Imaging and T₂ Relaxometry of Short-T₂ Connective Tissues in the Knee using Ultrashort Echo-Time Double-Echo Steady-State (UTEDESS)

Abstract

Purpose

To develop a radial, double-echo steady-state (DESS) sequence with ultra-short echo-time (UTE) capabilities for T₂ measurement of short-T₂ tissues along with simultaneous rapid, SNR-efficient, and high-isotropic-resolution morphological knee imaging.

Methods

3D radial UTE readouts were incorporated into DESS, termed UTEDESS. Multiple-echo-time UTEDESS was used for performing T₂ relaxometry for short-T₂ tendons, ligaments, and menisci; and for Dixon water-fat imaging. In vivo T₂ estimate repeatability and SNR efficiency for UTEDESS and Cartesian DESS were compared. The impact of coil combination methods on short-T₂ measurements was evaluated via simulations. UTEDESS T₂ measurements were compared to T₂ measurements from Cartesian DESS, multi-echo spin-echo (MESE), and fast spin-echo (FSE).

Results

UTEDESS produced isotropic resolution images with high SNR efficiency in all short-T₂ tissues. Simulations and experiments demonstrated that sum-of-squares coil combinations overestimated short-T₂ measurements. UTEDESS measurements of meniscal T₂ were comparable to DESS, MESE, and FSE measurements while the tendon and ligament measurements were less biased than those from Cartesian DESS. Average UTEDESS T₂ repeatability variation was under 10% in all tissues.

Conclusions

The T₂ measurements of short-T₂ tissues and high-resolution morphological imaging provided by UTEDESS makes it promising for studying the whole knee, both in routine clinical examinations and longitudinal studies.

Introduction

Magnetic resonance imaging (MRI) is a widely used modality in musculoskeletal imaging due its excellent soft-tissue contrast and diagnostic capabilities. A multitude of pathologies can be diagnosed with MRI with its capability of high-resolution imaging with varied contrasts (1,2). However, there are several tissues in the musculoskeletal system, especially in the knee, that exhibit very low T₂ relaxation times (3), which are challenging to image with high signal-to-noise ratio (SNR) and diagnostically viable contrasts. To overcome these challenges, techniques known as ultrashort echo-time (UTE) sequences have been used to achieve echo times of less than 100μs (4,5).

UTE sequences often rely on 3D radial k-space acquisitions as they obviate the need for slice-selection, phase-encoding, or readout dephasing gradients (6). Radial sampling offers isotropic resolution, high sampling efficiency, and excellent motion robustness, which has renewed interest in radial MRI for musculoskeletal applications (6–8). However, despite generating high SNR for short-T₂ musculoskeletal tissues, short echo times (TE) and repetition times (TR) in UTE sequences typically generate images with a proton density (PD) contrast where most tissues appear isointense. Fat saturation can be utilized to enhance tissue contrast in many musculoskeletal applications, however, performing fat saturation in UTE sequences with short scan times, remains challenging (9).

In addition to anatomical detail, MRI can also provide quantitative information regarding the biochemical state of the tissues. One of the most commonly studied quantitative parameters in musculoskeletal MRI is the T₂ relaxation time. In articular cartilage, T₂ has been proposed to be an indicator of the extracellular matrix state, which is sensitive to water and collagen concentrations, as well as tissue anisotropy (2,10,11). Consequently, T₂ mapping can serve as an important marker for therapeutic monitoring of tissue changes during disease progression or recovery (12,13).

Spin-echo (SE) sequences are commonly used to measure T₂ relaxation times. However, even with a multi-echo SE (MESE) acquisition, clinical translation is hampered due to a prohibitively long scan time for this 2D sequence. Faster variations of SE such as fast spin-echo (FSE) have been used commonly, but they have been found to overestimate T₂ values (14,15). The magnetization-prepared angle-modulated partitioned k-space spoiled gradient echo (3D-MAPSS) sequence can measure T₂ but its magnetization preparation can suffer from inherent time-accuracy tradeoffs (16). Fewer preparatory pulses achieve faster scans at the expense of SNR and T₂* decay-induced k-space modulation. In a recent study comparing various T₂ measurement techniques in articular cartilage (17), the 3D quantitative double-echo steady-state sequence (DESS) (18) allowed for rapid scans with a high dynamic range for T₂ measurements that were highly comparable to SE. Despite the myriad of methods available for T₂ quantification, low SNR limits the ability to accurately measure T₂ in short-T₂ tissues (19). Furthermore, phased-array-coil-combination methods such as sensitivity encoding (SENSE) (20) and sum-of-squares (SoS) (21), can limit accuracy of short-T₂ measurements due to varying noise statistics in low-SNR regions (22).

Current research suggests that osteoarthritis (OA) is not just a disease of the cartilage, but rather of the entire joint (23,24). Variations in cartilage T₂ have been shown to distinguish between healthy subjects and patients with varying degrees of OA, even before the onset of gross morphological changes (25). The short-T₂ menisci, tendons, and ligaments, are integral in joint mechanics and abnormalities in these tissues may impact joint degradation and the onset of OA (23,24,26). Therefore, evaluating T₂ variations in these tissues, between healthy subjects and patients with OA, may help provide sensitive markers for early OA diagnoses. While T₂* has also been linked to identifying OA progression and variations in tissue structure (27), it is sensitive to field inhomogeneities (28) and to chemical shift artifacts (29). Since there exists no consensus suggesting that one parameter is more diagnostically advantageous than the other (28), T₂ measurements were pursued in this study as T₂ might be more clinically robust than T₂*.

In this work, we present a new method termed ultrashort echo-time double-echo steady-state (UTEDESS) that provides T₂ relaxometry for short-T₂ tissues such as menisci, tendons, and ligaments, with high accuracy and precision. We demonstrate how in addition to quantitative imaging, UTEDESS can perform rapid, 3D, SNR-efficient, and high-isotropic resolution imaging in a sequence that provides UTE, T₂, and long-T₂-suppressed contrasts, as well as separate water and fat images. Simulations assess the sensitivity of the T₂ measurements to tissue and imaging parameters, as well to as coil combination methods. We validate the UTEDESS method in vivo in the knee by comparing SNR efficiency with Cartesian DESS, and by comparing UTEDESS T₂ measurements with those from Cartesian DESS and conventional spin echo methods.

Methods

UTEDESS Pulse Sequence

The DESS sequence, which incorporates two readouts separated by a spoiler gradient has been described previously (30–32) and modified to provide T₂ and apparent diffusion coefficient (ADC) measurements (18,33,34). The first echo of DESS (S+) generates a T₁/T₂ contrast common in gradient-spoiled sequences. Following the S+ readout, a spoiler gradient is applied and the second echo (S−) is sampled, which is weighted by both, diffusion and T₂. In this study, “DESS” will always refer to Cartesian DESS.

For UTEDESS, we replaced the DESS Cartesian readouts with 3D radial readouts (Fig. 1a). The S+ echo was sampled with 3D radial-out k-space segments while the S− echo was sampled with complementary radial-in k-space segments, without the need for preparatory or rewinder gradients. A multiple-echo-time version of this sequence was implemented by interleaving two constant duration TRs with two different TEs (Fig. 1c). The signals from the first interleaf were termed S₁+ and S₁−, while those from the second were S₂+ and S₂−. The S₁+ and S₂+ echo times were varied between TE+Δ and an original TE (Δ was a user defined offset). Consequently, the durations between the S₂− and S₁− echo-times and the following RF pulses were TE+Δ and TE, respectively. All four echoes traversed the same k-space trajectory, albeit in alternating directions for S+ and S−. Such an interleaved TR acquisition with varying echo-times has previously been demonstrated for simultaneous meniscal DESS T₂ and T₂* measurements (35) and for 3D radial steady-state imaging (36).

Figure 1.

(a) Illustration of the Ultrashort Echo-Time Double-Echo Steady-State (UTEDESS) pulse sequence. The S+ echo samples the free induction decay caused by the hard RF excitation. The S− echo samples the S+ signal from prior TRs that was dephased and then later rephased. (b) The 3D radial sampling strategy for UTEDESS where the dotted lines represent the trajectory of the radial spokes. The S+ echo is sampled radially outward – from the center of k-space to the radial endpoints (dots on the k-space sphere) while the S− echo is sampled radially inward – from the endpoints to the center of the k-space. (c) A multiple-echo-time version of UTEDESS has gradients in successive TRs shifted by a finite duration while maintaining a constant TR and symmetry across the second RF excitation. This creates a pair of S+ and S− images each with different TEs that can be used for Dixon water-fat separation. The time between S₂− and the subsequent RF pulse can be used for necessary sequence updates, so the Dixon scheme comes at a cost of little sequence dead time.

A non-selective RF pulse and gradient ramp sampling provided a 48μs TE for S₂+. Spoiler gradients (23.5 mT¹m⁻¹ms¹ area) were applied on all gradient directions to prevent steady-state banding, minimally sensitizing the S− echoes to diffusion. Water and fat images for both, the S+ and S− echoes, were generated using a flexible-echo-time Dixon water-fat separation method (37,38). The short-T₂ tissue anatomy was highlighted with a long-T₂ subtraction where twice the water S− image was subtracted from the water S+ image (39). Image reconstruction was performed by gridding the 3D radial data onto Cartesian grids (40). A Voronoi cell-based density compensation (41) was used to correct for ramp sampling. A gradient timing delay correction and a B₀ eddy current correction (42) was performed. For a given matrix size NxNxN, πN² radial spokes whose endpoints uniformly sample the surface of a sphere (Fig. 1b) were necessary for full Nyquist sampling (compared with N² for Cartesian phase encodes). However, radial undersampling does not lead to severe image degradation (43) and we utilized 7x angular undersampling in UTEDESS.

T₂ Mapping

To date, two primary methods have been used in conjunction with DESS to estimate T₂. One method analytically approximates T₂ (33) using the sequence TE and TR, and S−/S+ ratio in Eq. 1, while ignoring flip angle (B₁), T₁ recovery, and diffusion. The second method estimates T₂ by comparing acquired signals to a library of simulated DESS signals with varying T₁, T₂, and ADC values (18,44). In this study, we measured T₂ using an analytical relationship between signals S+ and S− in Eq. 2 (45), based on extended phase graphs (EPG) (46) and UTEDESS timings (Eq. 3).

$$\frac{S-}{S+}=e^{-\frac{2(TR-{TE}_{S+})}{T_2}}$$ [1]

$$\frac{S-}{S+}=e^{-\frac{2(TR-{TE}_{S+})}{T_2}-\left(TR-\frac{\tau }{3}\right)\mathrm{\Delta }{k}^{2}D}{sin}^2\left(\frac{\alpha }{2}\right)\left(\frac{1+e^{(-\frac{TR}{T_1}-TR\mathrm{\Delta }{k}^{2}D)}}{1-cos\alpha e^{-(\frac{TR}{T_1}-TR\mathrm{\Delta }{k}^{2}D)}}\right)$$ [2]

$${TE}_{S-}=2(TR)-{TE}_{S+}$$ [3]

In Eq. 2, α is the flip angle and D is the tissue diffusivity. The spoiler induced dephasing is denoted by Δk=γGτ, where G is the spoiler amplitude, τ is the spoiler duration, and γ is the gyromagnetic ratio. Assuming a low diffusion weighting (Δk≈0), a large T₁ relaxation time (T₁ approaches ∞), and a flip angle of 90°, Eqs. 1 and 2 are equivalent. However, for low flip angles and moderate T₁ values, Eq. 2 greatly improves T₂ accuracy (45). Acquiring separate scans with varying spoiler gradient areas and flip angles can be used for measuring T_2, T₁ and ADC simultaneously. However, non-contrast T₁ is not a commonly used musculoskeletal biomarker while accurate ADC measurements require a minimum of two scans (18,34). For this study, only a single rapid acquisition for T₂ measurement was pursued.

Simulations

The sensitivity of the T₂ fitting procedure to T₁, flip angle, and diffusivity in tendons, menisci, and cartilage (approximate T₂ = 5ms, 12ms, and 35ms, respectively) was evaluated. T₂ for these tissues was simulated with Eq. 2 for T₁ ranging from 0.5s–1.5s, flip angle ranging from 6°–20°, and diffusivity ranging from 0.5×10⁻⁹m²s⁻¹ – 2×10⁻⁹m²s⁻¹. The error between the original T₂ and the T₂ calculated in this study with Eq. 2 was measured, where all diffusivities were 1.25×10⁻⁹m²s⁻¹, all flip angles were 10°, tendon T₁ was 1151ms, meniscus T₁ was 998ms, and cartilage T₁ was 1167ms, as described below.

T₂ measurement variations induced by SoS and SENSE coil combination were explored with Monte Carlo simulations. Tissues with T₂ = 12ms and 30ms, representing meniscus and cartilage respectively, were simulated with spatially varying coil sensitivities and with DESS scan parameters listed in Table 1. Complex Gaussian noise was added to each coil image and coil combination was performed with SoS and R=1 SENSE, commonly referred to as the ‘Roemer’ combination (21), to compare voxel SNR and the estimated T₂ for the two methods.

Table 1.

Scan protocol for all volunteers in this study. The UTEDESS imaging volume was (16cm)³ while slice thickness for all other sequences were 3mm (except B₁ map which had 6mm thick slices). The legs of volunteers were positioned neutrally, with their right knee being scanned. UTEDESS and DESS were run twice each in order to assess T₂ map reproducibility and tissue SNR. UTEDESS was angularly undersampled by a factor of 7x, to include 45,957 radial spokes.

Sequence	TE (ms)	TR (ms)	Matrix	# Slices	Flip Angle	NEX	Bandwidth (kHz)	Scan Time
UTEDESS	0.05, 1, 10a, 11a	6	320×320	320	10°	1	± 250	8:23×2
DESS	6, 34a	20	320×320	40	30°	1	± 31	4:15×2
2D - MESE	10,20,30	2000	128×128	1	90°	0.5	± 31	2:36
2D - FSE	7,14,20,27,33	2000	128×128	1	90°	1	± 31	4:20
B1 Map	14	30	128×128	20	30°	1	± 15	2:48
PDw FS-FSE	30	3000	384×320	40	111°	2	± 31	3:18
MP-RAGEb	3.4	10	256×256	40	4°	1	± 31	3:28×3

^aUTEDESS and DESS TE_S− are longer than the TR since the S− refocusing occurs at 2*TR-TE_S+.

^bMP-RAGE sequences were run on two volunteers with inversion times of 150ms, 1200ms, and 4000ms, in order to assess T₁ relaxation times for the short-T₂ components.

Volunteer Study

All experiments were performed on a Discovery MR750 3.0T MRI scanner (GE Healthcare, Waukesha, WI) with 50 mT¹m⁻¹ maximum gradient amplitudes, 200mT¹m⁻¹ms⁻¹ maximum slew rates, and a 16-channel receive-only knee coil (Neocoil, Pewaukee, WI). 11 healthy volunteers (five male, age 29±4; six female, age 28±4) were scanned with institutional review board approval and informed consent with a protocol, listed in Table 1, for maximizing short-T₂ signals. DESS and multiple-echo-time UTEDESS were both repeated twice on all volunteers to assess SNR and repeatability of T₂ measurements. Single-slice sagittal 2D MESE and FSE were used to measure T₂ in the medial meniscal using a monoexponential fit (47), for comparisons with UTEDESS T₂ measurements. Low-resolution readouts were used with SE to minimize the TE and echo spacing. However, they could not achieve sufficiently low TEs to accurately characterize tendon and ligament T₂ values so they were used for meniscal T₂ comparisons only. All T₂ measurements were performed with R=1 SENSE coil combinations where coil sensitivity maps were generated by low-pass-filtering the k-space data from the lowest TE scans from every sequence (48).

The meniscus, patellar tendon (PT), and quadriceps tendon (QT) were manually segmented in all scans along with regions of interest of the posterior cruciate ligament (PCL) anterior cruciate ligament (ACL) at their tibial insertion sites. The meniscus was further sub-divided into six sections of the posterior, body, and anterior horns of the medial and lateral meniscus, to assess regional variation. A PD-weighted fat-saturated FSE (PDw FS FSE) sequence was used to aid in tissue segmentation.

Signal-To-Noise Measurement

SNR was measured in the above regions and the patellar cartilage with DESS and UTEDESS, where UTEDESS slices were downsampled to 3mm. DESS and UTEDESS noise was measured by the “difference method” (7,49). SNR efficiency (η) was calculated for UTEDESS and DESS by normalizing the single-slice SNR with voxel volume and square root of scan time. The tradeoff between S+ SNR and S− contrast-to-noise (CNR) in UTEDESS water was explored by scanning one volunteer with UTEDESS with flip angles of 10°, 15°, 20°, and 30°. Here, SNR was measured in all segmented tissues along with synovial fluid and fat, while CNR was measured between cartilage-fluid, cartilage-meniscus, cartilage-muscle, and cartilage-fat.

T₂ Mapping Data Analysis

UTEDESS and DESS scans were downsampled to an in-plane matrix size of 256×256. Five UTEDESS slices were combined using a complex sum for an effective slice thickness of 3.13mm, comparable to the 3mm slice thicknesses of DESS, MESE, and FSE. The DESS and UTEDESS S−/S+ ratios and Eq. 2 were used to generate pixel-wise T₂ maps in all manually segmented tissues. The multiple-echo-time UTEDESS TE’s were: TE_S1+=1.10ms, TE_S2+=0.05ms, TE_S1−=10.90ms, and TE_S2−=9.80ms. To measure T₂ in the meniscus, the UTEDESS S₂−/S₁+ echo pairing was used as it produces a higher S− SNR with a lower S₂− TE. For the tendons and ligaments, the S₁−/S₂+ echo pair was used to maximize the S₂+ signal. For illustrative purposes, the difference in T₂ values between a SoS and SENSE approach was performed in a DESS scan. An intra-subject coefficient of variation (CV) was measured (by dividing the T₂ standard deviation across the two scans by its mean) to assess scan-to-scan intra-subject variability for both, DESS and UTEDESS.

A magnetization-prepared rapid gradient-echo imaging (MP-RAGE) sequence (50,51) was used to measure the T₁ values of the segmented tissues, as well as the gastrocnemius muscle and the patellar cartilage for measurement validation, in two volunteers. A Bloch-Siegert B₁ map (52) was acquired to account for possible flip angle variations. T₂ was calculated using the measured T₁ and an approximate diffusivity of 1.25×10⁻⁹m²s⁻¹, for both, the nominal flip angle and the measured flip angle. Image reconstruction, signal simulation, and parameter fitting was performed using MATLAB (MathWorks, Natick, MA).

Results

Simulations

Simulations of the UTEDESS signal showed that the fitted short-T₂ has minimal sensitivity to diffusivity, low sensitivity to T_1, and moderate sensitivity to B₁ (Fig. 2). The error in T₂ estimation induced due to inaccurate diffusivity, T₁, and B₁, increases as the T₂ of the tissue increases. Fig. 3 indicates that tissues with a longer-T₂ (30ms) show minimal differences between T₂ measurements obtained from SENSE and SoS, while SoS substantially overestimates the true T₂ for low-T₂ (12ms) tissues. This is expected, since SoS coil combinations have known noise biases that are problematic for T₂ fitting when the signal approaches the noise floor (21).

Figure 2.

Sensitivity of the UTEDESS T₂ measurements in tissues with approximate T₂ of tendons (5ms), menisci (12ms), and cartilage (35ms), to parameters of diffusivity, T₁, and B₁. Reference T₂ values were calculated with all diffusivities = 1.25×10⁻⁹m²s⁻¹, all flip angles = 10°, tendon T₁ = 1151ms, meniscus T₁ = 998ms, and cartilage T₁ = 1167ms as described below. Simulated T₂ values were calculated with varying diffusivity, T₁, and B1 values, and the error between the reference and simulated values was measured. (a) The sensitivity profile shows minimal sensitivity to ADC, partly because a very low diffusion weighting was applied in the sequence. (b) The short-T₂ tissues show low sensitivity to T₁, where ~40% underestimations in T₁ only lead to ~6% variations in T₂. However, the sensitivity increases as the T₂ of the tissues increases. (c) T₂ measurements show the highest sensitivity to B₁, though the sensitivity decreases significantly with increasing flip angles.

Figure 3.

Monte-Carlo simulations with spatially varying coil sensitivities and varying levels of complex Gaussian noise were performed while using R=1 SENSE and SoS for coil combination and generating T₂ maps. (a) For low SNR values typically seen with meniscus (T₂ about 12ms), SoS significantly overestimates the estimated T₂. (b) For higher SNR typically seen with cartilage (T₂ about 30ms), SENSE and SoS have comparable T₂ estimates.

Volunteer Study

Consistent with simulations, the SoS short-T₂ overestimation in the meniscus, compared to SENSE, can be seen in-vivo with the DESS scan in Fig. 4. The T₂ variation seen between SoS and R=1 SENSE reconstructions is minor in cartilage, but substantial in the meniscus. This result demonstrates the importance of the R=1 SENSE combination, which we used for all subsequent T₂ mapping.

Figure 4.

In vivo T₂ differences between R=1 SENSE and SoS coil combinations with DESS. T₂ measurements in the femoral cartilage (a), lateral anterior (LA) meniscal horn (b), and in the lateral posterior (LP) meniscal horn (c) generated with a SENSE coil combination show consistently lower T₂ values in the meniscus compared to a SoS combination for the same regions of interest (d–f). (g–i) Unlike the T₂ in the meniscus, the T₂ measurements in cartilage are similar for both coil combination methods.

UTEDESS produced T₁-weighted images (S+), higher T₂-weighted images (S−), water-fat separated S+ and S− images, long-T₂-suppressed images, and T₂ maps in tendons, ligaments, and menisci in arbitrary scan planes (Fig. 5). The UTEDESS images had higher signal in the short-T₂ tissues and a comparable signal in the other tissues (Fig. 6 and 7). The higher signal in the tendons and ligaments (arrows, Fig. 6a–b) allowed for the creation of T₂ maps, which was not possible with DESS due to lower SNR (Fig. 6d and Table 2). The UTEDESS images were free of any obvious undersampling artifacts or image distortions that could arise from radial undersampling and gradient imperfections. The spatial resolution and SNR of UTEDESS were adequate to visualize some fascicular structure of the tendons in the S+ water image.

Figure 5.

Images from selected volunteers showing the contrasts that can be generated with UTEDESS. (a) The S+ UTE images are T₁-weighted and have high signal intensity from short-T₂ tissues such as the tendons (solid arrow) and the menisci (dashed arrow). (b) The S− echo is more highly T₂ weighted where the short-T₂ tissues have lower signal. (c) The water-fat separation method provides spectral separation with two echo times. There is clear separation of the short-T₂ tissues of the tendons (solid arrow) and the meniscus (dashed arrow) since the longest echo time used is only 1.1ms. (d) Implementing a Dixon-based water-fat separation, as opposed to using a fat-saturation RF pulse, generates a fat-only image. (e) Performing a weighted subtraction of the water S− image from the water S+ image helps suppress the longer-T₂ signals. This provides a similar contrast to the high SNR water S+ image but the weightings can be modified based on the short-T₂ anatomy that needs to be visualized. (f) In a different subject with fluid in the knee, the water S− image highlights the bright signal around the patellar cartilage (solid arrow) and the anterior horn of the lateral meniscus (dashed arrow). The water S− image has a T₂ weighting and has a lower dynamic range relative to the non-water-fat separated S− image. (g–i) Due to the isotropic resolution of the sequence, the scan planes can be formatted in arbitrary directions to create and view T₂ maps.

Figure 6.

Comparison of images generated with UTEDESS and DESS. Column 1 shows UTEDESS images, columns 2 and 3 show the UTEDESS retrospective water and fat decomposed images respectively, and column 4 shows the DESS water only acquisition. (a–d) In the S+ images, the quadriceps tendons (solid arrow), patellar tendon (dash-dot arrow), the anterior cruciate ligament (dotted arrow), and the posterior cruciate ligament (dashed arrow) show excellent signal with minimal blurring in the UTEDESS images. There is minimal signal in these tissues with DESS. (e–h) In the S− echoes for both sequences, there is bright fluid signal present posterior to the quadriceps tendon (dashed arrow), in the inferior section of the patellar cartilage (dotted arrow), around the tibial cartilage (dash dot arrow), and the anterior section of the posterior cruciate ligament (dashed arrow). This bright fluid signal can be better visualized with DESS.

Figure 7.

Comparison of signal-to-noise ratios (SNR) and SNR efficiencies (η) normalized to the scan time and voxel volume. The SNR values (a) show that UTEDESS outperforms DESS in all short-T₂ tissues, except in the S+ echo for meniscus. The S− SNR, however, is higher for all short-T₂ tissues with UTEDESS. This higher SNR makes accurate T₂ fitting possible. (b) The higher SNR efficiencies with UTEDESS show that when normalized for the resolution and scan time, UTEDESS provides a higher SNR in all short-T₂ tissues per unit time, compared to DESS, while maintaining a comparable SNR efficiency in the longer-T₂ tissues such as cartilage.

Table 2.

T₂ measurements, intra-subject repeatability CVs, and the SNR values for both echoes with UTEDESS and DESS. The reported T₂ values per sequence are averages of the two repeated scans across all volunteers. There is good agreement between T₂ measurements in the meniscus with DESS and UTEDESS, however, DESS overestimates the T₂ of tendons and ligaments, likely due to low-SNR in both echoes. The UTEDESS T₂ measurements show high repeatability with CV values lower than 10%.

	LAM	LBM	LPM	MAM	MBM	MPM	PT	QT	ACL	PCL
UTEDESS
T2 (ms)	10.6±1.5	9.7±2.1	9.7±1.5	11.4±1.6	10.9±1.4	10.9±1.8	6.0±1.1	5.2±0.5	9.3±0.8	5.8±0.6
T2 CV (%)	8.9	8.6	5.8	6.7	6.1	8.0	4.9	6.5	5.6	5.6
S+ SNR	33.1±6.4	35.3±6.1	34.2±5.4	37.8±3.6	35.2±3.9	34.9±4.6	47.0±3.2	45.2±5.8	28.2±3.9	30.2±3.3
S− SNR	19.9±2.1	19.3±3.1	18.8±1.8	20.4±1.7	19.8±1.9	19.6±2.0	20.2±1.9	17.7±2.8	20.1±2.5	19.0±2.3
DESS
T2 (ms)	12.1±1.4	12.1±1.3	11.5±1.3	11.6±1.1	12.1±1.3	11.8±1.7	21.0±6.0	12.9±1.4	33.4±6.6	22.3±8.0
T2 CV (%)	5.7	4.1	4.5	3.1	3.9	4.1	18.6	4.1	6.0	9.2
S+ SNR	39.6±6.6	41.5±6.6	38.7±5.8	43.3±8.7	40.1±5.1	37.9±4.5	15.1±3.3	15.1±2.9	38.7±5.0	19.2±5.5
S− SNR	7.3±1.5	6.4±1.0	7.1±1.8	6.3±1.7	6.3±1.9	6.2±2.0	9.7±1.8	7.0±2.8	13.1±4.2	7.0±1.3

Abbreviations: LAM – lateral anterior meniscus, LBM – lateral meniscus body, LPM – lateral posterior meniscus, MAM – medial anterior meniscus, MBM – medial meniscus body, MPM – medial posterior meniscus, PT – patellar tendon, QT – quadriceps tendon, ACL – anterior cruciate ligament, PCL – posterior cruciate ligament.

DESS had additional cartilage-fluid contrast in the S− echoes than UTEDESS, likely due to a longer TR and a higher flip angle, which was most apparent in the tibial cartilage (dash-dot arrow, Fig. 6f, 6h). However, the UTEDESS water-fat separation (Fig. 6e–f) helped generate additional fluid contrast. The water S+ images in Fig. 8 indicate that increasing the flip angles decreases meniscus (solid arrows) SNR, a trend consistent with other soft-tissues as well (Fig. 8i). However, in the water S− image (Fig. 8j), increasing the flip angles increases the cartilage-fluid (dotted arrows) CNR due to higher cartilage signal attenuation relative to the fluid. Overall, higher flip angles increase CNR at a cost of decreased SNR.

Figure 8.

SNR was measured in the patellar cartilage, the medial meniscus, the patellar tendon, the posterior cruciate ligament, the gastrocnemius muscle, femoral bone marrow (“Fat”), and synovial fluid. (a–d) Increasing the flip angle results in lower signal in the meniscus (solid arrows) in the water S+ images. (e–h) While decreasing overall signal from soft-tissues, higher flip angles also increase cartilage-fluid contrast (dotted arrows) in the water S− images. The increasing flip angle also results in better contrast between the synovial fluid and membrane. (i) By comparing the normalized SNR as a function of flip angle, the SNR for all soft-tissues decreases with an increasing flip angle while the fat and fluid SNR stays relatively constant. (j) An increasing flip-angle increases the cartilage-fluid CNR due to constant fluid signal and attenuated cartilage signals. At a flip of angle of 30°, there is considerably lower meniscal and muscle signal, which reduces the CNR between those tissues and cartilage.

The MP-RAGE-based T₁ estimates were: meniscus 998±4ms, tendons 1151±37ms, ligaments 1034±25ms, muscle 1392±41ms, and cartilage 1162±13ms. These values were used in the T₂ measurements with DESS and UTEDESS. T₂ measurements, CV, and SNR values, for all segmented tissues are reported in Table 2. T₂ measurements with MESE and FSE in the medial anterior meniscus were 12.4±1.7ms and 11.2±1.4m respectively, while those in the medial posterior meniscus were 12.0±1.4ms and 11.8±1.1m respectively. These showed excellent agreement with the T₂ measurements in the medial meniscus calculated with UTEDESS with DESS. The T₂ measurements from UTEDESS and DESS agreed well in the meniscus, however, for shorter-T₂ tissues such as the tendons and ligaments, DESS overestimated the T₂. The intra-subject repeatability CV in all tissues with UTEDESS was consistently under 10% for all tissues.

Discussion

In this study UTEDESS provided simultaneous quantitative and morphological imaging of the human knee, focusing primarily on short-T₂ tissues. The parameters for UTEDESS were chosen to provide accurate and repeatable T₂ measurements in menisci, tendons and ligaments. The UTEDESS T₂ measures in the meniscus showed good agreement with conventional T₂ mapping sequences such as MESE, FSE and DESS. Moreover, the 3D radial acquisition generated images with 3D isotropic resolution that allowed reformats in arbitrary scan planes. The multiple-echo-time version of the sequence provided high SNR efficiency with different contrasts such as T₁-weightings in the UTE S+ echoes, T₂-weightings in the S− echoes, water-fat separation in both S+ and S− echoes, and long-T₂ suppression.

Signal-to-Noise Ratios

UTEDESS was able to image the short-T₂ tissues with high-isotropic resolution and high SNR. DESS had a higher cartilage SNR and slightly higher meniscal SNR in the S+ echo, but this was likely due to the longer TR (20ms vs. 6ms) offering additional T₁ recovery and a longer data acquisition window (5.12ms vs. 0.72ms). EPG based simulations showed that tissues with a T₂ of 10ms had a 10–20% higher signal with DESS than UTEDESS for T₁ ranging between 800ms – 1200ms. The data acquisition duration variation should account for 2.7x higher SNR in DESS, but the short TE and TR in UTEDESS increases signal, resulting in comparable overall SNR between the sequences. UTEDESS did have higher S+ SNR in the tendons and ligaments compared to DESS (Fig. 7a) due to shorter TEs.

While the DESS S+ SNR was higher in the meniscus, UTEDESS provided higher S− SNR in the meniscus, tendons, and ligaments (Fig. 7a). Since the T₂ measurement method described in Eq. 2 relies on a two point ratio, the T₂ values are susceptible to the S− SNR, where noise causes T₂ overestimations (19). Thus, despite DESS offering marginally higher S+ meniscal SNR, UTEDESS provides increased robustness for T₂ mapping due to a higher S− SNR.

When the SNR from both sequences is corrected for scan time and voxel volume, UTEDESS outperforms DESS in terms of SNR efficiency (η) for all short-T₂ tissues in both echoes. To achieve a comparable SNR in the short-T₂ tissues, DESS would require 4–16 averages, which is not pragmatic. It should be noted that UTEDESS samples only 70% of k-space that is sampled with DESS, due to the spherical k-space filtering induced by the 3D radial trajectories.

T₂ Measurement

The UTEDESS images had high SNR and thereby could be used for accurate measurements of T₂ in the meniscus, tendons, and ligaments. The SoS method, overestimated the T_2, compared to SENSE, (Fig. 3) due to coil-combination noise biases. Since the T₂ fitting was performed on magnitude images, regions with a low signal had an increased noise bias where the increased signal was mistakenly attributed to a higher T₂. However, since SENSE uses a linear combination of the coil images, the noise remains zero-mean (20,21) after the coil combination, which reduces the noise bias. Compared to the lower-T₂ and SNR meniscus, coil combination minimally affects T₂ values in high-SNR tissues, such as the cartilage, since noise minimally biases the higher-mean signal (Fig. 4).

Although different statistical models accounting for low-SNR have been suggested for use in quantitative imaging, they rely on an accurate measure of the true noise statistics (19,53,54). Background ROIs cannot accurately depict true noise statistics in scans with phased array coils and parallel imaging (49). This makes low-SNR noise corrections challenging to implement clinically within reasonable scan times. An R=1 SENSE coil-combination does not require any additional information and can be performed retrospectively to measure short-T₂ relaxation times without coil combination noise biases. While this study does not perform a comprehensive noise analysis, it tries to address the importance of coil-combination for measuring short-T₂ values with radial undersampling.

T₂ Validation

The accuracy of the T₂ maps was verified by comparing the meniscus T₂ measurements to those obtained with MESE, FSE, and DESS. With SE sequences, the lowest TE should be lower than the T₂ being probed, which is challenging because the T₂ of the meniscus is around 10ms, while that of tendons and ligaments is even lower. Thus, the meniscus T₂ was used as a basis of comparison between the sequences.

The accuracy of the T₂ measurements could also be attributed to the modified exponential fit proposed in Eq. 2. Without accounting for the measured T₁ and flip angle values, the lower signal intensity in the S− echo can be mistakenly attributed to shorter T₂ which can underestimate T₂ by 10–20%. The simulations in Fig. 2 show that the short-T₂ tissues are somewhat insensitive to variations in the diffusivity, T₁, and B₁. The sensitivity to these parameters increases with increasing tissue T₂, however, with higher-T₂ tissues, it becomes easier to characterize the T₁ and diffusivity using other methods. The MP-RAGE T₁ values in muscle and cartilage were consistent with literature (55,56), and they provided an adequate and rapid T₁ estimate for short-T₂ tissues. However, depending on the tradeoff between scan time and T₁ accuracy needed, saturation recovery or variable-flip-angle UTE methods could be employed (57,58).

In the tendons and ligaments, DESS overestimated the T₂ values as compared to UTEDESS (Table 2). The tendons and ligaments in the DESS S− echo had low SNR and approached the noise floor, which may overestimate T₂. FSE can overestimate T₂ by 10% due to stimulated echo effects (16,17). Excluding the first echo from the multi-echo fit may mitigate this problem (47) but doing so significantly increases the lowest TE. Similarly, with MESE, a lowest TE that is shorter than 10ms would have been preferable. The lack of short-TEs made SE sequences insensitive to the fast-decaying short-T₂ components. Similarly, while UTEDESS had a UTE echo, it relied on a two-point fit with a TE_S− of over 10ms, which may have failed to capture the fast-decaying components. Thus, the similarity of the meniscal T₂ measurements with all sequences may be explained by the insensitivity to these fast-decaying short-T₂ components.

T₂ maps were generated with both, a nominal B₁, and with a measured B₁. The B₁ variations were low, and due to limited sensitivity of the T₂ estimates to B₁, there were minimal T₂ differences between the nominal and measured flip angles. Consequently, only the T₂ measurements with the nominal flip angles were reported. The only parameter unavailable in the fitting for T₂ was the diffusivity of the tissues. However, as Fig. 2 shows, UTEDESS and DESS sequences with low spoiling are very minimally sensitive to diffusion where even large variations in ADC only minimally affect the T₂. Thus, an estimate for short-T₂ ADC values, based on cartilage ADC of 1.25×10⁻⁹m²s⁻¹ (18,34), was deemed appropriate. The scan-to-scan repeatability CV was consistently lower than 10% in all tissues and was comparable to the DESS CV values, which showed excellent precision.

Tissues in the knee are sensitive to magic angle effects, wherein tissue T₂ is dependent upon its orientation with B₀ (59). The tibial insertion site of the ACL is mostly aligned with the magic angle while the insertion site of the PCL is mostly orthogonal to B₀ (60). Choosing these insertion sites also minimizes the impact of intrasubstance fluid on ACL T₂ measurements. Thus, magic angle could be one of the major causes of higher T₂ values in the ACL than in the PCL. Such effects with DESS T₂ have been reported earlier (61), which should be similar for UTEDESS and SE methods since the magic angle affects the actual tissue T₂. More importantly, however, one major goal for measuring T₂ and other quantitative parameters is to assess longitudinal changes. Since magic-angle-induced changes in T₂ would generally be consistent over time, T₂ is still efficacious in assessing temporal changes.

Overall, with tissue-specific estimates of the T₁ and diffusivity, the T₂ of short-T₂ tissues can be accurately measured with UTEDESS with good repeatability. As this is one of the few in-vivo studies exploring the T₂ values in short-T₂ tissues, normal spatial and longitudinal variations are not yet well known. However, our study indicates that UTEDESS may be a promising tool to study these variations, and to help evaluate new biomarkers of OA progression and response to therapy.

Other Meniscal T₂ Studies

The meniscal T₂ relaxation times for healthy volunteers measured with UTEDESS (on the order of 10–12ms) agree well with the reported measurements in prior studies (62–67). Most of these studies include the use of the 3D-MAPSS sequence. Compared to 3D-MAPSS, UTEDESS provided the ability to create T₂ maps in arbitrary scan planes, at a higher resolution, in lower scan times, and for not only the meniscus, but tendons and ligaments also.

Potential Clinical Utility

Most research applications of DESS, including its usage in the Osteoarthritis Initiative, rely on fat suppression to create contrast for cartilage segmentation and for diagnosing osteophytes, bone cysts, and bone attrition (68). A Dixon DESS sequence, with a shorter TE and TR, could have been used, but the benefit of a slightly shorter TR would have been mitigated due to the inefficiency in sampling four long Cartesian echoes. A sinc or hard pulse excitation without fat saturation could also have been used, but this would lower the image contrast. Thus, DESS with a spectral-spatial RF pulse was utilized in this study as a reference standard and also to evaluate if DESS could be used to study other tissues in the knee joint. For either implementation of DESS, however, UTEDESS offered a much higher sampling efficiency due to the lack of slice-select, phase-encode and rewinder gradients.

The two-point Dixon technique achieved adequate water-fat separation with UTEDESS. With only two echoes necessary, the TEs for all the input images in multiple-echo-time UTEDESS could also be kept low to achieve water-fat separation of even the short-T₂ tissues (Fig. 5c). The longest UTEDESS S+ TE of 1.1ms used in Dixon imaging at 3T, coupled with 0.7ms readouts, provided high signal and minimal blurring in the water image even in the tendons and ligaments. Such short TEs could not be achieved with water selective RF pulses since they are generally several milliseconds long.

The Dixon water-fat separation increased fluid contrast in the water S− images (Fig. 5f), synonymous to contrast provided by T₂-weighted sequences. The bright fluid signal could also be used for detecting inflammation, meniscal and ligament tears, synovitis, cartilage defects, bone marrow lesions, and cysts (69). DESS did achieve a better fluid contrast in the S− echo (Fig. 6f,6h), but this was due to a longer TR and higher flip angle. As can be seen in Fig. 8, there is an inherent tradeoff between short-T₂ tissue signal and bright fluid contrast, primarily as a function of flip angle. As the flip angle is increased, there is reduced stored longitudinal magnetization, which reduces the overall signal (Fig. 8i). Increasing the flip angle from 10° to 30° reduces the meniscus signal in the water S+ images (Fig. 8a–d) but creates cartilage-fluid contrast in the water S− images (Fig. 8e–h) since fluids experience less signal reduction due to their longer T₂. At higher flip angles (Fig. 8f–h), there also appears to be increased contrast between the synovial membrane and fluid. Simulations in Fig. 2c also suggest that higher flip angles may result in lower T₂ sensitivity to B₁. The low flip angle and short-TR in UTEDESS for this study was chosen to maximize short-T₂ tissue signal for accurate T₂ measurements. Future studies could be performed to investigate such scan parameter tradeoffs for quantitative and morphological imaging.

UTEDESS could assist in diagnostic imaging with its 3D isotropic resolution, which allows image reformatting with arbitrary slice thicknesses and orientations. The radial acquisition has also been shown to provide benign appearance of motion artifacts and excellent insensitivity to patient motion due to signal averaging at the center of k-space (70).

Future Work and Limitations

The pulse sequence design and parameter selection for UTEDESS was performed to image short-T₂ tissues with high SNR for accurate T₂ mapping. A high sampling bandwidth of ±250kHz was used to prevent blurring of short-T₂ tissues. This, however, led to minimal cartilage signal decay from the S+ to the S− echoes due to a TR of only 6ms. Decreased bandwidth readouts with a longer TR could potentially be used to trade-off between SNR of short-T₂ tissues and cartilage signal decay between S+ and S−, allow for cartilage T₂ measurements.

Regarding limitations, the volunteer pool for this study comprised only of 11 healthy young-adults, rather than a more typical OA population. Moreover, future studies could further investigate the role of magic angle and biexponential decay models on T₂ measurements, as discussed previously.

Conclusion

In this study, we developed UTEDESS, a new pulse sequence tailored to knee imaging, which provides short-T₂ relaxometry and morphological imaging. UTEDESS imaging was performed to maximize short-T₂ signal for measuring the T₂ relaxation times of menisci, tendons, and ligaments, with good repeatability, by using a noise bias-free R=1 SENSE coil-combination method. The UTEDESS T₂ measurements in the menisci showed excellent agreement with T₂ measurements obtained with DESS, SE sequences, and other studies from literature. This quantitative imaging was performed simultaneously with rapid and high-isotropic-resolution morphological imaging with multiple contrasts and high SNR efficiency. These results suggest that UTEDESS is a promising tool and an efficient technique for quantitative and morphological imaging of whole knee, suitable for routine clinical imaging as well as for longitudinal clinical studies.