A Simple Analytic Method for Estimating T2 in the Knee from DESS

Abstract

Purpose

To introduce a simple analytical formula for estimating T₂ from a single Double-Echo in Steady-State (DESS) scan.

Methods

Extended Phase Graph (EPG) modeling was used to develop a straightforward linear approximation of the relationship between the two DESS signals, enabling accurate T₂ estimation from one DESS scan. Simulations were performed to demonstrate cancellation of different echo pathways to validate this simple model. The resulting analytic formula was compared to previous methods for T₂ estimation using DESS and fast spin-echo scans in agar phantoms and knee cartilage in three volunteers and three patients. The DESS approach allows 3D (256×256×44) T₂-mapping with fat suppression in scan times of 3–4 minutes.

Results

The simulations demonstrated that the model approximates the true signal very well. If the T₁ is within 20% of the assumed T₁, the T₂ estimation error was shown to be less than 5% for typical scans. The inherent residual error in the model was demonstrated to be small both due to signal decay and opposing signal contributions. The estimated T₂ from the linear relationship agrees well with reference scans, both for the phantoms and in vivo. The method resulted in less underestimation of T₂ than previous single-scan approaches, with processing times 60 times faster than using a numerical fit.

Conclusion

A simplified relationship between the two DESS signals allows for rapid 3D T₂ quantification with DESS that is accurate, yet also simple. The simplicity of the method allows for immediate T₂ estimation in cartilage during the MRI examination.

1. Introduction

The Double-Echo Steady-State (DESS) sequence offers distortion-free, SNR-efficient 3D-imaging with good contrast flexibility [1–3]. DESS is a gradient-spoiled steady-state sequence, collecting two echoes per repetition (TR), with the spoiler gradient separating the two echoes. The two signals, often labeled S₁ (before the spoiler) and S₂ (after the spoiler), have different contrast. The S₁ signal is often said to have T₁/T₂ weighting and S₂ to have a mixture of T₂ and diffusion weighting, but in reality both the contrasts are quite complicated functions of the extrinsic scan parameters (TR, TE, flip angle α, gradient amplitude G and gradient duration τ) as well as the intrinsic tissue parameters (T₁, T₂, and diffusivity). While DESS has been used for various anatomies such as the breast [4,5] and the prostate [6,7], it has found particularly widespread use for musculoskeletal imaging, where for example it has been used in the Osteoarthritis Initiative (OAI) to determine Osteoarthritis (OA) progression [8].

To improve OA detection, DESS has been used for quantitative imaging of T₂ and apparent diffusion coefficient (ADC) [9–12]. These have been found to correlate with collagen network organization and proteoglycan concentration, respectively [13,14]. A number of signal models have been developed to quantitatively describe both of the DESS signals and their dependency on tissue parameters. These include both closed-form and non-closed-form analytical methods as well as numerical methods [15–20].

Some methods model the DESS signals completely and can be considered fully accurate for a single T₁, T₂, and diffusivity in a voxel [16,17]. However, owing to their complexity, using these models for parameter estimation typically requires solving them with a numerical search over a predetermined solution set. This inhibits immediate, automatic creation of parameter maps right at the scanner, which would be of great value for clinical scans. Furthermore, the method in ref. [16], which describes the magnetization with a continued fraction, does not allow for pulsed gradients. For these reasons, approximate models are often used to estimate the desired parameter in spite of known errors. For example, in ref. [9], the following model, originally suggested in ref. [1], was used to estimate T₂ from a single DESS scan:

$$\frac{S_2}{S_1}=e^{-\frac{2(TR-TE)}{T2}}$$ (1)

This relationship essentially describes a spin echo relationship between S₂ and S₁ in consecutive TRs. Although this approach may offer useful results in spite of errors in T₂, practical T₂ mapping would benefit from a comparably simple analytical formula that is even more accurate.

In this work, we introduce a new model that can be used for T₂ estimation from a single DESS scan. Our analysis demonstrates that higher-order echo pathways can often be ignored, based on pathway cancellation effects and signal decay, leading to a simple and accurate linear relationship between the signals from two DESS echoes. We validate the overall approach using simulations, phantom scans and in vivo knee scans.

2. Theory

2.1 Derivation of Simplified Signal Ratio

A simple yet accurate linear approximation of the relationship between the two DESS echoes can be derived by tracing the components that contribute to the magnetization, starting at S₂, until we arrive at S_1. We use the Extended Phase Graph (EPG) formalism, which decomposes magnetization into a basis of positively and negatively dephased transverse states (coefficients F₀, F₁, F₂, …) as well as sinusoidally varying longitudinal states (coefficients Z₀, Z₁, Z₂, …) [20,21]. The procedure is explained graphically in Fig. 1a–c, using similar graphical structures as those introduced by Scheffler et al. [18]. As explained by Weigel et al. [17,21], in the EPG formalism, the magnetization can be represented by a matrix of coefficents (F_n and Z_n) that multiply transverse states with n cycles of positive or negative dephasing, and longitudinal states with n cycles of sinusoidal variation:

$$M=\left[\begin{array}{cccc}{F}_0& F_1& F_2& \dots \\ F_0^{\ast }& F_{-1}^{\ast }& F_{-2}^{\ast }& \dots \\ Z_0& Z_1& Z_2& \dots \end{array}\right],$$ (2)

where an asterisk (*) denotes a complex conjugate. The observable signal is the DC transverse term F₀. Relaxation and diffusion over time are represented by scaling of the elements (T₁ relaxation is represented by an additive factor to the DC longitudinal term, Z₀) and the effect of a gradient is represented by shifting the transverse coefficients F_n.

Figure 1.

Diagrams showing dephasing across a voxel as a function of time using graphics similar to ref. [18]. In each panel, we consider the magnetization at the time of the right blue dot and examine the components from the time of the left blue dot that contributed to it. Graphical representations are shown of transverse and longitudinal states (above and below the dotted line, respectively). The accrued phase of the transverse magnetization is represented by the distance of the black line from the time axis. (a) The state that contributes to the measurable signal (S₂) before the RF pulse is the “negatively dephased” signal following the previous RF pulse. (b) Two relevant paths before the RF pulse contribute to the state in panel a [21], shown with black arrows. These are transverse and longitudinal states that have a net dephasing from one gradient. They also contribute to the longitudinal state following the RF pulse (gray arrows). (c) The measurable signal following the previous RF pulse (S₁), before dephasing by the gradient, contributes to the “positively dephased” transverse pathway from panel b. This is the desired endpoint so the pathway is not retraced further backward. The longitudinal pathway is not dephased by the gradient. The contribution of a third transverse pathway has been ignored since it comes from a doubly dephased signal, which can be neglected (shown with red x’s). This gives a recursive relationship between S1 and S2 that can be easily solved, resulting in Eq. (6).

The effects of T₁/T₂ decay, as well as diffusion, on state n, can be described by the scaling terms ${\stackrel{\sim }{E}}_{2,n}=e^{-\frac{TR}{T2}-\left[\left(n^2+n+\frac{1}{2}\right)TR-\frac{\tau }{6}\right]\mathrm{\Delta }{k}^{2}D}$ and ${\stackrel{\sim }{E}}_{1,n}=e^{-\frac{TR}{T1}-n^{2}TR\mathrm{\Delta }{k}^{2}D}$ (assuming the gradient in the center of the TR) [17]. The effect of an RF pulse with flip angle α and phase ϕ is represented by matrix multiplication with the matrix

$$R(\alpha ,\varphi )=\left[\begin{array}{ccc}{cos}^2\frac{\alpha }{2}& e^{2i\varphi }{sin}^2\frac{\alpha }{2}& -i{e}^{i\varphi }sin\alpha \\ e^{-2i\varphi }{sin}^2\frac{\alpha }{2}& {cos}^2\frac{\alpha }{2}& i{e}^{-i\varphi }sin\alpha \\ -\frac{i}{2}{e}^{-i\varphi }sin\alpha & \frac{i}{2}{e}^{i\varphi }sin\alpha & cos\alpha \end{array}\right]$$ (3)

Here, TR and TE represent the repetition and echo time, respectively, α is the flip angle, D is the diffusivity, and the dephasing per unit length induced by the unbalanced gradient is denoted by Δk =γGτ, where G and τ are the spoiler amplitude and duration, respectively, and γ is the gyromagnetic ratio. We label the states immediately before/after the RF pulse with −/+ superscripts, respectively. To determine a relationship between the two DESS signals, we look at the echo pathway of the observable signal immediately before the pulse, F₀⁻, and write down the components that contributed to it. We repeat this until we reach the observable signal immediately after the pulse, F₀⁺ (Fig. 1). As justified in Section 2.2, we will ignore signal contributions from states that have experienced two cycles of dephasing or more, as was done for the S₂ signal in ref. [22]. We will assume ϕ=90°, resulting in real valued states, without loss of generality.

$$\begin{array}{l}{F}_0^{-}={\stackrel{\sim }{E}}_{2,-1}{F}_{-1}^{+}={\stackrel{\sim }{E}}_{2,-1}\left(-{sin}^2\frac{\alpha }{2}{F}_1^{-}+{cos}^2\frac{\alpha }{2}{F}_{-1}^{-}-sin\alpha Z_1^{-}\right)\\ \hfill =-{\stackrel{\sim }{E}}_{2,-1}\left({sin}^2\frac{\alpha }{2}{\stackrel{\sim }{E}}_{2,0}{F}_0^{+}-{cos}^2\frac{\alpha }{2}{\stackrel{\sim }{E}}_{2,-2}{F}_{-2}^{+}+sin\alpha {\stackrel{\sim }{E}}_{1,1}{Z}_1^{+}\right)\hfill \\ \approx -{\stackrel{\sim }{E}}_{2,-1}\left({sin}^2\frac{\alpha }{2}{\stackrel{\sim }{E}}_{2,0}{F}_0^{+}+sin\alpha {\stackrel{\sim }{E}}_{1,1}{Z}_1^{+}\right)\end{array}$$ (4a)

$$\begin{array}{l}\hfill Z_1^{+}=\frac{1}{2}sin\alpha F_1^{-}+\frac{1}{2}sin\alpha F_{-1}^{-}+cos\alpha Z_1^{-}\hfill \\ \hfill =\frac{1}{2}sin\alpha {\stackrel{\sim }{E}}_{2,0}{F}_0^{+}+\frac{1}{2}sin\alpha {\stackrel{\sim }{E}}_{2,-2}{F}_{-2}^{-}+cos\alpha {\stackrel{\sim }{E}}_{1,1}{Z}_1^{-}\hfill \\ \hfill \approx \frac{1}{2}sin\alpha {\stackrel{\sim }{E}}_{2,0}{F}_0^{+}+cos\alpha {\stackrel{\sim }{E}}_{1,1}{Z}_1^{-}\hfill \\ \Rightarrow Z_1^{+}\approx \frac{1}{2}\frac{sin\alpha {\stackrel{\sim }{E}}_{2,0}}{1-cos\alpha {\stackrel{\sim }{E}}_{1,1}}{F}_0^{+}\end{array}$$ (4b)

Combining Eqs. (4a–b) and using some algebra and trigonometric identities yields

$$\frac{F_0^{-}}{F_0^{+}}\approx -{\stackrel{\sim }{E}}_{2,-1}{\stackrel{\sim }{E}}_{2,0}{sin}^2\left(\frac{\alpha }{2}\right)\frac{1+{\stackrel{\sim }{E}}_{1,1}}{1-cos\alpha {\stackrel{\sim }{E}}_{1,1}}$$ (5)

Using the stated expressions for the decay factors as well as the fact that the observed signal magnitudes are $S_1=F_0{e}^{-\frac{TE}{T{2}^{\ast }}}$ and $S_2=-{\stackrel{\sim }{E}}_{2,-1}{F}_{-1}{e}^{\frac{TE}{T2}}{e}^{-\frac{TE}{T{2}^{\prime }}}$ (where we have used that the two signals are out of phase and the second signal experiences T₂′ rephasing), we arrive at the following expression:

$$\frac{S_2}{S_1}=e^{-\frac{2(TR-TE)}{T2}-\left(TR-\frac{\tau }{3}\right)\mathrm{\Delta }{k}^{2}D}{sin}^2\left(\frac{\alpha }{2}\right)\left(\frac{1+e^{-\frac{TR}{T1}-TR\mathrm{\Delta }{k}^{2}D}}{1-cos\alpha e^{-\frac{TR}{T1}TR\mathrm{\Delta }{k}^{2}D}}\right)$$ (6)

This relationship, which assumes the echoes are acquired symmetrically during the TR (with the same duration from the RF pulse to S₁ as from S₂ to the subsequent RF pulse), provides a simple analytic estimate of T₂ given an estimate of T₁ and either minimal diffusion sensitivity or an estimate of diffusivity.

In the case of low spoiling (Δk ≈ 0), the model reduces to the following relationship:

$$\frac{S_2}{S_1}=e^{-\frac{2(TR-TE)}{T2}}{sin}^2\left(\frac{\alpha }{2}\right)\left(\frac{1+e^{-\frac{TR}{T1}}}{1-cos\alpha e^{-\frac{TR}{T1}}}\right)$$ (7)

For α = 90°, this further reduces to

$$\frac{S_2}{S_1}=e^{-\frac{2(TR-TE)}{T2}}\frac{1+e^{-\frac{TR}{T1}}}{2}$$ (8)

In the case of very long T₁ Eq. (7) becomes independent of flip angle and results in Eq. (1). This equation was presented by Bruder et al. and has been used by Welsch et al, for T₂ estimation in cartilage [1,9]. However, this relationship results in errors unless the flip angle is close to 90° or TR/T₁ is very small. The relationship also assumes weak spoiling so that diffusion effects may be ignored. Eq. (6) better approximates the signal over a large range of T₁, flip angle α, and (zeroth) spoiler gradient moments. This is demonstrated in Fig. 2a, which shows the approximated signal from Eqs. (1) and (6), as well as highly accurate numerical computation of the signal using EPG matrices. The signal ratio is modeled with a typical T₁ value of 1.2 s for cartilage [23], assuming both weak and strong spoiling (with spoiler moments of 0.001 and 156.6 mT/m•ms, respectively), as well as a long T₁ value of 5 s for the weakly spoiled case, and a TR of 22.5 ms. Other parameters were TE = 6.5 ms, T₂ = 40 ms, gradient duration τ = 3.4 ms, and ADC=1.6•10⁻³ mm/s. In order for Eq. (1) to be valid with smaller flip angles (about 20°–60°), T₁ clearly needs to be much larger than TR.

Figure 2.

(a) Comparison of proposed model of Eq. (6) (dashed) with complete simulations (colored solids) and with the simple exponential model of Eq. (1) (black solid) in a tissue with T₂ = 40 ms. The blue (bottom) curves correspond to strong spoiling (gradient moment 156.6 mT/m•ms) and the green (middle) curves to weak spoiling (gradient moment 0.001 mT/m•ms), both with T₁ = 1.2 s. The red (top) curves correspond to weak spoiling and T₁ = 5 s. The green curves represent DESS scans that can be used for T₂ estimation in cartilage. (b) Positive (dashed) and negative (solid) contributions to S2/S1 from pathways spending 2–12 TRs in the transverse plane. The black curves represent the sum of the different pathways. For pathways of order 6 and above, the positive (×) curves overlap with the negative (solid) curves. (c) Net contributions from the pathways in panel b shows that pathways with more than 2 TRs in the transverse plane contribute minimally. (d) The same simulations as in panel c, performed for synovial fluid at 3T (T₁ = 3.6 s, T₂ = 0.77 s), showing that the proposed model (blue signal) approximates the true signal (black curve) well for flip angles above 20°, even for long T₁ and T₂.

2.2 Cancellation of higher-order echo pathways

The model in Eq. (6) performs well for two reasons. The equation neglects the contribution of echo pathways between S₁ and S₂ that have spent more than two TRs in the transverse plane. This is partly justified by simple signal decay – the more time that the magnetization spends in the transverse plane, the more it will be attenuated both by T₂ relaxation and diffusion. However, a more important mechanism is the cancellation of echo pathways due to opposing phase, independent of relaxation. It is easy to show that all contributions to S₂ that have experienced two transverse TRs will have negative phase relative to S₁. However, this does not hold for higher order echo pathways. For example, if we denote TRs where the magnetization starts in the +/− n^th transverse state or the n^th longitudinal state by F_+/−n and Z_n, then the pathways F₀F₁F₋₂F₋₁ and F₀Z₁F₁F₋₂F₋₁ will have opposite sign. By accumulating the contributions of all such pathways, the positive and negative contributions (relative to S₁) of higher-order pathways can be modeled and compared. For example, by including pathways with two and four transverse TRs, Eq. (6) can be extended to

$$\frac{S_2}{S_1}=B_{2-}-B_{4+}+B_{4-}$$ (9)

where B₂₋ is the two-TR component given by Eq. (6), and B₄₊ and B₄₋ are the four-TR components with the same and opposite phase to S₁, respectively, given by

$$B_{4+}=2\mathit{CDEF}$$ (10)

$$B_{4-}=C(D^2+E^2)F$$ (11)

where

$$C=e^{-\frac{4TR-2TE}{T2}-\left(6TR-\frac{2\tau }{3}\right)\mathrm{\Delta }{k}^{2}D}$$ (12)

$$D={cos}^2\left(\frac{\alpha }{2}\right)$$ (13)

$$E=\frac{\frac{1}{2}{sin}^2\alpha e^{-\frac{TR}{T1}-TR\mathrm{\Delta }{k}^{2}D}}{1-cos\alpha e^{-\frac{TR}{T1}-TR\mathrm{\Delta }{k}^{2}D}}$$ (14)

$$F={sin}^2\left(\frac{\alpha }{2}\right)+\frac{\frac{1}{2}{sin}^2\alpha e^{-\frac{TR}{T1}-4TR\mathrm{\Delta }{k}^{2}D}}{1-cos\alpha e^{-\frac{TR}{T1}-4TR\mathrm{\Delta }{k}^{2}D}}$$ (15)

Fig. 2b displays the components B_n+ and B_n− for n = 2–12 over a range of flip angles with the same parameters as the green curve in Fig. 2a. The higher-order components with positive and negative phase have very similar magnitude. Therefore, they will have a negligible effect on the final answer, particularly for flip angles above 20° or short T₂. This is demonstrated in Fig. 2c. This leads to Eq. (6) simulating the signal ratio S₂/S₁ very accurately, even in cases where there is very little T₂ decay, such as in fluid. Fig. 2d shows the same simulation as Fig. 2c, but for synovial fluid at 3T, having T₁ = 3.6 s and T₂ = 0.77 s [23]. The higher-order pathways contribute more to the signal ratio in this case, especially for low flip angles, but Eq. (6) nonetheless approximates the true signal well for flip angles above 20°. The approximation of neglecting higher-order states was also used by Buxton [22]. In that work, only the S₂ signal was modeled. If S₁ in Eq. (6) is assumed to have no transverse pathway contributions, the two models agree, as further explained in the Discussion.

3. Methods

All scans were performed on a 3T 750 scanner (GE Healthcare, Waukesha, WI, USA). Informed consent was obtained from all subjects in accordance with the Institutional Review Board protocol at our institution. The volunteer scans used a 16-channel receive coil wrapped around the knee (GEM Flex by NeoCoil, Waukesha, WI, USA), the phantom scans used a single-channel transmit-receive wrist coil (GE BC-10 by Mayo Clinic, Rochester MN, USA), and the patient scans used an 8-channel transmit-receive knee coil (Knee Array Coil by Invivo, Gainesville, FL, USA).

3.1 Phantom scans

Three agar phantoms were scanned using DESS with scan parameters TR=13.4 ms, TE=3.9 ms, voxel size 0.47×0.47×3.0 mm³, spoiler moment 7.83 mT/m•ms and duration τ=3.4 ms, and flip angles α = 10°, 15°, 20°, 30°, 40°, and 50°. Fast spin-echo (FSE) scans with echo times TE = 17, 26, 34, 43, 51, 60, and 68 ms, TR = 1000 ms and voxel size 0.47×0.47×6.0 mm³ were performed and fitted to a monoexponential curve to obtain a reference T₂. The T₂ value was estimated from the DESS scans using Eqs. (1) and (7) and compared to the value resulting from the reference FSE scans. For Eq. (7), reference T₁ values known to be T_1-1 = 2000 ms, T_1-2 = 1700 ms, and T_1-3 = 1600 ms, measured with a 3TI MP-RAGE technique [24], and a flip angle measured with the Bloch-Siegert method [25], were used for processing each phantom.

3.2 Volunteer scans

To test Eq. (7) in vivo, a sagittal DESS scan was performed in the knee of three healthy volunteers. The scans used a spoiler with a moment of 15.66 mT/m•ms and duration 3.4 ms. Other parameters were α=25°, TR=22.5 ms, TE=6.5 ms, slice thickness 3.0 mm. A spectrally selective RF pulse was used. The T₂ of the cartilage was again estimated using the linear approximations of Eqs. (1) and (7). Reference FSE scans were acquired with TE = 9, 18, 27, 36, 45, 54, 63, and 72 ms and TR = 1000 ms. Both scans were run with FOV = 13cm×13cm and acquisition matrix 384×384, which was automatically interpolated to 512×512 by the scanner. For T₂ quantification using Eq. (7), T₁=1.2 s, typical for cartilage at 3T [23], and the prescribed flip angle of α=25° were used. Eq. (7) has low sensitivity to T₁, so small errors in the T₁ assumption should not cause large errors in estimated T₂. Furthermore, T₁ in knee cartilage does not change much whereas T₂ is sensitive to degeneration. For comparison, a T₂ map using a full numerical fit was produced from the DESS scan as well. This involved computing a dictionary of signal ratios for T₂ ranging between 10–100 ms (with a step size of 1 ms) using large EPG matrices (with 6 states) and then choosing the dictionary entry that best fitted each pixel. The total scan time was 6:26 min for FSE (acquiring 6 slices) and 5:11 min for DESS (acquiring 36 slices). In slices where both scans displayed clear anterior and posterior regions of the femoral cartilage, the resulting T₂ maps in those two regions were divided into deep and superficial regions, resulting in 4 regions per slice. This was possible in 10 slices, resulting in a total of 40 regions. The root-mean-squared (RMS) difference between the results from Eq. (7) and the FSE results was computed and compared to the difference between the results from Eq. (1) and the reference scans. The T₂ maps were computed in Matlab (The MathWorks, Natick, Ma) on a Linux terminal with 128 GB of RAM and 32 2.6 GHz CPUs.

3.3 Patient scans

As an initial demonstration of the value of Eq. (7) for routine diagnostic imaging, we selected data from three patients from a larger group scanned with DESS in a clinical setting (spoiler moment = 31.32 mT/m•ms, spoiler duration = 3.4 ms, α=20°, TR=20.4 ms, TE=6.4 ms, FOV = 16cm×16cm, acquisition matrix 416×512, 80 slices with thickness of 1.6mm interpolated to 0.8 mm (160 slices), 2× parallel imaging in the phase encode direction, scan time = 5:00). T₂ maps were constructed automatically at the scanner. A proton-density (PD) weighted scan was acquired for comparison.

4. Results

4.1 Phantom scans

The results from the phantom scans are shown in Fig. 3. The measured reference T₂ values were T_2-1 = 62 ms, T_2-2 = 44 ms, and T_2-3 = 31 ms for the three phantoms. Eq. (1) underestimates T₂ compared to the reference FSE estimate, and this becomes more pronounced for smaller flip angles. The estimate from Eq. (7) agrees very well with the reference scans for most flip angles (20° and higher), but gives a substantial overestimate in the case of a large T₂ and 10° or 15° flip angles. However, the error is much smaller than the error from Eq. (1).

Figure 3.

Results from estimating T₂ from DESS using Eq. (1) (dashed) and Eq. (7) (solid) as well as from reference FSE scans (dotted). For all phantoms, Eq. (7) clearly agrees better with the reference value than Eq. (1), and for flip angles of 20° or higher, Eq. (7) agrees very well with the reference scans.

4.2 Volunteer scans

Fig. 4 shows the T₂ estimates from the DESS scan, using Eqs. (1) and (7), as well as the results from the reference FSE scans and a full numerical fit. The maps from Eq. (7) are more visually similar to the reference scans than those of Eq. (1) (Fig. 4a–c). The RMS difference between the estimates of Eq. (7) and FSE was 3.7 ms, while for Eq. (1) it was 9.6 ms. Plotting the data points from Eq. (7) against the FSE data points and drawing a best fit line that crossed the origin yielded a slope of 0.96, while for Eq. (1) it was 0.77 (Fig. 4f). This demonstrates that Eq. (7) gave T₂ estimates that better agreed with FSE estimates than Eq. (1). The full numerical fit gives very similar results to those obtained by Eq. (7), as shown both by its T₂ map and by a difference map (scaled by 10), while taking around 60× longer to produce, with an average processing time of 577 s per DESS data set (36×512×512 data points) compared to 9 s for Eq. (7).

Figure 4.

(a) The first echo (S₁) from a sample sagittal DESS scan. (b) The second echo (S₂) from the same DESS scan. The windowing level is not the same as in panel a, in order to show the cartilage. (c) A sample T₂ map from FSE scans of the subject in panels a–b. (d) T₂ map of articular cartilage from a the DESS scan in panels a–b using Eq. (1). (e) T₂ map using Eq. (7). (f) T₂ map using a full numerical fit. The map looks very similar to the one in panel e, but took about 60× longer to produce. (g) A map showing the absolute difference between the maps in panels e and f, multiplied by 10. The difference in the cartilage is small, mostly 1 ms or less. Zero difference appears as transparent. (h) The T₂ estimates from 4 regions of femoral cartilage in a total of 10 slices from 3 subjects. The trend line for Eq. (7) (green) better agrees with the FSE scans.

4.3 Patient scans

Fig. 5 shows the scan of a patient with a history of knee injuries and osteochondral defects, visible on the DESS images. Fibrocartilage formation is present in the damaged hyaline cartilage region and in the surrounding subchondral bone. The corresponding region in the DESS T₂ map shows a focal decrease in T₂ values. In another patient (not shown), the availability of the DESS T₂ map made it possible to visualize an oblique meniscal tear in the anterior horn of the medial meniscus, which was challenging to delineate with only the morphological images. In a third patient (not shown), a T₂ map acquired with DESS indicated tendinopathy where the patellar tendon had increased T₂ values at its insertion site, which was also relatively challenging to visualize with only the morphological DESS images.

Figure 5.

(a) The first echo of a DESS scan of a patient with a chondral lesion (white arrow). (b) The second echo of the DESS scan. Panels a and b both show cartilage signal heterogeneity in the central femoral condyle (note the different windowing settings). (c) The T₂ maps resulting from applying Eq. (7) to the data in panels a–b. (d) A PD weighted scan, acquired for reference, shows low signal in the lesion.

5. Discussion

The DESS sequence allows 3D estimation of T₂ in cartilage with high SNR efficiency. The signal expressions are complicated, often necessitating imprecise simplifications or time-consuming numerical modeling of the signals. This study explores a linear relationship between the two DESS signals that can be used for accurately estimating T₂ from a single scan. We have focused on applying the method in cartilage, but the proposed method could also be used in various other anatomies. This could include T₂ mapping in breast cancer patients, which to date is impractical due to the time consuming nature of spin-echo-based T₂ measurements [26]. Also, this methodology has been used to perform relaxometry of short-T₂ tissues in the knee [27].

The proposed method can be viewed as an extension to Eq. (1), used in ref. [9], that accounts for T₁, flip angle, and diffusion. The expression results from applying a similar approximation to that used by Buxton [22], ignoring echo pathways that have spent more than two TRs in the transverse plane. It is important to note, however, that ref. [22] only dealt with sequences acquiring the second echo S₂. This required making the assumption that magnetization is transverse for only two TRs from the start of the sequence. In this work, the two-transverse-TR assumption is only applied between the echoes S₁ and S₂, which is a much weaker assumption, resulting in good performance in spite of a simpler model. Effectively, the model of ref. [22] can be viewed as the model proposed here in the special case of the S₁ signal having no transverse history, i.e., being the RF-spoiled or SPGR signal. If S₁ in Eq. (6) is replaced with the well-known expression for the RF-spoiled signal, the two models agree (resulting in the same expression for S₂).

The scan time for DESS used in this study was about 5 minutes. The patient scans were acquired at a higher resolution than the volunteer scans, but using parallel imaging, the scan time remained at 5 minutes at a cost in SNR. Scan time could be reduced by using lower resolution, higher readout bandwidth, fewer slices, or shorter RF pulses. However, in our experience, a 5 minute scan time is acceptable, especially considering that these 3D scans provide both morphological and quantitative information. Potentially, DESS could replace sagittal FSE scans that are widely used in many current protocols, but this needs to be studied further.

As mentioned, using the linear approximation of Eq. (7) for estimating T₂ from a single DESS scan requires an assumed T₁ value. This should not lead to large estimation errors, since the method has little sensitivity to T₁, which is also not expected to change much in cartilage. For example, using the same scan parameters as for the middle curve in Fig. 2a, with the same assumption of T₁ = 1.2 s, yields T₂ estimates of 38.8 ms and 41.8 ms (−3.0% and +4.5%) when the real T₁ is 0.96 s and 1.44 s (i.e., the actual T₁ is 20% lower or higher than the assumed value), respectively. Both values are relatively close to the actual value of 40.0 ms. A correct assumption of T₁ = 1.2 s yields T₂ = 40.6 ms (+1.5%), a slight overestimate due to the approximation made. Similarly, sensitivity to B₁ results in estimates of 36.9 ms and 43.1 ms (−7.75% and +7.75%) when the actual flip angle α is 20° and 30° (i.e., 20% lower or higher than the assumed value of 25°). The B₁ sensitivity could be mitigated by acquiring a B₁ map along with the DESS scan. The sensitivity of the T₂ estimate to errors in the T₁ and α assumptions is further demonstrated in Fig. 6. It should also be noted that assumptions of T₁ and flip angle will always be needed when estimating T₂ from a single DESS scan, regardless of the model used.

Figure 6.

The sensitivity of the T₂ estimation method in Eq. (7) to errors in the assumption of T₁ and B₁. The true T₂ is 40 ms. The flip angle α and T₁ are assumed to be 25° and 1.2 s, but actually vary from these values by +/− 20%. Other scan parameters were the same as in Fig. 2a. The black circle shows the point where the assumptions of α and T₁ are correct.

The proposed method can also be used in combination with other methods for quantitative DESS imaging. Cheng et al. proposed a multi-echo DESS sequence for imaging T₂ and T₂* that measures S₁ and S₂ at different echo times and uses the fact that T2′ causes dephasing of S₁ and rephasing of S₂ [28]. Put together, the two methods could potentially estimate other parameters such as T₁ or flip angle. Also, it has previously been noted by Bieri et al. that the ratio (S_2HS_1L)/(S_1HS_2L), where H and L denote scans with a large and small (Δk ≈ 0) spoiler gradient moment, respectively, has good diffusion sensitivity but low sensitivity to T₁ and T₂ [11]. This is consistent with the expression derived from Eq. (6):

$$\frac{S_{2H}{S}_{1L}}{S_{1H}{S}_{2L}}=e^{-\left(TR-\frac{\tau }{3}\right)\mathrm{\Delta }{k}^{2}D}\frac{1+e^{-\frac{TR}{T1}}\left(e^{-TR\mathrm{\Delta }{k}^{2}D}-cos\alpha \right)-e^{-\frac{2TR}{T1}}cos\alpha e^{-TR\mathrm{\Delta }{k}^{2}D}}{1+e^{-\frac{TR}{T1}}\left(1-cos\alpha e^{-TR\mathrm{\Delta }{k}^{2}D}\right)-e^{-\frac{2TR}{T1}}cos\alpha e^{-TR\mathrm{\Delta }{k}^{2}D}}$$ (16)

This relationship is independent of T₂, and largely insensitive to T₁.

In conclusion, we have demonstrated a simplified expression for the ratio between the DESS signals, providing good T₂ estimation. The expression disregards higher-order echo pathways, an assumption we have shown to be valid due to both decay and cancellation of such pathways. The expression takes into account signal dependency to T₁ and flip angle, and we have shown that the method works well by assuming a T₁ typical for the tissue and using the prescribed flip angle in calculations.

Highlights.

• A simple method for estimating T2 from a single DESS scan is proposed

• The method gave good results in simulations and in phantoms and in vivo scans

• Easy implementation and fast processing allow for near-instant T2 maps at scanner