Rapid Whole-Knee Quantification of Cartilage Using <inline-formula><tex-math notation="LaTeX">$T_{1}$</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">$T_{2}^*$</tex-math></inline-formula>, and <inline-formula><tex-math notation="LaTeX">$T_{RAFF2}$</tex-math></inline-formula> Mapping With Magnetic Resonance Fingerprinting

<italic>Objective:</italic> Quantitative Magnetic Resonance Imaging (MRI) holds great promise for the early detection of cartilage deterioration. Here, a Magnetic Resonance Fingerprinting (MRF) framework is proposed for comprehensive and rapid quantification of <inline-formula><tex-math notation="LaTeX">$T_{1}$</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">$T_{2}^*$</tex-math></inline-formula>, and <inline-formula><tex-math notation="LaTeX">$T_{RAFF2}$</tex-math></inline-formula> with whole-knee coverage. <italic>Methods:</italic> A MRF framework was developed to achieve quantification of Relaxation Along a Fictitious Field in the 2nd rotating frame of reference (<inline-formula><tex-math notation="LaTeX">$T_{RAFF2}$</tex-math></inline-formula>) along with <inline-formula><tex-math notation="LaTeX">$T_{1}$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$T_{2}^*$</tex-math></inline-formula>. The proposed sequence acquires 65 measurements of 25 high-resolution slices, interleaved with 7 inversion pulses and 40 RAFF2 trains, for whole-knee quantification in a total acquisition time of 3:25 min. Comparison with reference <inline-formula><tex-math notation="LaTeX">$T_{1}$</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">$T_{2}^*$</tex-math></inline-formula>, and <inline-formula><tex-math notation="LaTeX">$T_{RAFF2}$</tex-math></inline-formula> methods was performed in phantom and in seven healthy subjects at 3 T. Repeatability (test-retest) with and without repositioning was also assessed. <italic>Results:</italic> Phantom measurements resulted in good agreement between MRF and the reference with mean biases of −54, 2, and 5 ms for <inline-formula><tex-math notation="LaTeX">$T_{1}$</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">$T_{2}^*$</tex-math></inline-formula>, and <inline-formula><tex-math notation="LaTeX">$T_{RAFF2}$</tex-math></inline-formula>, respectively. Complete characterization of the whole-knee cartilage was achieved for all subjects, and, for the femoral and tibial compartments, a good agreement between MRF and reference measurements was obtained. Across all subjects, the proposed MRF method yielded acceptable repeatability without repositioning (<inline-formula><tex-math notation="LaTeX">$R^{2}\geq$</tex-math></inline-formula> 0.94) and with repositioning (<inline-formula><tex-math notation="LaTeX">$R^{2}\geq$</tex-math></inline-formula> 0.57) for <inline-formula><tex-math notation="LaTeX">$T_{1}$</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">$T_{2}^*$</tex-math></inline-formula>, and <inline-formula><tex-math notation="LaTeX">$T_{RAFF2}$</tex-math></inline-formula>. <italic>Significance:</italic> The short scan time combined with the whole-knee coverage makes the proposed MRF framework a promising candidate for the early assessment of cartilage degeneration with quantitative MRI, but further research may be warranted to improve repeatability after repositioning and assess clinical value in patients.

2 , and T RAF F 2 methods was performed in phantom and in seven healthy subjects at 3 T. Repeatability (test-retest) with and without repositioning was also assessed.Results: Phantom measurements resulted in good agreement between MRF and the reference with mean biases of −54, 2, and 5 ms for T 1 , T * 2 , and T RAF F 2 , respectively.Complete characterization of the whole-knee cartilage was achieved for all subjects, and, for the femoral and tibial compartments, a good agreement between MRF and reference measurements was obtained.Across all subjects, the proposed MRF method yielded acceptable repeatability without repositioning (R 2 ≥ 0.94) and with repositioning (R 2 ≥ 0.57) for T 1 , T * 2 , and T RAF F 2 .Significance: The short scan time combined with the whole-knee coverage makes the proposed MRF framework a promising candidate for the early assessment of cartilage degeneration with quantitative MRI, but further research may be warranted to improve repeatability after repositioning and assess clinical value in patients.

I. INTRODUCTION
E ARLY detection of cartilage damage in the knee is essential to prevent or decelerate progressive and/or irreversible joint damage [1].Noninvasive measurement of quantitative MRI biomarkers allows the detection of functional and nonmorphological changes beyond standard clinical qualitative MRI techniques [2].Several quantitative biomarkers have shown sensitivity to cartilage degeneration, including, T 1 relaxation times, which demonstrated sensitivity to changes in the proteoglycan content [3] or T * 2 relaxation times, which are sensitive to water content and collagen fiber network changes [4], [5].
Rotating frame of reference (RFR) relaxation measurements, such as T 1ρ mapping [6], have shown promising results as a quantitative biomarker due to high sensitivity to slow molecular motion in the cartilage [7].RFR measurements are most commonly obtained using spin-lock pulses to induce variable T 1ρ weighting in a series of images.However, spin-lock pulses are typically long pulses with high amplitudes, and their application at high field strength (3 T and above) is hampered by limits of the specific absorption rate (SAR) [8].Relaxation along a fictitious field (RAFF) is an alternative RFR contrast, which employs amplitude-and frequency-modulated pulses in the subadiabatic regime to create RFR contrast [9].Recently, the relaxation time of RAFF in the second rotating frame of reference (T RAF F 2 ) has shown promising results in the detection of cartilage degeneration in different ex vivo species (bovine [10], rabbits [11], [12], piglets [13], and ponies [14]).Compared to T 1ρ mapping, at the same peak RF amplitude, RAFF mapping has low SAR requirements with good resilience against B 0 and B + 1 inhomogeneities [9].Together, this renders T RAF F 2 as an extremely promising contrast mechanism and quantitative biomarker, for high-field imaging of the articular cartilage.
Conventional quantitative mapping techniques usually acquire only one single-parameter map per scan by generating multiple, single-parameter weighted images, which are then fitted voxel-wise to the corresponding signal model.However, different relaxometry-based biomarkers provide complementary information about the cartilage state and no single biomarker has been accepted as the standard [15].Therefore, multiparametric quantification bears promise for a more sensitive and specific diagnosis of cartilage degeneration [16].Independent acquisition of single-parameter maps requires long scan times, and the need to coregister the different maps hampers image fusion.Magnetic resonance fingerprinting (MRF) has been proposed as a promising approach to this problem by varying multiple acquisition parameters such as flip angle (FA), echo time (TE), and repetition time (TR) simultaneously throughout a single acquisition [17].The resulting signal evolution is subsequently matched to a simulated signal response for the underlying tissue parameters, resulting in inherently co-registered relaxation maps.Recently, MRF has been successfully applied for T 1 , T 2 , and T 1ρ tissue characterization of the human knee articular cartilage in 2D [18] and 3D [19], allowing for differentiation between mild osteoarthritis patients and healthy subjects.MRF was also applied to the human knee for simultaneous estimation of different relaxometry biomarkers (e.g., T 1 , T 2 ) and fat fraction [20], [21], and blood flow velocity maps [22].MRF based on Cartesian echo-planar imaging (EPI) readout (MRF-EPI) has been introduced, enabling rapid multiparameter mapping and efficient multislice coverage with interpretable and high-SNR baseline images [23], [24].
In this study, we explore the use of MRF-EPI for efficient, simultaneous multiparameter mapping of T 1 , T * 2 , and T RAF F 2 of the whole-knee.A slice interleaved MRF-EPI sequence is extended to measure T 1 , T * 2 , and T RAF F 2 in clinically acceptable scan times.Phantom and in vivo measurements are performed at 3 T to investigate the accuracy and repeatability of the proposed method.

A. Sequence Design
Fig. 1 shows a schematic of the proposed pulse sequence.A slice-interleaved multislice MRF-EPI sequence [24] was extended to incorporate the measurement of T 1 , T * 2 , and T RAF F 2 .Several gradient-echo planar imaging measurements per slice are acquired with varying FA and TE, to encode T 1 and T * 2 .A spectrally selective fat saturation module [25] is applied before each EPI readout excitation, as commonly used to suppress chemical-shift artifacts in EPI-based sequences [26].Adiabatic inversion pulses are distributed throughout the initial part of the sequence for enhanced T 1 encoding.T RAF F 2 quantification was achieved by adding RAFF pulse trains throughout the last part of the proposed sequence.The RAFF trains vary in duration and comprise one or more RAFF2 pulses [9] interleaved with short spoiler blips, as shown in Fig. 2.
Image reconstruction was performed inline at the scanner using the vendor's implementation of correction for gradient delay errors, eddy-current induced deviations, and distortion corrections [27].Dictionary calculation and matching were generated offline using MATLAB (The MathWorks; Natick, MA, USA).The dictionary was simulated using the Bloch equations assuming an isochromatic voxel.Gradient and RF spoiling was assumed to be complete, and simulated by complete loss of the transverse magnetization in the isochromat.Perfect inversion was assumed after the inversion pulses, which were considered to be ideal (180 • ) with no T 2 decay during the pulse.To compensate for deviations from the nominal flip angle excitation pulse, due to imperfect slice profiles excitation and inhomogeneities in the transmit field, a B + 1 correction was implemented within the dictionary [23], [28].Dictionary matching was performed by choosing the entry with the highest inner product between the magnitude of the dictionary entry and the magnitude of the measured signal.RAFF pulses were modeled as a two-parameter monoexponential decay function, S(t) = A • e − t T RAF F 2 , assuming negligible steady-state magnetization [9] in the dictionary generation.The Levenberg-Marquardt algorithm was used as the optimization method.T 1 times ranged from 100 to 3500 ms, and T * 2 and T RAF F 2 ranged from 10 to 400 ms, all with 5% increments.Unrealistic entries with T 1 < T * 2 were discarded.FA efficiency (B + 1 ) ranged from 0.6 to 1.2 with increments of 0.1, resulting in a total dictionary size of 2,765,544 entries.Total time for dictionary generation was 64 min, with the matching taking 7 min for 25 slices.Implementation of the proposed method is provided online (https://gitlab.tudelft.nl/mars-lab/mrf-epi-raff2).

B. MR Imaging
Phantom and in vivo imaging was performed on a 3 T MRI scanner (Ingenia; Philips, Best, The Netherlands) with an 8channel knee coil.
Phantom measurements, using the T1MES phantom [29], were performed to evaluate the bias and precision of the MRF sequence compared to the reference measurements.To assess repeatability of the proposed MRF method, phantom measurements were acquired once without re-positioning (test -re-test 1) and again after re-positioning (test -re-test 2).Additionally, to assess the performance of the MRF sequence, five different flip angle patterns were used and the measured T 1 , T * 2 , and T RAF F 2 were compared.
The following parameters were kept constant for MRF-EPI and reference maps for both the phantom and in vivo experiments: FOV = 160 mm × 160 mm, matrix size (base resolution) = 200 × 200 (0.8 mm × 0.8 mm), slice thickness = 3 mm.RAFF2 was performed with an RF peak amplitude of 500 Hz (corresponding to 11.74 µT B 1 ).
For T 1 quantification, a single-slice reference inversionrecovery gradient echo was performed with TI = 50, 100, 200, 400, 800, 1600, 3200 ms, FA = 35  2 .Reference T RAF F 2 maps were computed offline in MATLAB using qMRLab [31] by applying a two-parameter mono-exponential decay model, S(t) = A • e − t T RAF F 2 .Implementation of the proposed method is provided online (https://gitlab.tudelft.nl/mars-lab/mrf-epi-raff2) In vivo imaging was performed in 7 healthy subjects (5 female, 2 male, 29 ± 11 years, range: 19-52 years).The study was performed in accordance with the ethical principles of the Declaration of Helsinki.Ethical approval was obtained from the local research ethics committee.All subjects were informed of the study and provided written informed consent prior to the examination.Multislice MRF and single-slice reference maps were acquired in sagittal orientation, using the same imaging parameters as the phantom experiments.To assess the accuracy of the MRF technique, the areas of the tibial and femoral cartilage were manually segmented, based on landmarks, using MATLAB on single-slice reference scans and on the corresponding MRF slice.Mean T 1 , T * 2 , and T RAF F 2 of the cartilage segments were calculated.To assess repeatability, the MRF sequence was repeated twice, without repositioning (test -re-test 1).To measure intra-subject repeatability with repositioning, additional test/retest MRF measurements were performed (test -re-test 2), where the subject was asked to exit the scanner bore, and a new set of MRF images was acquired after repositioning [33].The three central slices for each medial and lateral compartments were Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.manually selected for all scans.Central slices were defined as the slice for which the femoral cartilage reached its most inferior position.Subsequently, manual segmentation of the femoral and tibial cartilage areas was performed in those six slices.Across all subjects, correlation and Bland-Altman (BA) analysis were performed for T 1 , T * 2 , and T RAF F 2 .

III. RESULTS
Fig. 3 shows a representative slice using the proposed MRF method and the reference methods, as well as the correlation analysis and the respective Bland-Altman plots of the three relaxation times in the nine vials.T 1 , T * 2 , and T RAF F 2 measured with the proposed MRF-EPI sequence were in excellent agreement with the reference protocols with R 2 = 0.99, 0.96, and 0.89, respectively.A bias of −54, 2, and 5 ms was observed for T 1 , T * 2 , and T RAF F 2 , respectively.Excellent repeatability with and without re-positioning was observed for T 1 , T * 2 , and T RAF F 2 in phantom (Fig. S1).Comparable T 1 , T * 2 , and T RAF F 2 was measured with five different flip angle patterns, as shown in Fig. S2.
Data acquisition was successfully performed for all the subjects, below the SAR limits for in vivo scans and no re-shimming was necessary.Fig. S3 shows example baseline images acquired with the proposed MRF method along with representative fingerprints in cartilage ROIs.Fig. 4 shows T 1 , T * 2 , and T RAF F 2 maps of the 25 acquired slices, for a representative subject, providing whole-knee characterization with medial to lateral coverage.The T 1 and T * 2 maps depict largely homogeneous signal in the cartilage.T RAF F 2 maps exhibited a higher level of signal variation in the cartilage and residual susceptibility to image artifacts.
Fig. 5(a) shows representative slices of T 1 , T * 2 , and T RAF F 2 maps acquired with the proposed MRF-EPI sequence and the reference protocols in two healthy subjects.A clear depiction of the cartilage was achieved, visually distinct from the bone area, on both the MRF and the reference maps.The MRF maps presented a higher level of image distortion, when compared to the reference maps, due to the EPI readout.Fig. 5(b) shows a comparison between the mean femoral and tibial values measured with the MRF and the reference protocols for the seven subjects.Across all subjects, the average T 1 ± SD values were 1010.5 ± 104.5 ms (femoral) and 1002.3 ± 157.9 ms (tibial) for MRF and 1128.7 ± 107.8 ms (femoral) and 1185.2 ± 133.5 ms (tibial) for the reference maps.T * 2 values were 29.6 ± 3.4 ms (femoral) and 24.8 ± 4.0 ms (tibial) for MRF and 27.1 ± 3.7 ms (femoral) and 20.0 ± 8.6 ms (tibial) for the reference maps.T RAF F 2 values were 36.7 ± 9.6 ms (femoral) and 38.5 ± 8.5 ms (tibial) for the MRF and 45.2 ± 11.3 ms (femoral) and 51.0 ± 16.9 ms (femoral) for the reference maps.The bias of the measured T 1 , T * 2 , and T RAF F 2 with the proposed MRF-EPI sequence and the reference protocols were −166, 4, and −8 ms, respectively, which were in line with the phantom results.For each subject, the average (± SD) T 1 , T * 2 , and T RAF F 2 , as well as the respective Coefficient of Variation (CoV) of the femoral and tibial regions are shown in Fig. S4 measured with MRF and the reference methods.The variability within the cartilage of single subjects is comparable between MRF and the reference methods.

IV. DISCUSSION
In this work, whole-knee quantification of T 1 , T * 2 , and T RAF F 2 in the articular cartilage was achieved in less than four minutes.In phantom, the proposed MRF-EPI sequence achieved high accuracy with respect to the reference methods.In healthy subjects, a satisfactory level of intra-subject repeatability with and without repositioning was obtained.This is in line with Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.previous MRF-EPI studies in the brain [24] and the kidneys [34] where accurate, whole organ, multiparametric mapping of T 1 and T * 2 was obtained.In the recent literature, a wide range of T 1 values of knee cartilage measured at 3 T have been reported.Mittal et al. [3] measured an average femoral and tibial T 1 of 912.5 ± 10.8 ms and 902.3 ± 22.0 ms, respectively, whereas Sharafi et al. [18] reported a global average T 1 of 778.3 ± 48.5 ms in healthy tissue.Our T 1 quantification yields slightly higher values compared to the literature and also shows a slight underestimation with respect to the reference methods, in phantom and in vivo.This underestimation can likely be attributed to imperfect inversion efficiency, as the proposed MRF sequence relies on multiple inversions to enhance the sensitivity to T 1 .To increase the robustness to B 0 and B + 1 , an adiabatic full passage pulse with relatively long pulse duration was used.During the pulse a nonnegligible amount of relaxation is incurred, which reduces the inversion efficiency [30].This can potentially be alleviated by alternative inversion pulses, or by taking the magnetization relaxation during the adiabatic inversion pulse into account.However, this would require a more complex dictionary and a larger number of fit variables at the cost of longer computation times and loss in precision.In this work, seven inversion pulses were used, as a trade-off between T 1 precision and scan-time efficiency gains, afforded by acquiring four interleaved slices after a single global inversion pulse, as previously shown [24].This number of pulses will impact the precision of the measured T 1 , and might not be the optimal value for different regions-ofinterest.Moreover, in this work, MRF maps were obtained from water-only baseline images due to the use of fat suppression.This may lead to different quantification results than, for example, inversion recovery-based methods, which generate images that include both fat and water signals.Further investigation of clinical sensitivity using water-only maps is warranted in the relevant patient cohorts.Additionally, distortion correction may lead to blurring of the point spread function.However, since it applies to all baseline images, it is not expected to impart bias.
The mean values of the femoral and tibial cartilage T * 2 obtained with the proposed MRF-EPI sequence were in good agreement with previously reported values for 3 T MRI by Hesper et al. (29.8 ± 5.0 ms) [35] and Zhang et al. (26.4 ± 2.6 ms and 21.8 ± 1.8 ms for the femoral and tibial regions, respectively) [5].Moreover, in this work, a negligible bias was observed for T * 2 obtained with MRF-EPI and a gradient echo reference sequence.Additionally, the fitting procedures used to estimate relaxation time constants assumed that the signal followed a mono-exponential curve.However, recent research suggests that more complex curves, such as multi-exponential functions, may be necessary to accurately model both longitudinal [36], transverse [37], [38], [39] or rotating-frame [40], [41], [42] magnetization behaviors, which can take into account factors like partial volume effects, flow effects, magnetization transfer, and chemical exchange [43], [44].Further research is warranted to investigate the promise of combining recently emerging multi-component or multi-exponential MRF techniques with the proposed framework [45].T RAF F 2 times have not previously been obtained for articular cartilage in vivo, but the quantification with the proposed MRF-EPI method shows only negligible bias compared to a RAFF2-prepared reference method.T RAF F 2 maps exhibited a higher level of signal variation and a residual susceptibility to image artifacts compared to the other parameters.However, as RAFF2 operates in a subadiabatic regime, increased RF power may be used to improve resilience of the preparation modules against system imperfections.In the present study the RF peak amplitude was limited to 500 Hz, due to the use of the body coil transmitter.Other scanners or custom transceiver knee coils allow for higher RF power amplitude, and are promising for further improvements in RAFF mapping quality.
MRF maps achieved a substantial reduction in scan time compared to single-slice methods.MRF simultaneously provided T 1 , T * 2 , and T RAF F 2 maps in ≈8 seconds per slice, in contrast to the reference maps, which took ≈345, 74, 295 seconds per slice for T 1 , T * 2 , and T RAF F 2 , respectively.The time efficiency of the proposed sequence makes it a suitable candidate for quantitative rotating frame of reference (RFR) imaging of the knee joint Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.with or without load in the clinic, which can provide valuable information about joint characteristics [46], [47], [48].Recently, multislice RFR based on T 1ρ mapping quantification has been explored, including 3D-MAPPS [49] and MRF-T 1ρ [18], [50].In comparison, the interleaved slice design with integrated T 1 modeling in the MRF-EPI framework allows for a substantially lower number of preparation pulses per slice.The proposed MRF-EPI sequence uses 40 RFR-prep pulses distributed over 25 slices (< 2 preparations per slice) compared with ≈ 16 preparations per slice (3D-MAPPS) and 6 and 5 preparations (MRF-T 1ρ ), which used a linear ordering for slice acquisition.This approach further reduced the SAR burden and allowed for time-efficient scanning even with SAR heavy-preparation pulses.The multislice acquisition was performed with slice order to maximize the time between the acquisition of adjacent slices.Dictionaries were calculated independently for each slice, assuming minimal slice cross-talk.A simulation of the exact slice profile, and joint matching of all slices, can be explored to further reduce bias due to the slice profile and residual slice cross-talk.However, this comes at the expense of greatly increased reconstruction times, and further research is warranted to make these reconstructions feasible.
This study has several limitations.In this work, only healthy subjects were scanned to study the feasibility of the proposed MRF-EPI sequence.Future work should be performed to validate the use of the proposed pulse sequence for the early assessment of cartilage degeneration in a cohort of patients with different pathologies.Moreover, further investigation needs to be performed to minimize the negative effect of re-positioning in terms of the repeatability of the proposed method.Furthermore, high-order RAFFn (n > 2) [81] could provide further SAR reductions, which could be desired for different anatomies or different field strengths.Additionally, to improve the image quality of the computed maps with the proposed MRF sequence, filtering [34] could be applied to the baseline images to minimize the effect of EPI k-space sampling.Also, the SNR of the parametric maps could be further improved by optimizing the FA/TE train of the pulse sequence for the range of T 1 and T * 2 values expected in the knee cartilage.No image registration was performed between MRF and reference maps.In this study, Gaussian noise is implicitly assumed, as the inner product was used as the loss function in dictionary matching.However, when the SNR is too low, the Rician noise distribution increasingly deviates from Gaussianity, which can lead to bias in the measurements.This can be mitigated by using phase-sensitive reconstructions, which are specifically designed to handle Rician noise, particularly if no partial Fourier is used.Finally, in the reference methods, conventional Cartesian k-space sampling was employed instead of EPI, to obtain the highest image quality.

V. CONCLUSION
In this study, T 1 , T * 2 , and T RAF F 2 mapping of the whole-knee articular cartilage was achieved in just under four minutes.An MRF-EPI sequence was extended to include T RAF F 2 resulting in inherently co-registered high-quality maps, with good visual map quality and with clinically tolerable specific absorption rates (SAR).The proposed method also achieved good levels of repeatability with and without repositioning in phantom and in vivo.The short scan time combined with the quantification of the whole-knee makes the proposed MRF-EPI sequence an excellent candidate for the early assessment of cartilage degeneration, but further research may be warranted to improve repeatability with repositioning and assess clinical value in patients.
Rapid Whole-Knee Quantification of Cartilage Using T 1 , T * 2 , and T RAF F 2 Mapping With Magnetic Resonance Fingerprinting Joao Tourais , Telly Ploem , Tijmen A. van Zadelhoff, Christal van de Steeg-Henzen, Edwin H. G. Oei, and Sebastian Weingärtner Abstract-Objective: Quantitative Magnetic Resonance Imaging (MRI) holds great promise for the early detection of cartilage deterioration.Here, a Magnetic Resonance Fingerprinting (MRF) framework is proposed for comprehensive and rapid quantification of T 1 , T * 2 , and T RAF F 2 with wholeknee coverage.Methods: A MRF framework was developed to achieve quantification of Relaxation Along a Fictitious Field in the 2nd rotating frame of reference (T RAF F 2 ) along with T 1 and T * 2 .The proposed sequence acquires 65 measurements of 25 high-resolution slices, interleaved with 7 inversion pulses and 40 RAFF2 trains, for whole-knee quantification in a total acquisition time of 3:25 min.Comparison with reference T 1 , T *

Fig. 1 .
Fig. 1.(a) Flip Angle and Echo Time (TE) are varied for each of the 65 measurements.(b) Seven adiabatic inversion pulses are distributed throughout the sequence for improved T 1 encoding.RAFF contrast was achieved by adding 40 RAFF trains of varying duration throughout the latter part of the sequence.

Fig. 2 .
Fig. 2. Schematic of the proposed multislice MRF pulse sequence framework.(a) Whole-knee imaging is performed using a slice-interleaved scheme with 25 slices being acquired for each measurement.Preparation pulses, inversion, or RAFF trains, are placed at different time points of the interleaved slice acquisition (i.e., the slice number immediately following the preparation vary throughout the acquisition).(b) RAFF trains comprise several RAFF pulses separated with spoiler blips.RAFF employs amplitude-(purple) and frequency-modulated (pink) pulses in the subadiabatic regime.

Fig. 3 .
Fig. 3. Comparison of T 1 , T * 2 , and T RAF F 2 obtained with (a) the proposed MRF sequence and (b) the reference protocols.(c) Excellent correlation, between MRF and the reference protocols, was achieved for T 1 , T * 2 , and T RAF F 2 in the phantom experiments.The black lines represent the best linear fit and the gray shading indicates a 5% deviation from the reference (yellow dashed lines).Coefficients of determination (r 2 ) and best fits are shown for each individual plot.(d) Corresponding Bland-Altman plots.Solid black lines represent mean bias, and dashed black lines represent the limits of agreement (±1.96SD).The inset shows the standard deviation (SD), the reproducibility coefficient (RPC) and the Coefficient of Variation (CV).

Fig. 4 .
Fig. 4. Simultaneous T 1 , T * 2 , and T RAF F 2 maps for all 25 slices acquired in a healthy subject.The in-plane resolution was 0.8 mm × 0.8 mm and the total scan time was 3 minutes and 25 seconds.The T 1 and T * 2 maps depict homogeneous signal in the cartilage.T RAF F 2 maps exhibited a higher level of signal variation in the cartilage and residual susceptibility to image artifacts.

Fig. 5 .
Fig. 5. (a) In vivo T 1 , T * 2 , and T RAF F 2 maps for one slice of MRF and reference scans, for two healthy subjects.(b) Boxplots showing mean femoral and tibial T 1 , T * 2 , and T RAF F 2 across all subjects using the proposed MRF-EPI and the reference methods.In each boxplot, the horizontal line depicts the median, the top and bottom of the box represent the upper and lower quartiles, whereas the whiskers indicate the minimum and maximum values excluding outliers.Additional outliers are indicated by (+).(c) Bland-Altman plot for T 1 , T * 2 , and T RAF F 2 for all subjects.