Simultaneous Dual-Nuclei Imaging for Motion Corrected Detection and Quantification of ¹⁹F Imaging Agents

Abstract

Fluorine MRI offers broad potential for specific detection and quantification of molecularly targeted agents in diagnosis and therapy planning or monitoring. Because non-proton MRI applications lack morphological information, accompanying proton images are needed to elucidate the spatial tissue context. Furthermore, low concentrations typical of targeted molecular imaging agents require long examinations for signal averaging during which physiological motion may lead to blurring, underestimation in signal quantification, and erroneous localization of the agent distribution. Novel methods for truly-simultaneous acquisition of dual-nuclei MR data are presented that offer efficient and precise anatomical localization of fluorine signals using accurate motion correction based on contemporaneous proton signals. The feasibility of simultaneous dual-nuclei MRI motion correction and corresponding dual-resolution reconstruction, providing nuclei-specific spatial resolution to retrospectively optimize the balance between signal-to-noise ratio and resolution, is shown on a clinical 3T MR system.

Introduction

¹⁹F-MRI based detection and quantification of exogenous imaging agents offer several advantages compared to hydrogen-based approaches, in particular for molecular imaging applications. Virtually without a biological background signal, ¹⁹F labels enable highly specific detection of imaging agents with the added benefit that local agent concentration can be quantified [1,2] without pre-/post-contrast MRI. ¹⁹F-MRI has been explored for a number of important medical imaging applications, e.g., for absolute pO₂ measurement in oncology [3], direct detection of fluorinated drugs and corresponding metabolic pathways [4], plaque detection in neuro-degenerative disease [5], molecular imaging in cardiovascular diseases like atherosclerosis [1,6,7] or inflammation [8,9], and for cell labeling [10]. Targeted nanoparticle (NP) based imaging agents have been designed [6,11,12] which label a molecular target with large amounts of ¹⁹F nuclei (~10⁸), thereby overcoming the low sensitivity for sparse epitopes. Indeed, initial results for fluorine based in vivo detection and quantification of targeted NP appear very promising [9,13]. However, such measurements usually require long examinations for signal averaging and are thus prone to physiological motion and mis-registration with an anatomical image.

As ¹⁹F-images lack anatomical information, a proton image - typically acquired separately - is required for anatomical co-localization and to elucidate the tissue context. In most reported ¹⁹F-MRI studies, different RF-coils have been applied for ¹⁹F and ¹H imaging, or the same coil has been retuned manually. Dual-tuned coils have also been used [14], primarily for separate ¹⁹F and ¹H imaging, but are a prerequisite for truly simultaneous ¹⁹F/¹H imaging as is presented herein.

Physiological motion during the non-proton image acquisition may lead to inconsistent anatomical co-localization and to signal blurring, which compromises detection and quantification. Motion effects can be corrected using navigator techniques, but the non-proton MR signal of labeled imaging agents is inherently too weak to be detected by navigators without averaging and typically lacks sufficient anatomical distribution for robust motion sampling.

Motion detection on the proton frequency could be interleaved during the scan, but may increase scan time and/or provide insufficient temporal resolution for many types of physiological motion. As a practical alternative, simultaneous dual-nuclei MR sequences are presented herein. A simultaneous readout on two resonant frequencies allows contemporaneous measurement of anatomy and motion information on the proton channel without interrupting the sequence for averaging of the non-proton signal. To date, dual-nuclei sequences only have been used in MR-spectroscopy [15], but not for concurrent dual-nuclei MRI and motion detection.

A simultaneous readout implies a fixed spatial resolution for both nuclei. Non-proton MRI usually necessitates a moderate resolution to obtain a sufficient signal-to-noise ratio (SNR). However, accurate motion correction and anatomical co-registration require a higher spatial resolution for the proton data. To address these differing requirements, a dual-resolution reconstruction technique is introduced.

The objective of the work presented herein is to demonstrate the feasibility of simultaneous dual-nuclei MRI for motion corrected quantification of ¹⁹F imaging agents. For this, simultaneous dual-nuclei MRI capability is implemented on a clinical MRI platform that allows the simultaneous acquisition of ¹H- and ¹⁹F-images with different spatial resolutions. The feasibility of dual-nuclei MRI and motion correction is demonstrated using in vitro and in vivo examples.

Methods

Hardware

This study was performed on two 3T clinical whole-body scanners (Achieva, Philips Healthcare, The Netherlands) for the in vitro and in vivo work, respectively, with modified spectrometers and software for concurrent dual-nuclei MRI. Separate transmitter and receiver devices for the non-proton nucleus were added as schematically shown in Figure 1. Waveforms for the prescribed RF pulses were generated independently for each nucleus and simultaneously fed into the RF amplifier. Employing a broadband solid-state RF amplifier, as available in many commercial MR systems, the dual-nuclei system was implemented through the incorporation of a standard power combiner (e.g. MiniCircuits ZSC-2-1) on the low-power input side of the amplifier (Fig. 1, “C”), permitting simultaneous transmission into the dual-tuned coil. For the present study, a dual-resonant ¹⁹F/¹H small-animal solenoid coil (Ø 7 cm) was used, capable of RF transmission or signal reception at both frequencies simultaneously. To accommodate the expected difference in order of magnitude between the proton and non-proton signal levels, the MR signal was deconvolved into two channels (Fig.1, power splitter “S”) for parallel acquisition and individual gain settings. SNR is not reduced significantly by the power splitter located after the pre-amplifier of the coil because noise is dominated by thermal noise from the coil or subject.

Figure 1.

Architecture of a spectrometer system capable of simultaneous dual-nuclei MR. The RF-waveforms generated in two separate transmitters (¹⁹F and ¹H) are combined by a low-power combiner “C” before the RF power amplifier. After transmission and reception by a dual-resonant RF coil and after pre-amplification, the signal is divided into two paths with individual gain settings using the splitter “S” and recorded by two independent receivers for the two nuclei. The highlighted components (red) are the additional components that transform a standard MR system into a simultaneous dual-nuclei system.

Data acquisition and reconstruction

The basic building blocks of simultaneous dual-nuclei MR sequences are concurrent RF pulses and data acquisition windows on both resonance frequencies [16], and, in general, any pulse sequence type can be modified for dual-frequency operation. For a simultaneous gradient-echo (GRE) acquisition as chosen in this work, the non-proton excitation pulse was complemented with a proton RF pulse at identical starting time but with the freedom to use a dedicated flip angle, RF waveform and frequency. The additional pulses were added to the modeling of specific absorption rate (SAR). Sequence timing and gradients were chosen to be set for ¹⁹F, but the gradient magnetic fields jointly affect nuclei at all frequencies. Due to the difference in the gyromagnetic ratio (γ), the resulting concurrent ¹H-image had to be rescaled by γ_19F/γ_1H. This was accomplished by adding the known linear proton gradient mismatch to the 3D gradient-linearity correction of the MR system, which was modified to act differently on the images depending on the nucleus. Accordingly, the actual size of the field-of-view (FOV) for the simultaneously-acquired proton data is reduced by 6%.

An inherent limitation of simultaneous dual-nuclei sequences, due to a single set of shared gradient values, is a fixed relation between the voxel sizes in the proton and non-proton acquisition. This disadvantage was mitigated by dual-resolution reconstruction in combination with a 3D isotropic radial k-space trajectory [17]. To achieve different voxel sizes, individual spherical weighting schemes were applied in k-space for the two nuclei. For ¹H-data, full resolution reconstruction was performed using a standard parabolic weighting [18] for k-values within the Nyquist-radius (radius of sphere, where sampling density fulfills the Nyquist-criterion), and uniform weighting was applied outside [19]. For fluorine, the k-space radius was reduced by a weighting factor η (which, for η<1, narrows the radius and reduces the uniform weighting of the periphery) to reconstruct an image with lower spatial resolution but increased SNR [19,20]. The fully-sampled k-space data was acquired and reconstructed on the clinical MR system using modified software. The raw k-space data was stored for additional a posteriori reconstructions at varying resolutions which were also performed on the MR system.

¹⁹F motion correction based on simultaneous dual-nuclei MRI was demonstrated using a correction technique according to Figure 2. A 3D `stack-of-stars' radial k-space trajectory (2D-radial and 1D-Cartesian phase encoding [21]) was applied. In an offline procedure, the <sup>1</sup>H-data was divided into sub-sampled volumes at subsequent time points. After individual gridding reconstruction of the sub-volumes, motion detection was performed in the spatial domain by mutual registration using the `TurboReg' algorithm [22]. Subsequently, the measured motion vectors were used to correct the fully sampled ¹⁹F and ¹H-data by transformation in k-space. To correct translational motion, a linear phase transformation was applied to the k-space lines according to the time point of acquisition (t_i) and orientation to the motion direction. Final gridding reconstruction of the ¹⁹F- and ¹H-data yielded motion corrected images for both nuclei.

Figure 2.

Schematic overview of the implemented dual-nuclei MRI based motion correction technique. The proton data is (i) divided in to sub-sampled volumes at subsequent time points (t1, t2, …), on which motion detection is performed in the image domain (ii). The derived motion vectors are then used for correction of the fully sampled ¹⁹F-data as well as the ¹H-data (iii) using phase transformations in k-space according to the time point of acquisition (t_i). Final gridding reconstruction (iv) of the ¹⁹F- and ¹H-data yields motion corrected dual-nuclei images.

Experimental Methods

Feasibility of dual-nuclei data acquisition was demonstrated in a phantom experiment (Fig.3a/b) using a bottle containing 7 tubes (inner diameter 9mm) filled with perfluoro-15-crown-5-ether (C₁₀F₂₀O₅, single fluorine resonance) surrounded by water. Acquisition parameters: 3D dual-nuclei GRE, voxel 0.4×0.4×1.5 mm³, TR/TE=20/9.7 ms, α_19F=α_1H=20°, matrix 176²×24, slices 1.5 mm, pixel bandwidth (pBW) 72 Hz. RF power adjustment for flip angle calibration was performed on the proton channel and applied to the fluorine RF-pulse without modification. Identical RF waveform types were applied for both nuclei (asymmetric two-lobed sinc-gaussian pulse), but adjusted to obtain the same flip angle. For validation in vivo, nanoparticles incorporating perfluoro-15-crown-5-ether (PFCE-NP, 20vol% C₁₀F₂₀O₅) as described previously [11] were used in a mouse model: four C57BL/6 black mice (wild type), under ketamine/xylazine anesthesia, received 2×10 μl i.m. injections in the longissimus dorsi at the thoracic level (see Fig.3c/d, parameters equal to Fig.3a/b but 16 signal averages, 29 min acquisition). Animal experiments were approved by the institution's animal studies committee.

Figure 3.

In vitro and in vivo feasibility of dual-nuclei MRI using truly-simultaneous GRE sequences. Due to the difference in γ, an uncorrected overlay of ¹⁹F- and ¹H-data shows a mismatch in size and position (a), which can be precisely corrected (b) (reduced FOV for ¹H is indicated by the dashed line). A perfluoro-carbon injection in a mouse model is visualized as a `hot-spot' in the ¹⁹F-image (c) and is precisely localized in an overlay with the simultaneously acquired ¹H anatomy image (d).

Feasibility of dual-resolution simultaneous MRI was shown using a phantom containing diluted PFCE-NP in a vial (Ø9 mm, Fig.4). Imaging parameters were: 3D radial dual-nuclei GRE, FOV=75 mm, matrix 128³, ¹H isotropic voxel Δx=0.60 mm, α_19F=α_1H=30°, pBW=106 Hz, TR/TE=15/6.2 ms, scanning time 4min (Fig.4a–c/f). For comparison, two data sets with lowered resolution were acquired, also within 4min: (i) matrix 64³, Δx=1.17 mm, 4 signal averages (Fig.4d) and (ii) matrix 48³, Δx=1.56 mm, 7 averages (Fig.4e). SNR was evaluated on the complex image data using SNR=S/σ, where S corresponds to the mean ¹⁹F signal intensity in an ROI at the center of the vial, and noise was estimated as standard-deviation σ=√[σ_real² +σ_imaginary²] at a signal-free image location. The threshold for the edge of the vial was defined as 0.8×S to measure the apparent diameter d_app and to quantify the effect of reduced resolution via the relative change of diameter δ=(d_app/d_nom)−1, normalized to the nominal diameter of the vial d_nom. Dual-resolution reconstructions (with η=0.55 and η=0.45), low resolution acquisitions (i and ii above), and a theoretical image model (i.e., a 9 mm circle convolved with Gaussian kernels corresponding to Δx) were compared in terms of δ. At a lower resolution, the apparent diameter is expected to be reduced (δ negative) for the chosen high signal threshold, because of the blurring effect.

Figure 4.

Dual-resolution reconstruction of simultaneous ¹⁹F/¹H radial data. By reducing the radius of spherical k-space weighting on the ¹⁹F-channel, images with reduced resolution and enhanced SNR (b, c) can be obtained from a single acquisition (a, matrix 128³). The inlays show a normalized intensity profile through the ¹⁹F-containing vial. The proton spatial resolution (f) is maintained. For comparison, two acquisitions matching the spatial resolution of (b) and (c) (matrix 64³ and 48³) are shown in (d, e) for ¹⁹F and (g, h) for ¹H. SNR and blurring effects in (d, e) agree well to the corresponding low-resolution reconstructions (b, c). The corresponding low-resolution ¹H images (g, h) exhibit the expected degradation.

Simultaneous dual-nuclei motion correction was explored in vivo. Since the breathing motion of mice is shallow and very rapid, beyond detection in this case, a model with irregular translational motion (moving the tray by hand, amplitude ≈5 mm, period about 5 sec) was chosen (Fig.5, agent and mouse model identical to Fig.3c/d). This allowed a direct comparison of acquisitions with and without motion effects. Imaging parameters were as follows: 3D dual-nuclei GRE, `stack-of-stars' k-space trajectory, voxel 1.1×1.1×3.0 mm³, matrix 128²×6, TR/TE=8.8/4.0 ms, α_19F=α_1H=20°, time resolution 0.8 sec, 128 frames, 96 projections/frame, 8×sub-sampling. For quantitative analysis, voxel intensity values were compared between images acquired in the static case, those affected by motion, and the corresponding motion-corrected images. The effect of motion correction was evaluated in terms of the normalized root mean square error (NRMSE, normalized to the intensity range I_max−I_min) over all voxels (in two animals, six slices each) with signal above a chosen threshold of SNR=3 in the images. This threshold, below the level of visual conspicuity (i.e., SNR=5), was chosen as a compromise to avoid having noise effects dominate the NRMSE analysis, but still include voxels with low SNR which may have resulted from motion-induced intensity reduction. In addition to NRMSE, the intensity results were analyzed in terms of the number of voxels recovered from below noise level by the motion correction.

Figure 5.

Dual-nuclei MRI based motion correction. Insets show ¹⁹F (top row) and ¹H (bottom row) images of perfluoro-carbon nanoparticles in a mouse. The first column (a, e) are images without motion. Repeating the same acquisition, but with cranial-caudal (arrow) motion of the whole mouse, the images (b, f) exhibit significant blurring and signal loss. Using the motion information derived from the ¹H-signal, both the ¹H- and ¹⁹F-data can be corrected, rendering images (c, g) similar to the case without motion. Example ¹⁹F signal intensity profile plots (d, iiii) for the lines indicated in (a, b, c) demonstrate the signal loss and mis-registration due to motion (i versus ii). For the corrected case (iii, c), both signal intensity and position are accurately restored and match the static case (i, a).

Results

Fluorine-proton simultaneous MRI was successfully demonstrated in vitro and in vivo. Figure 3 shows simultaneously acquired dual-nuclei gradient-echo image data sets (¹⁹F-image as green overlay) before and after correction for the γ-ratio. The 6% mismatch before γ-correction is clearly visible (Fig.3a) with the outline of the vials (¹H-image) not coinciding with the fluorine content (¹⁹F-image). After γ-correction (Fig.3b), there was a good match of position and size, observed by the concentric depiction of the tube outline (¹H), fluorine content (¹⁹F) and the tube wall (no signal). The dashed line marks the actual reduced FOV of the ¹H-channel and the voxel size decreased correspondingly. Application of ¹⁹F/¹H dual-nuclei imaging in a pre-clinical example is shown in Figure 3c/d. Given the high local concentration of the imaging agent, data for both nuclei could be acquired and reconstructed at a sub-millimeter spatial resolution, and co-localization of the injection site with the fluorine `hot-spot' was observed.

Figure 4 demonstrates simultaneous dual-nuclei MRI with reconstruction at variable lowered ¹⁹F resolution (Fig.4a/b/c). Comparing the obtained spatial resolution at the chosen Nyquist factors of η=0.55 and η=0.45, a change in the apparent diameter of the vial is observed to be δ= −13±1% and −15.5±1%, respectively, corresponding well to the low resolution acquisitions with a cubic matrix of 64 (δ= −12±1%) or 48 (δ= −14±1%) (Fig.4d/e) and to theory, which predicts δ= −12.0% and −15.7%, respectively. In good accordance, SNR was increased from SNR=6±1 (full resolution) to SNR=52±3 and 92±3, respectively, for the dual-resolution reconstructions, while the low resolution acquisitions showed SNR=54±3 and 91±3. The high-resolution ¹H-image obtained in dual-resolution imaging (Fig.4f) was clearly superior to ¹H-data from the low resolution acquisitions (Fig.4g/h).

In vivo motion compensation experiments were performed successfully. Figure 5 shows an example with 3 visible “hot spots” (top row: magnified ¹⁹F-images). A green overlay in Fig.5g shows the ¹⁹F-signal upon the ¹H-based anatomy (Fig.5e/f/g). Motion correction eliminated blurring in ¹H and ¹⁹F images and recovered the shape and intensity of the fluorine signal (Fig.5c vs. 5b), which was obvious in the exemplary intensity profile plots (Fig.5d, i–iii). The evaluation of the corresponding ¹⁹F intensity values showed that the NRMSE between the data with and without motion as measured over all image voxels and animals (N=102 above threshold in all 3 image types) was reduced from 17% (b) to 6% (c) due to the dual-nuclei motion correction. A significant number of voxels with signal intensity degraded by motion below the noise threshold was recovered in the correction (27% of N=255 above threshold in images without motion). Motion tracking detected and corrected a maximum displacement of (4.7±0.3) mm, corresponding to a precision of 0.3 voxels.

Discussion

A clear benefit of simultaneous dual-nuclei MRI is the examination time efficiency. Parallel to the large number of signal averages required for ¹⁹F agent detection and quantification, a high-resolution anatomical proton image can be acquired with excellent SNR. With the γ-correction performed on the scanner, the simultaneously-acquired proton image is suitable for anatomical co-localization, geometrical sequence planning and ROI-segmentation for analysis. A consequence of simultaneous acquisition is that the FOVs for the two nuclei are unequally scaled based on the γ difference. The FOV-reduction for proton is almost negligible for nuclei with high gamma, like ¹⁹F or ³He. For nuclei with lower gamma (i.e., ²³Na or ³¹P), the dual-nuclei technique would remain applicable if a large non-proton FOV can be used or if motion information can be obtained within the reduced proton FOV.

While GRE-sequences were shown in this work, the design of simultaneous dual-nuclei sequences could easily be extended to include virtually any pulse sequence available for single nuclei, e.g., fast spin-echo with dual-nuclei refocusing pulses or steady-state free precession (SSFP) sequences. For clinical applications, all additional RF pulses on the second nucleus have to be added to the model of RF energy deposition (SAR). The GRE-sequences in this work were not SAR-limited, such that the timing could be chosen identical to single nucleus versions.

The presented dual-resolution technique not only overcomes basic limitations, but offers a unique possibility to optimize the balance between resolution and SNR a posteriori. Since the concentration of targeted agents is a priori unknown and subject to variation, the resolution cannot be easily prescribed by a fixed protocol definition; either SNR could be insufficient for detection or the resolution could be chosen conservatively low. The dual-resolution technique allows fixed data acquisition with high spatial resolution and later flexible adjustment of the non-proton resolution as required to optimize the detection and localization of the agent. Furthermore, different reconstructions of the same dataset could serve for basic spatial placement and for quantification, the latter requiring a higher SNR for sufficient dynamic range. In the current experiments it was shown that SNR-efficiency can be maintained for a low-resolution fluorine image reconstruction from a high-resolution acquisition as compared to a direct acquisition at low spatial resolution. A limitation of this experiment was the equal TR applied for the low-resolution acquisitions to ensure comparability. In practice, a shorter TR could be chosen at low-resolution and, thus, more signal averages could be recorded within the same acquisition time. But the utility of the simultaneously obtained proton data is strongly reduced for low-resolution acquisitions.

A unique benefit of dual-nuclei acquisition is the possibility to obtain same-time motion information with high SNR and without examination time penalty from the concurrent proton acquisition. The presented technique is able to quantitatively correct motion effects on signal intensity as shown by the reduction in NRMSE. Motion corrected images and images without motion show very similar quality and reproducibility in signal intensity values. The correction removes blurring effects, like the reduction of peak intensities and the spread of signal into adjacent voxels, and also recovers signal reduced below noise level. The actual fraction of recovered voxels strongly depends on the distribution and concentration of the ¹⁹F imaging agent and the level of motional blurring. The NRMSE of 6% after motion correction, as derived from voxels above noise threshold (SNR>3), provides an estimate of a confidence interval for quantification of ¹⁹F signal intensity values. The sub-sampled time frames reconstructed from radial acquisitions are robust against motion because of the low artifact level [23]. Since the time resolution for motion tracking is limited by the image quality of the sub-sampled time frames, there is a balance between time resolution and tracking precision. The model of motion employed for this demonstration was simplistic and limited by design. In clinical settings, respiratory motion would have similar frequency, but greater amplitude (though normalized to a larger field of view). Cardiac motion would also add a different component of motion. Thus, sequence design has to be adapted to allow motion tracking and correction for clinical applications, but any successful proton based self-navigation technique (see e.g. [24]) may be complemented by a simultaneous dual-nuclei acquisition to correct data of both nuclei using the proton signal. While shown for in-plane motion, the presented 3D acquisition method is also applicable to through-plane motion. In general, the 3D sub-volumes can be used to track global translations, rotations, affine transformations or even local motion vector fields, and corresponding corrections can be applied by phase transformations in k-space to accommodate a wide variety of motion. Beyond the basic `stack-of-stars' trajectory, golden-means sub-sampling can be used [25], which offers a flexible way of choosing the sub-sampling a posteriori. Hence, the reconstructed time resolution could be adapted to the observed motion pattern for optimal tracking precision.

In molecular imaging, where quantification of the targeted fluorine agent concentration is vital, simultaneous dual-nuclei imaging with dual-tuned RF coils can offer further benefits. Accurate quantification requires precise flip angle calibration and the determination of spatial sensitivity profiles, which is a challenging task if using the fluorine signal of a sparsely distributed imaging agent. The presented dual-nuclei technique allows the calibration of non-proton data via the proton signal.

Conclusion

In conclusion, a clinical MR platform was successfully extended to allow simultaneous dual-nuclei pulse-sequences and reconstruction techniques. The structural changes are minor such that clinical translation could be straightforward. Dual-nuclei MR methodology was exemplified on truly-simultaneous ¹⁹F/¹H gradient-echo sequences and demonstrated in vitro and in vivo. Dual-resolution reconstruction, using 3D radial trajectories and modified spherical weighting, was shown to offer a good SNR-efficiency and a flexible a posteriori choice of spatial resolution. The prominent benefit of concurrent dual-nuclei data acquisition is simultaneous anatomical and motion information, which was shown to serve for self-navigated motion correction for both nuclei.