Anisotropy of T2 and T1ρ relaxation time in articular cartilage at 3 T

The anisotropy of R2 and R1ρ relaxation rates in articular cartilage contains information about the collagenous structure of the tissue. Here we determine and study the anisotropic and isotropic components of T2 and T1ρ relaxation parameters in articular cartilage with a clinical 3T MRI device. Furthermore, a visual representation of the topographical variation in anisotropy is given via anisotropy mapping.


INTRODUCTION
Osteoarthritis (OA) is a complex degenerative joint disease that is a common cause of disability among the aging population.One key symptom of OA is the progressive degradation of internal structure of articular cartilage (AC) as the disease develops. 13][4] AC consists primarily of water, collagen (Type II), proteoglycans, and chondrocytes.The internal structure of AC can be divided into distinct histological zones based on the depth-wise variation on the average orientation of collagen fibers within the tissue 2,5 (Figure 1A).
Transverse magnetic relaxation (T 2 ) has been found to depend on the orientation of AC in relation to the B 0 field. 2,6,78][9] Longitudinal relaxation in a rotating frame of reference (T 1ρ ), while initially perceived as a marker of proteoglycan content, 6,12 has also been found to display dependence on collagen orientation. 6,12Furthermore, the anisotropy of T 1ρ has been found to depend both on the type of spin-lock preparation and on the frequency (amplitude) of the spin-lock pulse. 6,8,10,11,13Relaxation anisotropy has been detected in various tissues, including cartilage, tendons, and white matter. 11,14,15][17][18][19][20] As OA develops, the internal structure of the collagen network within AC degrades, reducing the structural organization of the tissue. 1 The anisotropic relaxation components of both T 1ρ and T 2 , previously reported to be sensitive to cartilage degeneration, [21][22][23] also carry information on the degree of organization of the collagen network within cartilage.This relationship has the potential to allow noninvasive assessment of the condition of cartilage through quantitative MRI (qMRI) at clinical field strengths.Furthermore, accurately modeling the relaxation anisotropy has the potential to help mitigate its effects in clinical imaging, where it is often regarded as an artifact.In many of the previous studies, the cartilage samples have consisted of small osteochondral plugs, measured only at a small subset of sample orientations in relation to the main magnetic field, 13,[24][25][26] or using a single relaxation parameter. 27n this work, the relaxation anisotropy and depth dependence of T 2 , continuous-wave T 1ρ (CW-T 1ρ ), and adiabatic T 1ρ in AC were investigated by separating the total relaxation into its isotropic and anisotropic components.These relaxation components were extracted from measurements using an extensive set of sample orientations ranging from 0 to 200 • with respect to B 0 .Unlike in most of the previous studies using small samples and/or high field strengths, the presented measurements were performed on whole ex vivo bovine stifle joints using a clinical 3T MRI scanner.This work elucidates the significant role of the anisotropic relaxation component for clinical cartilage imaging.Furthermore, we aim to clarify the confusion in interpreting T 1ρ measurements in articular cartilage being dependent on either proteoglycan content or collagen orientation.

Measurements
Eight ex vivo bovine (Bos taurus) stifle joint samples containing intact cartilage surfaces of healthy adult animals were acquired from a local abattoir.The samples were stored at −20 was prepared for imaging by introducing a small amount of Dulbecco's phosphate buffered saline (Sigma-Aldrich) into the synovial space of each joint capsule to prevent the cartilage surfaces from drying during scanning.Afterward, the samples were vacuum-sealed using a commercial vacuum sealer.
The measurements were performed using a clinical 3T whole-body MRI scanner (MAGNETOM Skyra; Siemens Healthcare, Erlangen, Germany) using both an 18-channel receiver body coil and a 32-channel receiver coil within the patient bed, resulting in voxel dimensions of 0.52 × 0.52 × 3.00 mm 3 , interpolated to 0.26 × 0.26 × 3.00 mm 3 .The study used four 2D imaging sequences, summarized in Table 1.T 2 -weighted scanning was performed using a multi-echo spin-echo sequence with echo train length of 12 and echo spacing of 10.7 ms.CW-T 1ρ imaging used a continuous-wave refocused rotary echo T 1ρ magnetization preparation, 28 with magnetization flipped along +Z axis at the end of the magnetization preparation block and before the readout. 29,30CW-T 1ρ scanning was repeated twice with two different spin-lock frequencies of 300 and 500 Hz, using five spin-lock durations of 0, 20, 40, 60, and 80 ms.Adiabatic T 1ρ -weighted (Ad-T 1ρ ) imaging was performed using hyperbolic secant pulses (HS4, with maximum spin-lock frequency of 600 Hz) 31,32 and five spin-lock durations of 0, 24, 48, 76, and 96 ms.The Ad-T 1ρ imaging obtained three averages per scan to maintain a comparable SNR to CW-T 1ρ and T 2 measurements.All T 1ρ acquisitions used a spoiled turbo gradient-echo readout with TE of 3.43 ms and flip angle of 15 • .This flip angle has previously been shown to provide good SNR with minimal saturation in Ad-T 1ρ imaging. 32,33s both Ad-T 1ρ and CW-T 1ρ use the same readout, the same flip angle was also used in both sequences for consistency.Each measurement sequence used a TR of 3.5 s.
Sample reorientation between measurements was performed manually using a 3D-printed sample rotation platform (Figure 2A).The rotation platform itself was fixed onto the patient bed of the MRI device during measurements (Figure 2B).Each sample was initially positioned to simulate a patient laying on the sagittal plane on the patient bed, with the tibia aligned in parallel with B 0 ( = 0 • ).
Each sample was imaged using one of the previously listed sequences at 22 different orientations in relation to the B 0 field, ranging from 0 • to 200 • at 10 • increments, with an additional scan at  = 55 • .Six of the imaged samples were additionally scanned using all four sequences at sample orientations of 0 • and 55 • .The exact combinations of imaged sequences for each sample are provided in Table 2.The scan time at a single orientation varied between 5 and 10 min, depending on the sequence.The total measurement time of a single sample was approximately 4-5 h.

T A B L E 2
Measurements performed for each sample (a-h).Note: Column "22-angle parameter" describes the quantitative MRI parameter measured at 22 orientations.Column "Additional scans" lists the additional parameters scanned at sample orientations 0 and 55.Columns 3 and 4 display the number of ROIs selected from the cartilage surfaces of each sample.Abbreviations: Ad-T 1ρ , adiabatic T 1ρ -weighted; CW-T 1ρ , continuous wave T 1ρ -weighted; ROI, region of interest.

Analysis
The T 2 or T1ρ-weighted images were each coregistered to a single geometry using a combination of manual image rotation and elastix (version 5.0). 34,35All images were coregistered to the orientation at which the tibia of the sample was parallel to the main magnetic field.After coregistration, the T 2 or T1ρ relaxation time maps were created from each echo series via pixelwise fitting of a mono-exponential relaxation equation.The fits were performed using a nonlinear least-squares estimation (MATLAB, ver.R2019b; Natick, MA, USA).Multiple regions of interest (ROIs) were selected from the relaxation time maps of both femoral and tibial cartilage of each scanned sample.These ROIs were selected from regions of cartilage with minimal curvature of both the bone-cartilage interface and cartilage surface.In general, 2-3 suitable rectangular ROIs could be selected from each femoral and tibial cartilage surface (see example in Figure 1B), with the exact number of drawn ROIs given in Table 2. On average, each ROI had a width of 11 pixels and a depth of 10 pixels.After image coregistration, the ROIs were manually transferred to the same coordinates in each rotated image.Pixels with relaxation time values longer than 300 ms were discarded from the drawn ROIs.In some cases, the coregistered image sets encountered pixel shifts of 1-3 pixels, resulting in the transferred ROIs being slightly off target.These cases were addressed by manually adjusting the location of the ROI to better match the desired cartilage location.Each ROI was further split depth-wise into two segments of approximately similar size: a surface segment containing the superficial 50% of cartilage volume and deep 50% segment consisting of the remaining cartilage closer to the bone-cartilage interface.Relaxation rates at each sample orientation for each ROI segment were calculated from the corresponding averaged relaxation times (R = 1/avg[T]).The relaxation components of each segment were then estimated using a model presented by Momot et al. 17 1 where R i and R a represent the isotropic and anisotropic fitting components of relaxation, respectively;  represents the nominal sample orientation; and Δ represents an added offset term between the nominal sample orientation and estimated collagen fiber orientations.Fitting quality was estimated using RMS error (RMSE) analysis.
Anisotropy of each parameter was estimated by calculating the R a /R i ratios for each ROI segment, whereas the depth dependency of both R a and R i within each ROI were estimated by calculating the ratios of R a (Deep)/R a (Surface) and R i (Deep)/R i (Surface).To study the anisotropy of the parameters within-sample, paired t-tests were performed for relaxation rates measured at sample orientations 0 • and 55 • .The ROIs used in these tests were selected from areas in femur and tibia with both minimal tissue curvature and articular surface normal parallel to B 0 .Furthermore, the within-sample comparisons of anisotropy of the different parameters were performed using paired t-tests of the relative differences . Topographical changes in relaxation anisotropy across both cartilage depth and along the cartilage surface of the scanned samples were illustrated via the creation of anisotropy maps.The anisotropy values (A) of these maps were calculated pixelwise from the corresponding relaxation time maps using the Michelson contrast as follows 6,36 : where T k,max and T k,min describe the maximum and minimum relaxation times over the different physical orientations of the sample.

RESULTS
A selection of representative plots (Figures 3, 4, S1, and S2) displays the differences in orientation dependence of the measured parameters.In particular, the R 2 of both femoral (Figure 3A) and tibial (Figure 4C) deep segments show notable differences in the measured relaxation rates across different sample orientations, indicating strong anisotropy.Similar behavior can be seen in deep ROI segments of R 1ρ (300 Hz) (Figures 3B and 4C), although to a lesser extent.The observed orientation dependence becomes notably lesser in R 1ρ (500 Hz) (Figures 3C  and 4C) and has disappeared more or less entirely for R 1ρ,Ad (Figures 3D and 4D).In comparison, the surface segments display a low to negligible degree of anisotropy in most cases, although some orientation dependence can be seen in Figures 3A,C and 4A.The fit quality of the representative plots was in general good.RMSE values of both femoral and tibial fits were mostly at or below 1.4.The worst fit quality was seen in the deep segment fits of R 1ρ (300 Hz), with femoral and tibial RMSE values of 6.23 and 2.58, respectively.
The anisotropic and isotropic relaxation components of all fitted femoral and tibial ROI segments (Figure S3) show in general faster anisotropic relaxation in the deep cartilage compared with the surface cartilage.In the surface cartilage (Table 3), the isotropic relaxation Orientation dependence of the relaxation rates and anisotropy model fits of a representative subset of collected deep and surface segments of femoral cartilage regions of interest.
Orientation dependence of the relaxation rates and anisotropy model fits of a representative subset of collected deep and surface segments of tibial cartilage regions of interest.

T A B L E 3
Mean relaxation components and SDs for surface and deep cartilage segments.components were faster than their corresponding anisotropic components and showed no notable variation across the measured parameters.In deep cartilage (Table 3), the mean isotropic components showed only comparatively small differences, whereas the differences between the anisotropic of different sequences were at their most prominent.In deep cartilage, the mean R 2,a = 36.61/s (∼27.3 ms) was found to be the fastest, followed by R 1ρ,a (300 Hz) = 12.5 1/s (∼82.0 ms), R 1ρ,a (500 Hz) = 7.5 1/s (∼133.3ms), and R 1ρ,Ad,a = 1.5 1/s (∼666.7 ms).In surface cartilage segments, R 2,a was found to comprise an average of 42% of total relaxation, whereas R 1ρ,a (300 Hz) comprised an average of 29%; R 1ρ,a (500 Hz) comprised of 32%, and R 1ρ,Ad,a comprised 18% of the total relaxation.In the deep cartilage segments, R 2,a was found to comprise on average of 71% of the total relaxation rate, whereas R 1ρ,a (300 Hz) comprised 40%, R 1ρ,a (500 Hz) comprised 35%, and R 1ρ,Ad,a comprised 16% of the total relaxation.

Superficial cartilage
In both femoral and tibial cartilage, the ratios of R a /R i for R 2 , R 1ρ (300 Hz) and R 1ρ (500 Hz) were found to be on average higher in deep segments compared with surface segments, whereas R 1ρ,Ad had similar and small ratios in both segments (Figure 5A,B).R 2 had the highest average ratios of R a /R i in both femoral (surface segment: 0.6 ± 0.4; deep segment: 2.5 ± 1.0) and tibial cartilage (surface segment: 1.3 ± 0.8; deep segment: 3.7 ± 2.4), whereas R 1ρ,Ad had the lowest.The ratios of R 2 displayed large amounts of variation compared with the other parameters.The average R a /R i ratios for R 1ρ (300 Hz) and R 1ρ (500 Hz) were found to be within each other's errors across all studied segments.The ratios of isotropic and anisotropic relaxation components between deep and surface cartilage segments (Figure 6C,D) displayed low depth dependence for isotropic and a highly varying amount of dependence for the anisotropic components.For the anisotropic relaxation component, the average R a (Deep)/R a (Surface) ratios followed the previously shown order of depth dependence of R 2 (femur: 9.2 ± 11.2; tibia: 4.3 ± 3.2), R 1ρ (300 Hz) (femur: 3.7 ± 2.3; tibia: 2.6 ± 1.8), R 1ρ (500 Hz) (femur: 2.6 ± 1.6; tibia: 1.6 ± 1.3) and R 1ρ,Ad (femur: 0.8 ± 0.23; tibia: 1.2 ± 0.4).In comparison, the R i (Deep)/R i (Surface) ratios showed no depth dependence in the isotropic relaxation components.
The comparisons between relaxation rates at sample orientations 0 • and 55 • (Table 4) show significant differences for R 2 , R 1ρ (300 Hz), and R 1ρ (500 Hz).Ad-R 1ρ in comparison showed no significant difference between the two sample orientations.The relative difference in relaxation rates (Table 4) was significantly larger in R 2 than R 1ρ (300 Hz), whereas in R 1ρ (300 Hz) the difference was significantly larger than in R 1ρ (500 Hz).No significant difference was found between the R 1ρ (500 Hz) and Ad-R 1ρ .
Visual inspection of the anisotropy maps computed with Eq. ( 2) showed large differences in anisotropy among the measured qMRI parameters (Figure 6).Many of the maps also displayed notable differences in anisotropy across the topology of the considered samples.The largest and most consistent variations in anisotropy values were observed for R 2 (Figure 6A,B).In these maps, deep cartilage showed notably large anisotropy values ranging between 60% and 75%, whereas the cartilage near the surface showed comparatively low anisotropy values of 30%-50%.The anisotropy maps of R 1ρ (300 and 500 Hz) (Figure 6C,F) demonstrated a more complex set of results, Anisotropy maps of the scanned joint samples.The quantitative MRI (qMRI) parameters of the samples were T 2 (A,B), T1ρ (300 Hz) (C,D), T1ρ (500 Hz) (E,F), and adiabatic T 1ρ -weighted (G,H).

T A B L E 4
Pairwise comparisons between the relative differences in mean relaxation rates at sample orientations 0 and 55.
Parameter R(0 with notable differences between femoral and tibial cartilage of the studied samples.For example, Figure 6D shows large anisotropy values in the deep femoral cartilage, whereas in the deep tibial cartilage of the same sample, R 1ρ (300 Hz) was observed to be largely isotropic.Some of the maps (Figure 6C) also displayed large portions of the map as being 100% anisotropic.The anisotropy maps of R 1ρ,Ad (Figure 6G,H) displayed a low amount of anisotropy across all layers of both femoral and tibial cartilage.

DISCUSSION
The purpose of this work was to study the anisotropy of multiple clinically feasible magnetic relaxation parameters in bovine articular cartilage at 3T field strength.The results showed that R 2 and, to a lesser degree, R 1ρ , are significantly affected by the orientation of the joint surface with respect to B 0 .The observed R 2 and R 1ρ anisotropies were found to exhibit similar magic-angle behavior to studies performed at both higher and lower field strengths. 7,10,13,24,26,37The anisotropic relaxation components were found to be, in general, faster in the deep cartilage segments compared with the surface segments.R 2 was found to be the most anisotropic of the studied qMRI parameters, with R 2,a comprising on average of 71% of total relaxation rate.Conversely, the isotropic relaxation components showed only small variation across different qMRI parameters.Similar to a previous study by Zheng et al., 27 R 2 was found to be anisotropic in both surface and deep cartilage.The observed CW-R 1ρ anisotropy was found to be inversely proportional to the spin-lock strength, as seen in previous studies performed at both 4.7 T and 9.4 T. 11,12 The relative insensitivity of Ad-T 1ρ on cartilage orientation, previously demonstrated at 9.4 T by Hänninen et al., 11 likely results from the shapes of the amplitude and frequency modulation functions of the hyperbolic secant pulse used in the measurement sequence.It is worth highlighting that, despite the lower maximum B 1 amplitude used at 3 T imaging, the same effect of negligible orientation dependence was seen.Where observed, the anisotropy was found to exhibit similar depth dependence to the findings made in 9.4 T studies. 10,11elaxation parameters in the deep segments of cartilage were in general more anisotropic as compared with the corresponding surface segments.Especially in the case of R 2 , deep segment R a was almost 3 times larger than the corresponding R i , indicating high tissue anisotropy.This coincides well with previous knowledge of the strength of the magic-angle effect being related to the collagen structure within deep cartilage, which is more uniformly aligned compared with transitional or superficial cartilage. 4In the case of R 2 , the deep segment mean anisotropic relaxation component was over 4 times faster than its surface segment counterpart.Any relaxation averages taken across cartilage depth would in these cases have been dominated by the deep segment relaxation.
Of the fitted anisotropic relaxation components, those of R 2,a were in general fastest, followed by R 1ρ,a (300 Hz).R 1ρ,a (500 Hz) were comparatively slow, whereas R 1ρ,Ad,a were close to zero.The isotropic relaxation components in the surface cartilage segments showed only small differences across the relaxation parameters.The relaxation anisotropy model was found to function well, with most of the fits having RMSE values at or below 1.4.The R a /R i and R(Deep)/R(Surface) ratios calculated from the mean relaxation rates confirm the previously presented order of anisotropy and degrees of depth dependence.The same order of anisotropy for the parameters can also be seen in the comparisons calculated across two sample orientations and between the parameters.
The visual inspection of the full-slice anisotropy maps also showed the different levels of anisotropy across the studied parameters.In the case of T 2 , the anisotropy maps showed systematic changes in anisotropy across cartilage depth, with deep cartilage being more anisotropic compared with the cartilage closer to surface.The high overall anisotropy in these maps coincides well with total R 2 being dominated by R 2,a .The faster deep segment relaxation components also appear to coincide with the larger anisotropy values in areas of deep cartilage.As the anisotropic relaxation component becomes less dominant in other qMRI parameters, so too does the overall anisotropy of the maps decrease, following the same pattern of decreasing anisotropy in order of R 1ρ (300 Hz), R 1ρ (500 Hz), and finally R 1ρ,Ad .
As shown here and in previous studies, 10,11 T 2 was found to be very sensitive to both cartilage orientation with respect to the direction of the B 0 field and to the histological depth of the studied cartilage segment.CW-T 1ρ relaxation at low spin-lock and B 1 fields was also found to be affected by collagen orientation, whereas in previous studies the frequency of the spin-lock field appeared to affect the specificity of the measurements to proteoglycans. 12,38Nevertheless, as shown by the current study and many others, 6,23,39 the collagen-related anisotropic relaxation is also dependent on the spin-lock amplitude.These findings suggest that using T 1ρ as an indicator for solely proteoglycan content in cartilage 12,[40][41][42] is not feasible at low spin-lock amplitudes.As the bone ends of both femur and patella are naturally curved, a knee joint placed in a magnetic field will contain cartilage-and by extension collagen fibers-at various orientations in relation to the main magnetic field.Thus, the basic prerequisite for the magic angle effect is present in practically all MRI of a healthy knee joint.Although a robust model for the orientation dependence in cartilage could potentially allow the mitigation of the effect in clinical imaging, it is not within the scope of this study.
Ad-T 1ρ was found to be the parameter least affected by collagen orientation as compared with T 2 and CW-T 1ρ .As seen both here and in previous studies on patellar cartilage at both preclinical 6,11 and clinical field strengths, 43 Ad-T 1ρ in cartilage is seen to be orientation invariant, and the parameter appears to have the potential to serve as a baseline for isotropic relaxation time in clinical measurements.This, combined with the reported sensitivity of Ad-T 1ρ relaxation times to cartilage damage, 13,33 alludes to the possibility for the use of the parameter as a tool for estimating the condition of the internal structure of cartilage without having to account for the magic-angle effect.

Limitations
The scanning time of a single sample at 22 angles was very long, between 4 and 5 h, even with compromises made to shorten the scanning speed, such as omission of multiscan averaging from T 2 and CW-T 1ρ sequences.This effectively prevented fully scanning each sample with more than one sequence, necessitating instead the additional scans at sample orientations 0 and 55 as seen in Table 2.The cartilage ROIs used in analysis were limited to intact regions of cartilage containing comparatively straight and parallel surfaces and bone-cartilage interfaces.This selection process was chosen to ensure collagen within each cartilage section was aligned as uniformly as possible.The two-segment ROI scheme was chosen as a method of performing a depth-wise analysis without including any selection bias that would have resulted from attempting to segment the cartilage in a zone-wise manner.This method does, however, lead to some uncertainty in the exact collagen orientation in each individual segment and ROI.Depending on the thickness of the underlying cartilage zones, the superficial segments can contain a mixture of superficial, transitional, and possibly even deep cartilage, whereas the deep segments can contain a mixture of transitional and deep cartilage.The extent of transitional and radial zone relative to cartilage thickness can vary not only across measured joints, between different anatomical locations within a joint (femoral/tibial cartilage), but also between different locations across a cartilage surface.These variances in the underlying collagen orientation within the measured segments are a likely cause for the observed intrasample and intersample differences (Figure S4), and more in general the variance in the observed magic-angle effect.A detailed comparison with histological measurements, such as performed by Hänninen et al., 13 would have been needed to both accurately compare collagen fiber alignment to the measured anisotropy, and to segment the ROIs into more accurate cartilage zones.In this study, the samples were rotated manually between each scan, introducing the potential for human error during sample reorientation.Although an automated sample rotation platform similar to one used in a recent high-field study by Leskinen et al. 10 was considered, it was in this case deemed impractical due to constraints imposed by both the scanning environment, as well as sample size.During the measurements of each sample, the flex coil had to be manually lifted and lowered back onto the sample after each sample rotation.The shifting location of the receiver coil around the sample likely caused small sample movements, which are the most likely cause for the image shifts observed between subsequent scans of a single sample.These shifts and the resultant inaccuracies in image coregistration are also believed to be the primary source of errors in both the anisotropy mapping, likely significantly contributing to the large 100% anisotropic regions seen in Figure 6C as well as in the fits for 22 angles shown in Figure 3. Partial volume effect, possibly compounded by unintended sample movement between measurements, is assumed to be the main reason for varying fit quality.It is worth noting that ROI-wise analysis is less affected by this error, as multiple voxels were used, and the ROI locations were manually adjusted to match as well as possible.
Translating these types of measurements to in vivo would introduce a variety of new challenges beyond the scope of this study (e.g., maintaining sufficient rotation angles for fitting).As such, the principal purpose of this study was to demonstrate the varying degrees of orientational dependence of the studied contrasts in 3 T.However, low-field scanners using open-core magnets or rotating B 0 fields could allow a sufficient number of sample orientations to be scanned, and as such hold potential for future in vivo application.

CONCLUSIONS
In conclusion, we have demonstrated that T 2 and T 1ρ relaxation demonstrates notable variation depending on the sample orientation, magnetization preparation, and studied articular cartilage layer.T 2 and CW-T 1ρ relaxation at 3 T demonstrates significant dependence on collagen orientation, and the amount of anisotropy in CW-T 1ρ is dependent on spin-lock amplitude.Adiabatic T 1ρ was found to be nearly isotropic and can be used to suppress the anisotropic relaxation component of T 1ρ .The anisotropic component of relaxation should be considered a significant source of variation in qMRI of curved articular surfaces.
both the collagen structure within articular cartilage and the surface/deep segmentation.(B) Representative sample illustrating region-of-interest selection and segmentation into surface and deep segments.
of the sample rotation platform.(B) Image of the scanning setup.The joint sample has been fixed onto the sample rotating platform, with the body coil placed on top of the sample.The rotation platform itself has been fixed onto the patient bed of the clinical 3T MRI device.

5
Relaxation parameter ratios of femoral and tibial cartilage segments.(A,B) The averaged ratios between Ra and Ri for surface and deep segments in tibial (A) and femoral (B) cartilage.(C,D) The ratios between deep and surface cartilage for Ra and Ri in tibial (C) and femoral (D) cartilage.
Imaging parameters of the quantitative MRI sequences.
• C and thawed to room temperature at 21 • C for a period of 24 h before imaging.Each thawed sampleT A B L E 1Abbreviations: Ad-T 1ρ , adiabatic T 1ρ -weighted; CW-T 1ρ , continuous wave T 1ρ -weighted; HS4, hyperbolic secant pulse; MESE, multi-echo spin echo.a One sample scanned with 20 slices.b FOV and matrix were adjusted to accommodate varying sample sizes while maintaining the same resolution.