calculated by measuring the T₁, T₂ and T₂* of each sample and performing a linear regression using Eq. 2. T₁, T₂, and T₂* were respectively measured with 1) an inversion-recovery fast spin echo (FSE) sequence (TR/TE=4s/9.3 ms, ETL=4, BW=±31.25 kHz, 5-mm slice thickness, 256×78 image matrix, and 0.47-mm in-plane resolution) using 14 inversion times (TI) ranging from 50 ms to 3.5 s; 2) a multiecho spin echo sequence (TR=1.5 s, BW=±31.25 kHz, 3-mm slice thickness, 256×128 image matrix, and 0.47-mm in-plane resolution) using eight evenly spaced TEs ranging from 8.5 ms to 68 ms; and 3) a multiecho gradient echo sequence (TR=300 ms, BW=±31.25 kHz, θ=25°, 1-mm slice thickness, 192×192 image matrix, and 0.63-mm in-plane resolution) with 16 evenly spaced TEs ranging from 3.8 ms to 57.8 ms.

On each image, signal intensities were measured by drawing ROIs in the center of each vial. T₁ quantification was performed by curve-fitting the analytical function A×|1-B×exp(−TI/T₁)| to the signal, with A, B, and T₁ as free parameters. Using B as a free parameter allowed us to account for the not-infinitely long TR, and for actual flip angles that differed from the nominal one (a 180° flip angle ideally corresponds to B=2 when TR is infinitely long). This significantly improved the goodness of fit.

T₂ and T₂* were extracted by fitting [(C×exp(−TE/T₂⁽*⁾))² + D²]^1/2 to the measured signal decay, with C, D, and T₂⁽*⁾ as free parameters. The offset term (i.e., D) was added to the usual exponential model in order to account for noise. It significantly improved the goodness of fit when the SNR was poor at the longer TEs (i.e. at high iron concentration). T₁ and T₂⁽*⁾ nonlinear least square fitting algorithms were implemented using Matlab (The MathWorks Inc., Natick, MA, USA).

To compare the theoretical predictions derived from the previous section, further measurements were performed using the in vitro samples. Several multiecho 2D UTE sequences were acquired with varying TEs (from 8 μs to 33 ms) and θs (10, 30, 50, and 70°) with TR fixed at 300 ms. Other MR parameters were: NEX=2, BW=±62.5 kHz, 2-mm slice thickness, 256 readout data points, and 299 radial trajectories. The data were reconstructed into a 256×256 image matrix with a 7.5-cm field of view (FOV). UTE sequences were performed in multiecho mode in order to produce SubUTE images and to assess contrast variation with TE. Additionally, this allowed extraction of baseline T₂₀* from the decay curve under the same conditions in which the UTE imaging was performed. This was important for comparing theory with results since it significantly impacts β as shown in Fig. 2.

In vivo molecular imaging study

All in vivo studies were conducted with the approval of the University’s Institutional Animal Care and Use Committee. Nude mice bearing orthotopic 22Rv-1 human prostate tumors were imaged before and 7 hours after IV injection of 5 mg iron per kg of a CR(NMe)EKA-coated nanoworm IONPs. A nontargeted form of the nanoworms was used as a control. To reduce particle uptake by the reticulo-endothelial system, each animal received an intravenous injection of nickel liposomes (0.2 mmol of Ni) 1 hr prior to the nanoworm injection (43). Mice were scanned using a custom made 3-cm-diameter bird-cage coil.

An anatomic scan consisting of a T₂-weighted FSE (TR/TE=6.4 s/70 ms, ETL=32, NEX=3, BW=±15.63 kHz, 1-mm slice thickness, 18 slices, 160×160 image matrix, 3.5-cm FOV, total scan time of about 6 min) was first acquired to show the tumor location. Next, a multiecho 2D UTE sequence was run (TR=89 ms, TEs=0.008/6.2/12.4/18.6 ms, θ=50°, NEX=6, BW=±27.78 kHz, slice thickness/spacing=2/2 mm, 2 slices, 256 readout data points, and 299 radial trajectories, total scan time of 2 min, 45 s). These data were reconstructed into 256×256 image matrix with a 3.5-cm FOV. Corresponding SubUTE images were computed. For post-injection UTE scans, axial slices were carefully matched to the previous scans by measuring the height of the sections and comparing the vascular patterns and other landmarks in the images and using the FSE scan as a guide.

Results

Optimal MR contrast, theoretical results

Using the derivative approach (i.e., small concentration limit, Eq. 4), we derived optimal MR parameters as a function of tissue and IONP characteristics. Optimization rules are summarized in Table 1. The IONP positive contrast regime (CNR_T1) is optimized under the same condition such that only T₁ varies, as there is no interference with the T₂* contrast in this case (TE→0). IONP negative contrast (CNR_T2*) and synergistic T₁-T₂* positive contrast (CNR_T1–T2*) regimes lead to specific results, different from the usual independent T₂* contrast, because of the needed balance between the two relaxation effects. For those last two regimes, it is not possible to derive analytical expressions for optimal TEs and θs using simple mathematical functions because the corresponding equations are non-algebraic. However, specific methods can be used to find numerical approximations. The negative contrast to which T₁ and T₂* effect are opposed is typically optimized with a TE ≈ T₂₀* (the optimal TE is actually slightly longer than T₂₀*) and a relatively low flip angle (θ_T2*<θ_Ernst) in order to reduce signal enhancement due to T₁ (consistent with (37)); whereas the SubUTE contrast benefits from a slightly longer TE and a higher flip angle (θ_T1–T2*), which is a compromise between the two optimal angles θ_T1 and θ_T2*.

Table 1.

Optimal MR parameters derived under the small variation assumption.

	Optimal TR	Optimal TE	Optimal θ
T1 contrast efficiencya	TR < T1	TE ≪T2*	θ = θT1 > θErnst
T2* contrast efficiencya	TR < T1	TE = T2*	θ = θErnst
IONP contrast efficiency	TR < T10	TE ≪ T20*	θ = θT1 > θErnst
Positive contrast regime (CNRT1)
IONP contrast efficiency	TR < T10b	TE = TET2*	θ = θT2* < θErnst
Negative contrast regime (CNRTT2*)
IONP contrast efficiency	TR < T10	TE = TET1–T2*	θ = θT1–T2*
Synergistic T1-T2* (CNRT1–T2*)			θT2* < θT1–T2* < θT1

TR < T₁ means that the optimum is essentially flat, as opposed to TE ≪ T₂* here.

^aT₁ and T₂* variations treated separately. See (37) for an analytical expression of θ_T1.

^bWhen β < 1, a low flip angle does not reduce T₁ effects enough at short TR, and a longer TR (> T₁) may slightly improve CNR_T2*. However in such cases T₁ contrast is the dominant mechanism and thus is a better strategy than T₂* contrast.

B_opt and the SubUTE CNR-gain were calculated using optimal MR parameters for all three contrast mechanisms (see Table 1; TR≪T₁ was used). Despite the lack of a closed form, we used numerical methods to evaluate those two quantities using a broad range of parameters covering all realistic values, and found that they vary with β but do not directly depend on T₁₀, T₂₀*, r₁, r₂*or the corresponding optimal MR parameters. Hence Figure 3 displays B_opt and the SubUTE CNR-gain as a function of β only. Such curves allow prediction of which contrast mechanism should be used for a particular application, assuming relative relaxation time variations are known. Two particular [β, B_opt] value sets are emphasized in Fig. 3-a): 1) B_opt≈0.39 for β=1; and 2) B_opt=1 for β≈2.2. The former shows that for identical relative variations of both relaxation times, optimal T₂* contrast is only about 39% of optimal T₁ contrast, while the latter shows that to obtain identical image contrast from both mechanisms, the relative T₂* variation has to be 2.2× that of T₁. This proves that T₂* contrast is intrinsically less efficient than T₁ contrast for SPGR sequences, consistent with the findings of Buxton et al. (37). In Fig. 3-b), a SubUTE CNR-gain > 1 indicates that the synergistic combination of T₁ and T₂* contrasts will increase IONP detection sensitivity. This happens when both mechanisms have comparable intensities and corresponds to 1.3< β <3.9.

Figure 3.

Theoretical comparison of optimal T₁-, T₂*-, and T₁-T₂*-contrasts as a function of relative relaxation times variations. a) B_opt is defined as CNR_T2*/CNR_T1 where both contrasts are calculated using respective optimal MR parameters. B_opt = 1 defines the boundary between dominant T₂* contrast (B_opt > 1) and dominant T₁ contrast (B_opt < 1). b) When SubUTE CNR-gain > 1, SubUTE improves detection sensitivity as compared to SPGR because T₁ and T₂* synergistic combination overcome the √2 noise penalty. In the worst-case scenario (β ≪ 1 or β ≫ 1), SubUTE will yield a CNR √2 times lower than the dominant contrast mechanism because the other mechanisms has a negligible contribution to SubUTE contrast.

Figure 4 displays the numerical calculation of CNR generated by IONPs as a function of TE and θ for four different baseline tissues, two IONP concentrations and measured IONP relaxivities. TR is arbitrarily set to 100 ms here as the shortest T₁ is about 174 ms (for 1mM iron and a 0.8s T₁₀). A discrete contrast definition was used for the calculation presented here (C= S_[IONP] – S_{no IONP} instead of Eq. 4). For each concentration and tissue baseline condition, isocontours were calculated in relation to the maximal achievable contrast among the three regimes (i.e. T_1, T₂* or synergistic T₁-T₂*). The plots corresponding to 1 μM represent the low concentration limit (i.e. identical curves are obtained using the derivative approach since the first-order linearity assumption is satisfied under such small variations), whereas the 1 mM curve set display a high concentration example. These show how optimal MR contrast and parameters vary with IONP concentration. Intermediate concentrations were also studied (not shown). These show a smooth transition to the high concentration regime. We found that the same general rules as with standard T₁ and T₂* contrast apply when concentration varies, namely i) the shorter the T₂*, the shorter the TE needed to obtain a high CNR in the T₂* regime, and ii) the shorter the T₁, the higher the flip angle required to optimize CNR in the T₁ regime.

Figure 4.

IONP contrast-to-noise ratio as a function of TE and θ for regular SPGR (including UTE) and SubUTE, calculated for different baseline relaxation times and IONP concentrations (TR = 100 ms, r₁ = 4.5 mM⁻¹.s⁻¹ and r₂* = 145 mM⁻¹.s⁻¹). Isocontours are calculated in relation to the maximal achievable contrast (indicated by the 99% contour) for each particular concentration and tissue baseline conditions. SPGR exhibits the two usual contrast regimes: positive T₁ contrast toward short TEs, and negative T₂* contrast when longer TEs and lower θs are used. The 0% isocontours indicate the settings when T₁ and T₂* effects are perfectly balanced, resulting in no effective contrast from the background tissue. Conversely, SubUTE provides a single positive contrast regime: the synergistic T₁-T₂*-contrast as described on Fig 1.

Negative T₂* contrast proved more efficient than T₁ contrast only at low IONP concentration and for a moderate T₁₀/T₂₀* ratio (a)). The corresponding B_opt correctly predicts the ratio of optimal T₁ and T₂* contrast obtained for the 1 μM case, but fails to predict the correct ratio at high concentration. This is directly related to the validity of the derivative approach that is no longer correct at high concentration: in this latter case, optimal MR parameters are different from those used to calculate B_opt. However, we observe that T₁ contrast is always superior to T₂* contrast at high concentration, and that the relative efficiency of T₂* contrast (decrease from b), d), f) to h)) follows the corresponding β values. Note that for the particular case where β =1 over the entire concentration range (e) and f)) contrast is very similar at both low and high concentration.

At high IONP concentration, the UTE approach is very important to the efficient probing of the T₁ effect since the contrast falls off rapidly beyond TE=0.1ms. As expected from Fig. 3-b), when T₁ or T₂* contrast intensities are comparable (as in a) and c)), SubUTE provides optimal contrast because of its synergistic combination of the two relaxation effects (both positive). In the other situations, it does not improve sensitivity because of the √2 noise penalty.

Experimental results

Measured T₂ and T₂* relaxation times of IONP samples were nearly equal up to about 0.9 mM, showing that the motional narrowing regime (4) is satisfied for these particles in this environment (1.8% agarose gel) and concentration range. At higher concentration, T₂* was significantly shorter than T₂. Therefore, a unique transverse relaxivity was calculated ignoring the highest concentration (1.8mM) data point so that a consistent contrast mechanism could be described. Measured T₁ and T₂⁽*⁾ relaxivities were r₁ ≈ 4.5 mM⁻¹.s⁻¹ and r₂⁽*⁾ ≈ 145 mM⁻¹.s⁻¹. A large T₂₀* difference was found between multiecho SPGR (1-mm slice and T₂₀* ≈ 63 ms) and multiecho UTE (2-mm slice and T₂₀* ≈ 25 ms), consistent with large-scale magnetic field inhomogeneities (44) in the slice direction.

Figure 5 displays a comparison between the multiecho UTE measurements and numerical computation for θ=10° and 50° (other angles are not shown). Calculations were performed using the measured relaxivities and baseline relaxation times (T₁₀=2.7 s and T₂₀*=25 ms). Comparing SPGR experimental data at 50° and TE→0 with those at 10° around TE ≈ T₂₀*, we observed that the T₁ contrast was greater than T₂* contrast for the studied IONPs and “tissue” in our experimental condition.

Figure 5.

Experimental (a) and c)) versus simulated (b) and d)) MR signal as a function of TE for two different θs (other flip angle not shown). The left column displays SPGR signal (from TE = 8 μs − i.e. UTE − to TE = 33 ms) whereas the right one shows SubUTE data. Theoretical curves were calculated using TR = 300 ms, T₁₀ = 2.7 s, T₂₀* = 25 ms, r₁ = 4.5 mM⁻¹.s⁻¹ and r₂* = 145 mM⁻¹.s⁻¹, consistent with presented in vitro data. Both experimental and simulated datasets are normalized to allow clear comparison (0 mM SPGR MR signal set to 100 at TE = 0 ms and θ = 50°).

In vivo results are displayed in Fig. 6 and 7. Figure 6 compares a pre- and post-injection dataset for CR(NMe)EKA-coated IONPs and nontargeted IONPs. CR(NMe)EKA-coated IONPs accumulate in the tumor by recognizing fibrin and fibrin-associated clotted plasma proteins in tumor vessels, while nontargeted IONPs do not (39). Such particles can cause additional clotting in tumor vessels, which creates more binding sites for the peptide (42,43). This self-amplifying tumor homing effect makes the CR(NMe)EKA-coated IONP system a good candidate for MR molecular imaging as the effect provides significant accumulation of contrast media at the targeted site. Tumors appeared hyperintense relative to muscle on T₂-weighted images before IONP injection and are consequently easily identifiable in row a). The signal was fairly homogeneous inside the tumor after intravenous injection of the nontargeted IONPs, suggesting that little or no iron was present in the tumor (the bright signal seen on the T₂-weighted FSE, nontargeted IONPs post-injection, was probably due to a necrotic core). In line with the previous report (39), signal voids were visible on both FSE (a)) and the 6.2 ms 2^nd echo UTE (c)), 7 hours after injection of the CR(NMe)EKA-coated IONPs (middle column). These areas of signal loss became hyperintense on UTE (b)) and SubUTE (d)) due to T₁ and synergistic T₁-T₂* effects of IONPs, respectively. The combination of T₁ and T₂* contrast mechanisms increased the specificity for IONPs. Note that the brightness was more evident on the subtraction images (d)) due to partial background suppression (see Fig. 7 for different echo subtraction images). This improves the specificity for short T₂* components. Those images illustrate the capacity of positive-contrast UTE strategies to detect IONPs in vivo.

Figure 6.

In vivo molecular imaging results. Axial images of tumor-bearing mice. Images were acquired before (left column) and after the injection of 5 mg iron per kg of targeted (middle column) or non-targeted (right column) IONPs to 2 different mice. Gd and IONP tubes were added as reference standards. 7h after injection of targeted IONPs, particles are visible as signal voids on T₂ and T₂* weighted images (a) and c)) and as hyperintense area on the UTE and SubUTE images (b) and d)). No particles are visible either pre injection or 7h post injection when using non-targeted IONPs.

Figure 7.

SubUTE in vivo results. Axial images of tumor bearing mouse acquired 7h after injection of 5 mg iron per kg of tumor-targeted IONPs. Different contrasts are illustrated here, obtained from the multiple echoes and corresponding subtraction image. The 2^nd echo subtraction provides a better background suppression than later echoes (note the Gd signal suppression). All those images are obtained using a single acquisition, except the T₂-weighted FSE, that is displayed here as a complementary contrast to distinguish the tumor.

Discussion

Using the standard theory of describing MR signal and IONP relaxivity as our starting point, we derived significant and easy-to-implement criteria for IONP contrast optimization. Our theoretical ideas were validated by in vitro experiments although there are some potential limitations.

The theoretical analysis presented here shows that, although the transverse relaxivity of IONPs is usually much higher than the corresponding longitudinal relaxivity, T₁ contrast strategies may be the optimal way to detect them. This is explained in part by endogenous T₁₀s that are often much longer than T₂₀*s (except for fluids or highly liquid tissues), and corresponding low β values; the other reason being that T₂* contrast is intrinsically less efficient than T₁ contrast for SPGR sequences. T₂* variations have to be 2.2 times greater than T₁ variations to produce similar CNR under optimal imaging conditions (see Fig. 3). Hence we have derived the following criteria that predict optimal contrast strategies at low concentrations: when (r₂*T₂₀*)/(r₁T₁₀) > 2.2, negative T₂* contrast is the preferred option, whereas when β < 2.2, T₁ contrast is preferred. Indeed, in the low concentration limit, B_opt is a general indicator of contrast efficiency, as presented in Fig. 3. Note the importance of the r₂*/r₁ ratio as usually reported to characterize IONPs. At high IONP concentration, β→1 independent of baseline relaxation times and relaxivities, and T₁ relaxation is expected to be the dominant contrast mechanism assuming a short enough TE can be employed.

This makes UTE detection particularly appealing in the high concentration range. We have shown that subtraction images (SubUTE) combine the usually-antagonist T₁ and T₂* contrast in a synergistic way. Although this does not always improve CNR because of a √2 noise penalty, it does increase detection sensitivity for IONPs when T₁ and T₂* effects make similar contributions (see Fig 3.). Moreover, because of its long T₂ suppression capabilities (background suppression), significant improvement in specificity is anticipated with SubUTE (as illustrated in Fig. 7).

Despite the aforementioned advantages, IONPs are usually detected by MRI using either T₂- or T₂*-weighted sequences. An explanation for this may be the lack of efficient ways to detect T₁ effects at high iron concentrations. The sensitivity toward T₁ effects is reduced when relatively long TEs are used, and image interpretation becomes ambiguous because both positive and negative contrast can occur simultaneously. UTE and SubUTE approaches represent a new step forward in addressing this limitation.

Another explanation is the reduction of IONP T₁ relaxivity and the increase in T₂* relaxivity that are usually observed in vivo when particles are compartmentalized into cells (19–23). This is a major concern, especially for molecular imaging applications (2). A better understanding of the mechanisms involved in those compartmentalization-related effects is of great importance for the purpose of quantification and for distinguishing free iron from targeted iron (20). However, this is outside the scope of this paper. The major consequence to consider in our context is that r₁ is observed to decrease with compartmentalization, whereas r₂* increases; this typically results in a reduction of T₁ contrast and an increase in T₂* contrast for in vivo applications as compare to in vitro.

However, as we noticed in our experiments, large-scale magnetic field heterogeneities (due to imperfect shimming or BMS effects) may reduce T₂₀*, and hence have a direct effect on CNR in any practical experimental setting (see Fig. 2 and 3). This is important, especially in 2D imaging, where the slice thickness is usually the largest voxel dimension and causes signal averaging. Although these effects may be corrected to improve quantification (44), they imply a reduction of T₂* contrast efficiency to the benefit of the T₁ effect.

In the in vivo experiment we described, we were not able to estimate β because we lacked the corresponding IONP relaxivity. However TR was set to an optimal value (TR of 89 ms for a tumor T₁₀ of about 1s) and a 50° flip angle was used. This appeared to be a good compromise for optimizing both UTE T₁-contrast and Sub-UTE synergistic contrast over a wide range of IONP concentrations, regardless of the actual β value, as illustrated by Fig. 4 for a slightly longer TR. We expect that application of our theory to realistic estimates of in vivo relaxivity and unenhanced relaxation times should prove to be a good indicator of in vivo contrast efficiency, as we have shown in the simple case of in vitro experiments where we measured corresponding relaxivity. Here we have successfully demonstrated that in vivo molecular imaging can be done with IONPs and T₁ contrast using UTE. In the presented case, however, the problem of limited T₁ relaxivity is unlikely to occur because CR(NMe)EKA-coated IONPs do not undergo cell internalization (42). Further studies will address compartmentalization effects using a system that causes cell internalization of the IONPs such as the iRGD peptide system (45).

Although the curves of the theoretical and experimental in vitro results do not match exactly, especially at high IONP concentrations, the agreement as shown in Fig. 5 is strong. We expect that the observed differences are due to the non-zero actual TE that was used experimentally, artifacts, or further limitations in our simple model such as nonlinear relaxivity at high concentration, non-exponential T₂* decay (as seen in Fig. 5), or break down of the validity of the usual SPGR signal (Eq. 1) when T₂* is of the same order as the RF pulse duration (46). In addition, some discrepancies could be attributed to RF-inhomogeneities as shown in Fig. 5-c for the case of 10° flip angle and UTE: contrast should be essentially null on such a proton density weighted image.

In contrast to the general optimization performed by Hendrick et al. (47), we did not consider the lengthening of signal readout time as TE increases (i.e., the reduction of the readout bandwidth, and thus a noise reduction). This explains why, in their case, optimal T₁ CNR is obtained for TE ≠ 0. Our formulation is more appropriate for UTE imaging, where the signal intensity is essentially independent of the observation time (center of k-space sampled at the initial peak of the FID, albeit with blurring and size-dependant signal attenuation (27) from T₂* decay).

Conclusions

In this study we show that contrary to conventional wisdom, optimizing T₁ contrast may be a good strategy for imaging IONPs. Hence, UTE-based sequences are of considerable interest for IONP detection. T₁ effects are measured at ultrashort TE (i.e., with minimal signal loss and susceptibility effects), providing a positive contrast even at relatively high IONP concentrations. Furthermore, a multiecho UTE sequence yields registered datasets and complementary types of contrasts in a single acquisition (T_1, T₂* and SubUTE T₁-T₂*). As compared to single T₂* contrast strategies, such a sequence enables the additional specificity afforded by measurement of the T₁ weighted UTE signal. SubUTE may increase detection sensitivity, by combining T₁ and T₂* contrast in a synergistic way, and specificity by offering background signal suppression. Using UTE and SubUTE, we show in a preliminary in vivo molecular imaging study the benefit of such additional specificity for voxels with both short T₁ and T₂*, the signature of IONPs.