Spatially resolved free‐induction decay spectroscopy using a 3D ultra‐short echo time multi‐echo imaging sequence with systematic echo shifting and compensation of B0 field drifts

1 INTRODUCTION

Signals from biologically interesting compounds in tissue (e.g., choline, creatine, N-acetyl-aspartate,1 collagen,2 or proteins leading to the so-called macromolecular signals in brain spectra3-5) are usually frequency shifted compared to free water. These frequency shifts are mainly caused by the well-known chemical shift effects, but water protons in molecules attached to proteins can experience additional induced frequency shifts. Myelin water signals, for example, have been reported to be slightly frequency shifted when compared to free water signals.6, 7

Due to their fast transverse relaxation water signals of myelin8 and 1H signals from water and collagen in tendons and bones9 are only detectable at very short echo times (TEs), as enabled by ultra-short TE (UTE) imaging or pulse-acquire spectroscopy sequences.

Spectroscopy techniques are required for proper assessment of multiple frequency components. Common chemical shift imaging (CSI) or MR scpectroscopic imagingMRSI10, 11 methods with 2D or 3D Cartesian spatial encoding of fre-induction decay (FID) signals apply phase-encoding strategies between RF excitation of the magnetization and the start of signal acquisition, leading to an inevitable delay of at least about 1 ms.

In contrast, UTE imaging sequences are capable of acquiring signals immediately after excitation (TEmin ≈ 0.05 ms), since spatial encoding is mainly performed after recording data for the center of k-space. Therefore, UTE imaging is more sensitive to rapidly decaying signal components. However, the frequency spectrum of contributing signals cannot be decompensated. In previous applications of UTE sequences for characterization of signal components with short T2 times, only the signal amplitude values were determined and analyzed as a function of the TE.12, 13 While this method is very well suited to estimate the relaxation behavior of a single frequency component (i.e., when no other frequency-shifted signals need to be considered)14 it does not allow a distinct assessment in the case of multiple frequency components contributing to the UTE signal. As mentioned above, a distinct analysis of the signal components requires phase-sensitive measurements at different delay times as it is done by spectroscopy sequences.

The present study aims to record a complex time signal with a UTE imaging sequence, thus providing spatial encoding and spectral information simultaneously. Therefore, this is essentially a spatial-spectral encoded MRSI sequence based on 6 to 15 temporal interleaves acquired with a stack-of-spirals 3D UTE sequence. The comparably high number of interleaves is necessary to ensure the required spectral range. However, since the underlying UTE imaging sequence in contrast to common MRSI sequences15 enables the detection of ultrashort TEs this approach is more specifically referred to as UTE-FID. The FID time signal of the UTE sequence is intended to correspond largely to a non-selective pulse-acquire spectroscopy, but with an advantage that this is recorded separately for each imaging voxel. To limit the measurement time, a multi-echo 3D spiral UTE was used to record six equidistant echoes after each excitation. The time points in-between the six TEs were recorded by repeated application of the UTE sequence with suitably shifted TEs. Since active gradient switching in UTE sequences results in a significant temporal and slightly location dependent B0 field drift over time,16 appropriate phase correction procedures had to be implemented.

The method was evaluated in phantoms with solutions of collagen powder, which has several different chemical shift components with different relaxation properties. Collagen was chosen since its behavior is comparable to that of other macromolecules, and quantification studies based on UTE measurements already exist. Siu et al17 presented a bi-exponential approach using a single collagen frequency to fit the magnitude of the signal obtained by UTE imaging. This approach was reported to allow the detection of collagen from a concentration of approximately 10 g per 100 mL.

2 METHODS

All measurements were performed on a 3T whole body imager MAGNETOM PrismaFit (Siemens Healthcare, Erlangen, Germany). The UTE imaging measurements on phantoms with multiple samples were conducted using a 20-channel head-neck coil, while standard non-localized FID-spectroscopy and single-voxel localized STEAM-spectroscopy according to Ref. [18] of a sphere filled with 50% collagen solution were executed with a wrist coil. Furthermore, in vivo UTE imaging measurements on a knee were acquired with a knee coil. All coils were selected as they best matched the respective phantom size and geometry, therefore ensuring the highest SNR.

2.1 UTE measurements

A prototypical 3D UTE prototype sequence using a non-selective rectangular RF excitation pulse (with duration of 60 µs) was applied for data acquisition. Slice encoding was achieved by phase-encoding. k-Space is sampled by stack-of-spiral trajectories.19, 20 The UTE sequence allowed recording of raw data for six TEs after each RF excitation. To obtain a complete time signal for each voxel with equidistant temporal spacing, the UTE sequence was applied repeatedly with shifted TEs (shifts ΔTE of 0.25 ms or 0.5 ms were used; dwell time is corresponding to ΔTE) analogous to the concept of temporal interleaving in spectral-spatial encoding. A scheme of the sequence is depicted in Figure 1.

image

Timing scheme of the UTE-FID sequence. Six equidistant TEs (TE1, TE2, …, TE6) are recorded after each RF excitation. k-Space for each TE is sampled by a stack of spirals. To achieve a temporal spacing of sampling points suitable for spectroscopy, further acquisitions are required using shifted TEs. The TE spacing ΔTE corresponds to the desired dwell time for the spectrum

The 3D UTE sequence recorded images of 64 slices with a 2 mm thickness for six different TEs within one acquisition. For a first data set, this sequence was repeated 15 times with a repetition time (TR) of 49 ms to obtain 90 equidistant TEs between TE1 = 0.05 ms and TE90 = 44.55 ms (resulting in a spectral dwell time of ΔTE = 0.5 ms, which translates into 2 kHz spectral bandwidth and, thus, covers the entire collagen spectrum, which extends over approx. 1200 Hz at a field strength of 3T). All the TEs of the UTE imaging sequence are thus corresponding to delay times TD (with TD being the time delay between RF excitation and time of recording central data of k-space) in the time domain for the FID spectrum. The measurement time for recording this data set was approximately 67 minutes, which limits clinical applicability. Therefore, a second data set with 72 TEs between TE1 = 0.05 ms and TE72 = 17.8 ms (ΔTE = 0.25 ms) was recorded within 24 minutes using a reduced TR of 22 ms. This strategy led to oversampling regarding the chemical shift range of collagen signals at 3T, but would be adequate for higher field strengths as 7T. Subsequently, for further analysis of the collagen solutions at 3T every second TE of this data set was omitted (resulting in ΔTE = 0.5 ms) to evaluate the feasibility of a further reduction of measurement time to 12 minutes.

2.2 Measurements of static magnetic field drift and phase correction

Frequent gradient switching of the UTE sequence led to heating of the system (especially the shim irons) and consequently to significant drifts of the static magnetic field B0 over time as reported by Hui et al for 95 different 3T MR scanners.21 Therefore, undesired influences on the frequency and phase of the signals recorded several minutes after the start of the measurement occur. For this reason, B0 drift-related phase shifts were mapped, and a respective phase correction was implemented to compensate for these effects. Before evaluating the complex data sets, a temporally and spatially dependent phase correction was performed.

The phase shifts resulting from B0 drifts must be corrected when the signals (separately for each pixel) are used as complex data in the time domain for the Fourier transform, which finally leads to complex spectra in the frequency domain. For an adequate temporal phase correction, a reference region in or near the investigated sample is needed in which only a single frequency contributes to the signal (e.g., pure water). The average phase within this region was fitted for each UTE acquisition as a function of TE by a linear model. Ordinate intercept and gradient of the measured reference phase over time are used to determine correction factors (initially uniform for the entire spatial region).

For extended samples a location-dependent phase correction can be applied. In the phantom measurements, a mask of areas containing pure water was created, and B0 field maps were derived from the phase images of the successive UTE recordings. A 2D third-degree polynomial function was used to compensate for spatial differences in the B0 field drift in the images. Location dependent zero- and first-order phase factors are added to correct for the spatially dependent B0 field drift for each individual recording.

2.3 Calculation of localized UTE-FID spectra

Localized 1H FID spectra for selected regions of interest (ROIs) in the phantom (with uniform content) were calculated from the UTE data. For this purpose, the complex UTE image data from the respective region were first corrected for time- and location-dependent field drift effects. In this process, the complex signal values for each point in the time domain were averaged over all pixels in the ROI. This formed the basis for extrapolation of the measured FID to longer delay times TD (which were not recorded) by a 60th-order auto-regressive model based on Burg’s method22 for the data set with 72 and 90 data points in the time domain, and 30th-order for 36 points in the time domain, respectively. Subsequently, a Gaussian filter (σ = 10 ms) was applied to the FID, and then a Fourier transformation was performed. The effect of the extrapolation and filtering on the spectrum and FID respectively are depicted in Supporting Information Figure S1, which is available online.

2.4 Standard FID spectroscopy

For comparison, a non-localized standard FID (pulse-acquire) spectroscopy sequence from the manufacturer was applied, providing a minimum TE of 0.15 ms for the first data point in the time domain. This sequence recorded 2048 time points with a dwell time of 0.5 ms. Spectra with and without water suppression were recorded gathering 32 acquisitions with a TR of 1500 ms.

2.5 Volume-localized STEAM spectroscopy

Results were further compared to those of spatially resolved single-voxel spectroscopy: A volume-selective STEAM sequence according to Ref. [18] with TE = 5.4 ms, transverse magnetization (TM) = 75 ms, and TR = 1500 ms was applied, recording 1024 time points in the time domain with a dwell time of 0.5 ms. For each spectrum, 16 acquisitions were acquired from a volume of 10 × 10 × 10 mm3.

2.6 Collagen phantoms

Different collagen concentrations in aqueous solution were obtained by dissolving 6 g, 12 g, 24 g, 36 g, 48 g and 60 g of NEOCELL (Nutranext, Sunrise, Florida, United States) collagen type-1 and -3 powder in 60 g distilled water. Six blue caps containing these different solutions ranging from 9.1% to 50.0% were placed in a cylinder (diameter 13 cm, height 16 cm) filled with distilled water. This phantom with six known collagen concentrations was used for UTE measurements with 90 and 72 TEs. An ROI of 10 × 9 pixels was placed in the pure distilled water between blue caps for temporal phase correction (Figure 3B), while for location-dependent phase correction the entire area of distilled water was used (see Figure 7A,E). Circular ROIs with a radius of 10 pixels were evaluated for each solution to calculate localized FID spectra for all collagen concentrations.

A quantitative determination of collagen concentration was also attempted using the UTE-FID approach with the phantom mentioned above. In addition, to evaluate the sensitivity to low collagen concentrations, the blue caps originally filled with 16.6%, 28.6%, 37.5%, and 44.4% collagen solution (in mass percent) were replaced with blue caps filled with 2%, 4%, 6%, and 8% collagen solutions. Only a UTE data set consisting of 72 TEs was recorded for this purpose.

For non-localized FID spectroscopy and volume-localized STEAM spectroscopy, a thin-walled glass sphere with 20 mm radius was filled with 50% collagen solution (mass percent) again using the NEOCELL collagen type-1 and -3 powder.

2.7 Estimation of proton-density-collagen-fraction

Diefenbach et al23 presented a generalized method to separate different chemical species based on multi-echo gradient-echo-based complex imaging data. Their open-access Python computer programs were used to separate collagen and water signal based on the corrected complex UTE datasets. A bi-exponential model was chosen allowing different relaxation times for collagen and water (somewhat simplified as the collagen signal components show different relaxation behavior). The pixel-by-pixel estimation of the collagen fraction is based on a signal model derived from a non-localized FID spectrum with 32 Hz water saturation of a 50% solution of the collagen powder in distilled water. MATLAB® R2020a was used to fit 11 peaks (7 between 0.3 and 2.3 ppm and 4 between 3.2 and 4.2 ppm) with frequencies below (up-field) water and 8 peaks (between 6 and 9 ppm) above the water frequency (down-field). Fitting was performed with a multi-Gaussian model provided by MATLAB® based on non-linear least squares allowing for a maximum of eight components. The measured collagen spectrum, the extracted collagen model and the residuum are depicted in Figure 2. The chemical shift of each signal contribution (0.644 ppm, 0.942 ppm, 1.122 ppm, 1.286 ppm, 1.398 ppm, 1.709 ppm, 2.024 ppm, 3.307 ppm, 3.633 ppm, 3.872 ppm, 4.141 ppm, 6.592 ppm, 6.730 ppm, 7.099 ppm, 7.313 ppm, 7.992 ppm, 8.261 ppm, 8.492 ppm, 8.723 ppm), and its normalized magnitude fraction from the integral of each Gaussian curve (0.144, 0.028, 0.081, 0.009, 0.066, 0.116, 0.039, 0.007, 0.150, 0.005, 0.153, 0.013, 0.011, 0.016, 0.007, 0.029, 0.056, 0.021, 0.081) were extracted and taken as a-priori knowledge.

image

Spectrum of 50% collagen solution with 32 Hz water saturation (blue), fitted 19-peak collagen model (red), and residuum (yellow)

The fitting performance of the chosen model was improved by initial B0 field mapping based on the phase images of the first acquisition. Analogous to the proton-density-fat-fraction according to Reeder et al24 a proton-density-collagen-fraction (PDCF) was calculated based on the resulting “collagen images” and “water images.” The fitting program provides a map showing the Euclidean norm of the residual. This norm was used to exclude voxels with an unsatisfactory fitting result (Euclidean norm of residual > 1000).

2.8 In vivo UTE imaging measurements of the knee of a healthy volunteer

UTE images of the knee of a 32-y-old, healthy, female volunteer were acquired after giving informed consent for 36 different TEs with analogous measurement parameters to the phantom data set with corresponding number of TEs. Temporal and spatial correction were conducted in the adipose tissue. Therefore, the phase shift of fat compared to water was calculated for each TE based on chemical shifts (1.3 ppm, 0.9 ppm, 2.05 ppm, 2.75 ppm, 2.3 ppm, 5.28 ppm, 4.18 ppm) and relative contributions (0.51, 0.24, 0.09, 0.02, 0.025, 0.035, 0.08) of a seven-peak fat model. After subtraction of the calculated phase shift, the algorithms of temporal and spatial correction used for the water areas in the phantom study were applied. A localized UTE-FID spectrum is calculated for a 2-by-2-pixel ROI within the subcutaneous adipose tissue.

3 RESULTS 3.1 Static magnetic field drift and phase correction

Precise spatial-temporal mapping of B0 field drift was performed by repetitive measurements with a UTE sequence with unchanged TEs. Thus, temporal B0 drifts result in easily accessible phase shifts of recorded signals. Figure 3A depicts the progression of phase shift in a ROI containing pure water for 12 consecutive measurements (each of them with six unchanged TEs: 0.05 ms, 3.05 ms, 6.05 ms, 9.05 ms, 12.05 ms, 15.05 ms). For the series of measurements, the same sequence with a duration of 2 minutes was repeated 12 times without intermission. For example, the signals recorded at the longest TE (15.05 ms) exhibit a phase drift of more than 0.6 rad corresponding to a frequency drift of ~6 Hz after 24 minutes of continuous UTE imaging.

image

(A) Temporal progression of the drift of averaged phase values within a water ROI for all six echoes in a series of 12 successive UTE measurements with a duration of 2 minutes without intermission. The gradient of the phase values increases from acquisition to acquisition with a faster increase in the beginning. (B) Phase image recorded at TE = 17.8 ms (6th echo of the 12th acquisition of the recording with TE offset). The development of the mean phase value inside the white rectangle serves the subsequent temporal phase shift correction. Resulting phase image with temporal phase shift correction only (C) and with temporal and spatial correction (D). All images are corrected regarding field map bias. A B0 map was calculated from the images of the first acquisition and used to remove influence of the initial field inhomogeneities

Figure 3B–D show phase shift images (unwrapped phase images corrected by the initial field inhomogeneities of the first acquisition) at TE = 17.8 ms (6th echo of the 12th acquisition of the recording with shifted TEs). Figure 3B shows the remaining B0 drift-related phase shift in a slice without correction, whereas Figure 3C depicts the same phase shift image after temporal drift correction of all pixels with the same value obtained from the water area marked in Figure 3B. For some areas remote to the reference area, phase shifts up to 0.3 rad (~3 Hz remaining spatially dependent field drift) still occur for the sixth echo of acquisition number 2. Figure 3D shows the situation after temporal and spatial phase correction. It is obvious that further spatial correction clearly reduces remaining local phase shifts. Without correction, the overall phase drift over time was found approximately two times higher than maximum spatial differences in our phantom with a diameter of 13 cm.

3.2 Localized UTE-FID spectroscopy

All areas of the examined phantom with water and different collagen concentrations could be analyzed. Both the correction of the B0 field drift and the analysis of the FID time signals worked sufficiently well. Localized FID spectra were calculated from the 90, 72, and 36 TE data sets. The spectra obtained for the different collagen concentrations are shown in Figure 4 for all three data sets. A high correlation between signal intensity and collagen concentration was found for all contributions to the signal pattern.

image Localized FID spectra taken from the ROIs depicted in Figure 6A based on different numbers and timings of echoes for six different collagen concentrations (values given in mass percent)

Figure 5 shows localized UTE-FID spectra without correction for field drift effects, with a uniform correction over time and with correction for temporal and spatial effects. Localized spectra were calculated from the UTE data set consisting of 72 TEs for six different collagen concentrations. It is obvious that the temporal correction is particularly important as it brings a clear improvement in spectrum quality. Since the field drift in the 3T MRI system used is not very spatially dependent, an additional spatial correction provided only minor advantages for areas situated close to the reference water area.

image

Localized FID spectra obtained from a UTE-FID data set with 72 TEs, ΔTE = 0.25 ms and TR = 22 ms without correction for field drift effects (top), with a uniform correction over time (middle), and with correction for temporal and spatial effects (bottom). Six different collagen concentrations (values given in mass percent) are considered

Figure 6 shows the comparison of a STEAM spectrum recorded with TE = 5.4 ms with a non-localized pulse-acquire FID spectrum and a UTE FID spectrum (from the UTE data set with 72 TEs) of 50% collagen solution. All three spectra were recorded without water suppression.

image

Spectra of 50% aqueous collagen solution: STEAM spectrum with TE = 5.4 ms and TM = 75 ms (top), a non-selective FID spectrum (middle), and a UTE-FID spectrum calculated from the set of UTEimages with 72 different TEs (bottom)

The signal pattern in the localized STEAM spectrum shows clear deviations from the localized UTE-FID spectrum. In contrast, spectra from the non-selectively recorded FID (from a spherical sample) and the UTE-FID spectra (from the phantom with multiple samples) show more similarity, but the spectral lines are wider in the UTE-FID spectrum due to the Gaussian filtering in the time domain. According to Wishart et al25 the resonances between 0.5 and 2.5 ppm in the FID spectra are attributable to the aliphatic protons of valine, hydroxyproline, proline, and lysin, which are the most common amino acids of collagen. The resonances evident downfield of water in all three spectra are similar to those found by Winkler et al26 to disappear when the solvent was changed to D2O. Their origin is assumed to be in exchangeable protons of amine, amide, and OH groups.

3.3 Estimation of PDCF

The arrangement of the collagen solutions in the phantom and their concentrations in mass percent are incorporated in Figure 7A. PDCF maps were calculated for the UTE data sets with 90, 72, and 36 TEs; they are shown in Figure 7B–D, respectively. For all samples of collagen solution, there is a clear deviation of the determined collagen concentration from surrounding pure water.

image (A) Magnitude image recorded at TE = 0.05 ms indicating the collagen concentration in mass percent for each area of evaluation. The area of distilled water (0%) enabled temporal (ROI depicted in Figure 2A) and spatial drift correction (entire distilled water area). (B) PDCF map obtained using the UTE sequence with 90 TEs, ΔTE = 0.5 ms and TR = 49 ms. (C) PDCF map obtained from 72 TEs with ΔTE = 0.25 ms and TR = 22 ms. (D) PDCF map obtained from 36 TEs with ΔTE = 0.5 ms and TR = 22 ms. (E) Magnitude image at TE = 0.05 ms indicating the lower collagen powder concentrations in mass percent for each area of evaluation in the modified phantom. Again, the area of distilled water (0%) was used for B0 drift correction. (F) PDCF map of the modified phantom (according to (D)) obtained using the UTE sequence with 72 TEs, ΔTE = 0.5 ms, and TR = 49 ms

To better identify the detection limit for collagen, further measurements were performed in which the four mean concentrations of the phantom were replaced by lower concentrations in the range of 2% to 8% (arrangement according to Figure 7E).

The UTE-FID sequence with 72 TEs and ΔTE = 0.25 ms was used. The resulting PDCF map is shown in Figure 7F. From a concentration of 4%, the collagen solution is distinguishable from the surrounding distilled water region.

The mean and SD of the PDCF were calculated for each sample in a circular ROI with a radius of 10 pixels. In addition, the entire area of distilled water was used to define a PDCF reference value at 0% collagen concentration. The results are shown in Figure 8A for the three data sets with 90, 72, and 36 different TEs, respectively.

image

(A) Mean values and SD of the PDCF maps calculated from complex image data at 90 TEs, 72 TEs, and 36 Tes plotted against collagen concentration in mass percent. (B) Mean values and SD of the PDCF maps calculated from complex image data of 72 TEs for all measurements including those with low collagen concentrations

A fit of the measured values to a linear model with a weighting of the individual values with the normalized reciprocal values of the SD was performed separately for all three data sets with high collagen concentrations. The fit parameters are summarized in Table 1.

TABLE 1. Fit parameters of the linear correlation between estimated PDFC and collagen concentration in mass percent No. of TEs Ordinate section Gradient Adjusted R-square 90 0.529 0.501 0.995 72 2.220 0.444 0.962 36 2.473 0.447 0.965

While the acquisition with 90 TES and ΔTE = 0.5 ms yielded the highest adjusted R-square, no reduction of R-square was found when reducing the number of TEs from 72 to 36 and simultaneously increasing ΔTE = 0.25 ms to ΔTE = 0.5 ms (according to the sampling theorem). Therefore, a dwell time of 0.5 ms is sufficient for the PDCF estimation at 3T but should be shortened accordingly at higher field strengths.

The mean values and SD of the PDCF obtained from the UTE measurements with 72 TEs for all collagen concentrations are shown in Figure 8B (with PDCF values of the samples with high collagen concentrations shown in red, and PDCF values of the samples with low collagen concentrations in blue). The mean values and SDs show significant differences from distilled water already at a concentration of 2%.

3.4 In vivo UTE imaging measurements of the knee of a healthy volunteer

Figure 9A shows a UTE image of the knee of a 32-y-old volunteer for the first TE (0.05 ms). The blue square indicates the 2-by-2-pixel ROI selected for calculation of a localized UTE-FID spectrum. The resulting spectrum is plotted in Figure 9B and is comparable to well-known fat spectra. A dominant CH2 resonance is evident at approximately 1.3 ppm. Furthermore, the fat resonances of CH3 (~0.9 ppm), glycerin (~4.1 ppm), and vinyl (~5.4 ppm) are visible. The allylic and α-methylene resonance merge to one line at 2.1 ppm. Additionally, spectral lines at 4.7 ppm and 2.9 ppm are attributed to water and the diallylic fat resonance, respectively.

image

(A) UTE image of a healthy knee at TE = 0.05 ms. The blue square indicates the 2-by-2-pixel ROI selected for calculation of a localized UTE-FID spectrum. (B) Localized UTE-FID spectrum of subcutaneous adipose tissue. The spectral lines are assigned to (CH3 at 0.9 ppm, CH2 at 1.3 ppm, allylic fat resonance and α-methylene at 2.1 ppm, diallylic fat resonance at 2.9 ppm, glycerin at 4.1 ppm, water at 4.7 ppm, and vinyl at 5.4 ppm

4 DISCUSSION AND CONCLUSIONS

The proposed UTE-FID spectroscopy represents a new approach to localized spectroscopy that is particularly sensitive to rapidly relaxing signals. It was possible to reproduce the signal patterns of dissolved collagen powder known from the non-selective pulse-acquire FID spectrum with good quality, demonstrating an approximately linear dependence of the signal intensities on concentration.

In addition to examining collagen structures such as tendons and ligaments, a possible clinical application is in the detection of organ fibrosis, which according to Zeisberg and Kalluri27 is responsible for one third of natural deaths worldwide. Excessive accumulation of extracellular matrix is usually caused by prolonged organ damage and can occur in liver, kidney, heart, and lung,28 but also in skeletal muscle.29 In a healthy liver, there is about 2% collagen content, whereas in an alcoholic liver in the final stage, an increase to sometimes more than 8% can be detected.30 To detect the accumulation of collagen I/III-rich tissue, liver biopsy, despite its local limitation, still serves as the gold standard today.31 However, as this method is associated with various risks, a non-invasive detection would be preferable, especially for follow-up and early detection.32 In addition to ultrasound (fibroscan), MRI offers various possibilities: Current MR methods use contrast agents or measure liver fibrosis indirectly via tissue elasticity, relaxation, diffusion, or perfusion.33 The study presented here for direct detection of collagen signals compares well with previous measurements by Siu et al,12 who, however, only used signal amplitude in UTE sequences and not phase information (as in our approach). The method presented here appears to be advantageous as it allows the discrimination of collagen solution and distilled water in the PDCF map from a concentration of about 4% (compared to previous work with a minimum concentration of about 10%).

Comparing the PDCF maps and FID spectra obtained with different numbers and timings of TEs, the best results could be achieved with 90 TEs. However, UTE-FID records from recording 36 TEs within 12 minutes recording time are comparable. Shorter recording times advantageously also lead to lower phase drift problems. In general, at least on the 3 T whole-body system used, it has been shown that the temporal B0 field drifts are quite pronounced but can be successfully corrected. The spatial dependence of the B0 field drift is rather low, and its correction can certainly be omitted in many possible in vivo applications, especially when the evaluation is restricted on small areas. In the collagen phantom with a diameter of 13 cm, the temporal field drift was two times the maximum spatial difference in an axial view. El-Sharkawy et al16 reported a temporal field drift of 0.7 ppm in 4 h, while over a comparable distance in axial view a maximum difference in spatial field drift of about 0.4 ppm was observed. The relative amounts of temporal and spatial B0 field drift are expected to depend on the number and positions of shim irons used to correct for slight magnet manufacturing inaccuracies and are unique to each MR device.

For example, the 2D MRSI spectroscopy at 3T presented by Chang et al34 acquires a matrix of 32 × 32 voxels with a size of 6.25 × 6.25 × 10 mm3 within 14 minutes. This method has a comparable measurement time to the presented UTE-FID spectroscopy, but a much lower spatial resolution. However, one disadvantage of the UTE-FID method is the significantly higher stress on the gradient system and the associated heating of the iron pieces used for passive shimming, resulting in B0 field drifts. The spatial encoding of fast relaxing signals by UTE sequences is also not optimal, since a large part of the spatial information is only read out with a longer delay after the RF excitation. Furthermore, with the UTE-FID approach, only a relatively short time range can be read out directly after the RF excitation when short TRs (and short examination times) are desired. To obtain narrow lines in the spectrum, the time signals must be extrapolated for longer TD. However, this is usually less problematic, especially since rapidly relaxing signal components for longer TD disappear in the noise. Additionally, motion of body parts might cause interleaving artifacts especially for measurements with a high number of interleaves, which however was not the case for the in vivo knee study.

For many tissues particularly outside the brain, fat signals are to be expected, which often overlay the desired fast relaxing signals of protein (e.g., collagen). For a meaningful evaluation, the signal model underlying the PDCF determination must be supplemented by the signals of the fat spectrum. However, it can be assumed that by using an approach with multiple signal components, an improvement can be achieved compared to simpler models with often one single frequency for rapidly decaying signals.

The UTE-FID spectrum extracted from a 2-by-2-pixel ROI within subcutaneous tissue exhibits resonances attributable to different fat resonances and thereby demonstrates the feasibility of in vivo measurements with the presented 1H localized FID spectroscopy.

ACKNOWLEDGMENTS

The MRS package containing the applied STEAM sequence was developed by Edward J. Auerbach and Małgorzata Marjańska and provided by the University of Minnesota under a C2P agreement. Open access funding enabled and organized by ProjektDEAL.

CONFLICT OF INTEREST

Thomas Benkert is an employee of Siemens Healthineers. All other authors declare no conflict of interests.

留言 (0)

沒有登入
gif