Abdominal T2-weighted Imaging and T2 Mapping using a Variable Flip Angle Radial Turbo Spin Echo Technique

Abstract

Background:

T2 mapping is of great interest in abdominal imaging but current methods are limited by low resolution, slice coverage, motion sensitivity, or lengthy acquisitions.

Purpose:

Develop a radial turbo spin-echo technique with refocusing variable flip angles (RADTSE-VFA) for high spatio-temporal T2 mapping and efficient slice coverage within a breath-hold and compare to the constant flip angle counterpart (RADTSE-CFA).

Study Type:

Prospective technical efficacy.

Phantom/subjects:

Testing performed on agarose phantoms and 12 patients. Focal liver lesion classification tested on malignant (n=24) and benign (n=11) lesions.

Field Strength/Sequence:

1.5T/RADTSE-VFA, RADTSE-CFA

Assessment:

A constrained objective function was used to optimize the refocusing flip angles. Phantom and in vivo data were used to assess relative contrast, T2 estimation, specific absorption rate (SAR), and focal liver lesion classification.

Statistical Tests:

t-tests or Mann Whitney Rank Sum tests were used.

Results:

Phantom data did not show significant differences in mean relative contrast (p=0.10) and T2 accuracy (p=0.99) between RADTSE-VFA and RADTSE-CFA. Adding noise caused T2 overestimation with a higher degree for RADTSE-CFA and low T2 values.

In vivo results did not show significant differences in mean spleen-to-liver (p=0.62) and kidney-to-liver (p=0.49) relative contrast between RADTSE-VFA and RADTSE-CFA. Mean T2 values were not significantly different between the two techniques for spleen (T2_VFA=109.2±12.3 ms; T2_CFA=110.7±11.1 ms; p=0.78) and kidney-medulla (T2_VFA=113.0±8.7 ms; T2_CFA=114.0±8.6 ms; p=0.79). Liver T2 was significantly higher for RADTSE-CFA (T2_VFA=52.6±6.6 ms; T2_CFA=60.4±8.0 ms) consistent with T2 overestimation in the phantom study. Focal liver lesion classification had comparable T2 distributions for RADTSE-VFA and RADTSE-CFA for malignancies (p=1.0) and benign lesions (p=0.39).

RADTSE-VFA had significantly lower SAR than RADTSE-CFA increasing slice coverage by 1.5.

Data Conclusion:

RADTSE-VFA provided noise robust T2 estimation compared to the constant flip angle counterpart while generating T2-weighted images with comparable contrast. The VFA scheme minimized SAR improving slice efficiency for breath-hold imaging.

INTRODUCTION

Radiologists use T2-weighted MRI of the abdomen as one of the core imaging acquisitions for the detection and characterization of an array of important pathologies. Diagnostic analysis is generally dependent on the qualitative interpretation of images by an experienced radiologist. Sequence parameters that affect contrast, such as the echo time, may affect a radiologist’s interpretations and these parameters may be inconsistent between studies, scanners or centers. As an alternative, T2 mapping is a quantitative approach that may provide standardization and reproducible results (1,2). T2 mapping may also reveal tissue features related to disease that may otherwise be unobserved by standard qualitative imaging interpretations (1). To date, T2 mapping of the abdomen has been reported sparingly. Most reports have focused on the characterization of focal liver lesions with the goal of differentiating malignancies from the most common benign lesions (bile duct hamartomas and hemangiomas) (3–6). A few more recent reports have explored the utility of T2 mapping to assess changes in the liver parenchyma and other abdominal organs (7–9).

T2 mapping techniques applied to the abdomen have been based primarily on Cartesian free breathing spin-echo or turbo spin-echo methods (used in combination with respiratory gated techniques) (3,5,10,11), or on faster techniques based on breath hold dual echo turbo spin-echo (6), echo planar (4), single-shot fast spin-echo (4,12), or balanced steady-state free precession (bSSFP) readouts (13), where the acquisition of the multiple echo-time (TE) data sets required for T2 estimation can be fit within a breath hold. Free-breathing, respiratory-gated spin-echo based methods can provide good spatial and temporal resolution as well as good slice coverage (11) but are not efficient in terms of imaging time, which is exacerbated when the subject’s breathing pattern is irregular. These techniques are also more sensitive to motion, caused by residual breathing, peristalsis, and blood flow, affecting the intrinsic spatial resolution of the technique (8,11) which in turn may affect T2 estimation accuracy due to volume averaging (3). The breath hold techniques with single shot readouts are robust to motion at the expense of low spatial and temporal resolution (e.g., 2–3 TEs), limited slice coverage (1–2 slices) per breath hold, and misregistration when TE data sets are acquired consecutively through the breath hold (14). MR fingerprinting, a technique recently developed for multi-parametric imaging, has also been reported for T2 and T1 mapping of the abdomen (15). The technique has limited spatial resolution and slice coverage (1 slice per breath hold).

Multi-shot radial turbo spin echo (RADTSE) based methods have been proposed for T2 weighted (T2w) imaging and T2 mapping (9,16–19). In the RADTSE pulse sequence, all TEs sample the center of k-space providing high temporal resolution (as high as the spacing between refocusing pulses) for encoding the T2 signal decay. Moreover, due to the oversampling of the center of k-space, radial trajectories are robust to undersampling compared to Cartesian trajectories so only a few radial views (e.g., 6–8) are needed to reconstruct the TE images, rendering an efficient technique for T2 mapping (19–21). Relevant to abdominal imaging, the RADTSE technique allows for the acquisition of data with high spatial and temporal resolution for T2 mapping in a single breath hold (20). Due to the nature of the TSE acquisition (where TE data are acquired within a window of ~200 ms in each TR), the resulting abdominal TE images are co-registered simplifying the computations of T2 maps. RADTSE is also motion robust compared to Cartesian multi-shot TSE methods (20,22) where ghosting due to blood flow and residual breathing can affect image quality even in breath held acquisitions (20). Algorithms based on echo sharing (16), compress sensing (16,19–22), and convolutional neural networks (23) have been developed to reconstruct highly undersampled RADTSE data (e.g., 3–5% sampling relative to Nyquist). These algorithms have been shown to yield accurate T2 values even in the presence of RF pulse imperfections (e.g., when the signal from the refocusing pulses is < 180°) (24).

In turbo spin-echo pulse sequences, the use of a long echo train length (ETL) is desired to reduce the total acquisition time as well as to accommodate longer sequence repeat times (TRs) to increase SNR and the number of slices acquired per TR. However, the higher number of refocusing pulses associated with long ETLs results in an increase of specific absorption rate (SAR), limiting the number of slices that can be acquired within a given TR (25,26). Moreover, long ETLs suffer from a loss of signal in the latter echoes, due to T2 decay (25). Although, the use of refocusing pulses with flip angles less than 180° is an alternative for reducing SAR, the latter echoes are still affected by noise due to signal decay which in turn can affect the accuracy of the T2 estimation (27).

Techniques based on variable refocusing flip angle (VFA) schemes have been proposed (28–30) to prolong the signal evolution and reduce SAR in very long echo train TSE pulse sequences (19,21). The aim of this work is to design a VFA scheme for improved T2 mapping and efficient slice coverage with RADTSE for breath hold abdominal imaging. The radial turbo spin-echo with variable refocusing flip angle (RADTSE-VFA) technique is demonstrated in phantoms and in human subjects in vivo abdominal imaging and compared against its constant flip angle counterpart.

METHODS

Refocusing Flip Angle Design

In Cartesian imaging, flip angle designs have focused on minimizing spatial blurring while maintaining low SAR (29,30). A practical design parametrizes the refocusing flip angles by four control angles $\overrightarrow{\alpha }= \left[{\alpha }_{max},{\alpha }_{min},{\alpha }_{cent},{\alpha }_{end}\right]$ which are computed using a prospective extended phase graph (EPG) algorithm (29). The control angles are then used to set the target signal intensity for the pseudo-steady state signal along the echo train. For T2w imaging and T2 mapping of the abdomen with RADTSE-VFA, we designed an analytical framework for the refocusing flip angle scheme. To improve the T2 estimation accuracy, the control angles were chosen to minimize the Cramer-Rao lower bound (CRLB) for T2 estimation while increasing the area under the T2 decay curve (AUDC). The latter constraint increases the SNR for T2 estimation by prolonging the signal evolution curve and minimizes the effect of noise on the late echoes. This can be expressed in the following optimization

$$\underset{\overrightarrow{\alpha }}{\text{min}} crlb(\overrightarrow{\alpha )}$$ [1]

$$snr\left(\overrightarrow{\alpha }\right)>ϵ ;$$

where, $crlb\left(\overrightarrow{\alpha }\right)$ is the analytical Cramer-Rao lower bound for liver T2 and $snr\left(\overrightarrow{\alpha }\right)$ is defined as the area under the decay curve (AUDC).

In order to maintain a desired T2w contrast for the images, an additional constraint was added with the goal of maximizing the relative contrast between liver and malignant hepatic lesions using T2_liver = 40 ms, T1_liver = 550 ms, T2_lesion = 80 ms, T1_lesion = 1000 ms (4,14). Further, a constraint was added to maintain SAR within acceptable limits. Thus, the flip angle design problem can be expressed as a constrained optimization:

$$\underset{\overrightarrow{\alpha }}{\text{min}} crlb\left(\overrightarrow{\alpha }\right)$$ [2]

$$s.t. sn{r}_{liver}\left(\overrightarrow{\alpha }\right)+ sn{r}_{lesion}\left(\overrightarrow{\alpha }\right)>ϵ ;$$

$$rel{C}_{liver, lesion}\left(T{1}_{liver},T{2}_{liver},T{1}_{lesion},T{2}_{lesion},B1,T{E}_{eff},\overrightarrow{\alpha }\right) >\tau ;$$

$$b{1}_{+rms}\left(\overrightarrow{\alpha }\right)<\eta ;$$

$$\varphi < {\alpha }_{min}<{\alpha }_{max}; \varphi <{\alpha }_{cent}<{\alpha }_{max}$$

where $rel{C}_{liver,lesion}(.)$ is the relative contrast between liver and lesion at the desired effective TE $\left(T{E}_{eff}\right)$ and is defined as $rel{C}_{liver,lesion}= \frac{|S_{lesion}\left(T{E}_{eff}\right)-S_{liver}\left(T{E}_{eff}\right)|}{S_{liver}\left(T{E}_{eff}\right)}$. $T{E}_{eff}$ is the contrast equivalent echo time defined as the TE that would achieve the same signal decay as a mono-exponential function (resulting from a perfect 180° refocusing flip angle) (29). As described in (29), the effective TE is calculated using the acquired signal and the transverse coherence component of the signal when relaxation effects are ignored. $b{1}_{+rms}$ constraint is used as a subject independent measure of SAR and is defined as a function of the total RF power ($R{F}_{total}$) and the total scan time ($T_{scan}$) of the pulse sequence: $b{1}_{+rms}=\sqrt{\frac{R{F}_{total}}{T_{scan}}}$. An additional lower bound constraint was added to reduce motion-related signal dephasing at very low refocusing flip angles, especially in regions near the heart (31). Due to the non-convex nature of the optimization problem, Equation 2 was solved using a genetic algorithm.

Simulations

All simulations were implemented in MATLAB (MathWorks, MA, USA).

A CRLB analysis as performed to understand the effect of the control angles on the estimated T2 values. The CRLB was derived using the SEPG signal model with an independent and identically distributed noise assumption. In the CRLB analysis, T2 and B1 were the parameters of interest.

Monte-Carlo simulations were performed to study the T1 sensitivity on T2 estimation due to the use of refocusing flip angles < 180°. A numerical phantom was constructed consisting of three spheres with the following relaxation times based on a normal distribution: (i) T1=800±50 ms and T2=50±5 ms, (ii) T1=1004±50 ms and T2=80±5 ms, and (iii) T1=1337±50 ms and T2=148±5 ms. These values cover typical T1 and T2 values for abdominal organs (liver, spleen, kidney) as well as malignant and benign lesions at 1.5T (5,10). Slice profiles corresponding to the RF pulses used in the RADTSE pulse sequence (constant and VFA implementations) were generated using the Shinnar–Le Roux algorithm (32) assuming an excitation and refocusing slice thickness of 10 mm. The signal at each voxel was simulated using the Bloch equation model for the TSE sequence assuming the following refocusing pulse flip angles: (A) constant flip angle 180°, (B) constant flip angle 150°, and (C) the proposed VFA. One hundred TE images were generated from the forward model and independent and identically distributed Gaussian noise was added to the images. T2 values were estimated by fitting TE images to the SEPG model using: T1 = ∞, T1 = 2000 ms, or by estimating T1 along with T2 during fitting.

MRI Experiments

The RADTSE-VFA and RADTSE with constant flip angle (RADTSE-CFA) refocusing pulses were implemented on a 1.5T MRI scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). RADTSE-CFA data were acquired using a train of 150° refocusing flip angles with the Hennig correction factor(33) resulting in the first refocusing pulse being 165°.

Phantom Experiments

To evaluate relative contrast and T2 estimation performance of the RADTSE-VFA pulse sequence, nickel-doped (1 mM NiCl₂) agarose gel phantoms were prepared using 4 different concentrations of agarose (3.6, 2.5, 1.5, 1%) to target T2 values in the 35–135 ms range; T1 for all phantoms was ~ 1000 ms. The acquisition parameters were: field of view (FOV) = 12 cm, base resolution = 256, radial views = 192, ETL = 32, echo spacing = 8.2 ms, TR = 2500 ms, slice thickness = 5 mm. T2 estimation accuracy was evaluated against T2 estimates obtained using a single-echo spin-echo pulse sequence with: FOV = 12 cm, acquisition matrix = 128x128, TR = 5 s, and 12 different TEs from 10 ms – 240 ms.

Fully sampled phantom data (i.e., 402 radial views per TE) were also acquired in the phantom to evaluate the effect of noise on T2 estimation. We used fully sampled TE data to perform the evaluation independent of the iterative reconstruction. Monte-Carlo simulations (50 noise realizations) were performed by adding complex Gaussian noise to the acquired k-space data to yield noise levels corresponding to noise standard deviations of 0.3 and 0.5. TE images were reconstructed using a non-uniform Fast Fourier Transform algorithm.

In vivo Experiments

The study was approved by the institutional review board. Subjects were imaged after obtaining informed consent.

The RADTSE-CFA and RADTSE-VFA pulse sequences were added to the clinical abdomen-pelvis protocol and breath-held data were acquired on 12 clinical patients (7 men and 5 women) with a median age of 65.5 (range=34–84 y/o). The patients included in this study had standardized contrast-enhanced comprehensive MRI exams at our center. The diagnoses for liver pathology and tumors were derived from the radiological report and confirmed in the patient electronic records. Images were also reviewed for agreement by a liver MRI expert with over 20 years of experience (DRM). Based on this analysis, 2 subjects had liver metastases, 7 subjects had chronic liver disease (one with hepatocellular carcinoma), and 3 subjects had normal livers.

Acquisition parameters for RADTSE-VFA and RADTSE-CFA data were: FOV = 40 cm, ETL = 32, echo spacing = 6.7 ms, base resolution = 256, radial views = 192, TR = 2500 ms, slice thickness = 8 mm, breath-hold time = 18 s. Fat suppression was employed using chemical-shift fat saturation. Super-inferior saturation bands were used to reduce streaking artifacts due to flow from the abdominal aorta and vena cava. With these parameters, the maximum number of slices per breath hold was 7 and 11 for RADTSE-CFA and RADTSE-VFA, respectively.

Image Reconstruction

The undersampled RADTSE TE k-space data were reconstructed using a principal component (PC) subspace constrained algorithm (18,24) with locally low rank regularization (34). The PC subspace basis was generated by simulating the signal decay over TEs using the slice resolved extended phase graph algorithm (SEPG) (35). This model accounts for the additional coherence pathways arising from use of slice-selective non-180° refocusing flip angles. The subspace constrained reconstruction problem was solved using 4 PCs to represent the T2 decay. The T2 and B1 ranges for the subspace basis were 20 – 500 ms and 0.6 – 1.3, respectively. T2 maps were generated from the reconstructed TE images by fitting them to a library of pre-computed T2 evolution curves (36) based on the SEPG signal model.

All reconstruction algorithms were implemented offline using MATLAB (MathWorks, MA, USA) and the Gadgetron framework (37) in a hybrid CPU-GPU approach resulting in a reconstruction time of ~10 minutes / slice on a workstation with a 3.6 GHz Intel Xeon processor E5–1620, 64GB RAM and an NVIDIA GeForce GTX 780 GPU.

Image Analysis

The relative contrast was evaluated on images corresponding to TE_eff = 90 ms For the phantom experiments relative contrast was computed against the vial with T2 value closest to liver (reference) according to: $rel{C}_{phantom}= \frac{|S_{phantom}\left(T{E}_{eff}\right)-S_{reference}\left(T{E}_{eff}\right)|}{S_{reference}\left(T{E}_{eff}\right)}$. In vivo relative contrast for a specific organ against liver was according to: $rel{C}_{organ}= \frac{|S_{organ}\left(T{E}_{eff}\right)-S_{liver}\left(T{E}_{eff}\right)|}{S_{liver}\left(T{E}_{eff}\right)}$.

The AUDC was computed by adding all echo images after each image was normalized by the echo image with maximum signal intensity.

Region-of-interest (ROI) drawing on the abdominal images was performed by MIA and reviewed by an expert radiologist (DRM). ROIs were drawn excluding visible blood vessels as well as areas of necrosis (on malignant lesions). For relative contrast evaluation, ROIs in the liver were drawn at the same level as those corresponding spleen and kidney. T2 estimation (mean and standard deviation) was evaluated on liver, spleen, kidney medulla, and focal liver lesions (22 metastases, 2 hepatocellular carcinomas, 1 hemangioma and 10 cysts).

Statistical Analysis

All continuous variables were tested for the normality assumption (Shapiro-Wilk’s test). A two-tailed t-test was used for all comparisons involving normally distributed data. For cases when the distribution was not normal a Mann Whitney Rank Sum test was performed. The significance level was 0.05.

RESULTS

Refocusing Flip Angle Design

Two-dimensional surface plots of the objective function and constraints used to optimize the variable flip angle train using Equation 2 are shown in Figure 1 as a function of the two design parameters ${\alpha }_{min}$ and ${\alpha }_{cent}$. Figure 1a shows a plot of the relative contrast between liver and lesion at TE_eff= 90 ms, the TE commonly used by radiologists for detecting focal liver lesions(10). Figure 1b shows the cumulative AUDC, used as a metric of SNR across all TEs. From the plot, it can be observed that the AUDC is predominantly determined by ${\alpha }_{min}$. To minimize noise in the late echoes we want to operate in regions of high AUDC, i.e. ${\alpha }_{min}\le {90}^{\circ }$. A plot of the CRLB lower bound on the variance of the T2 estimator is shown in Figure 1c. Note that the T2 estimation performance is a non-linear function of the design parameters ${\alpha }_{min}$ and ${\alpha }_{cent}$. Figure 1d shows a plot of the computed $b{1}_{+rms}$ metric that is used as a patient independent measure of SAR. Since the $b1{+}_{rms}$ is proportional to the total RF power in the pulse sequence, it linearly increases with the control angles. According to the constraints shown in Figures 1(a–d), one feasible region for the search space of ${\alpha }_{min}$ and ${\alpha }_{cent}$ is $\left[{60}^{\circ }\le {\alpha }_{min}<{90}^{\circ }\right]$ and $\left[{70}^{\circ }<{\alpha }_{cent}\le {100}^{\circ }\right]$. Based on these constraints, the control angles chosen for abdominal imaging were $\overrightarrow{\alpha }= \left[130, 60, 100, 40\right]$. The flip angle evolution is shown in Figure 1e.

Figure 1:

(a-d) 2D plots of the objective function and constraints as a function of the two design parameters ${\alpha }_{min}$ and ${\alpha }_{cent}$ for the relative contrast between liver and malignancies (simulated as T2=40ms, and T2=80ms species, respectively), the cumulative (liver and lesion) area under the decay curve (AUDC), the variance of the T2 estimator (based on the CRLB), and $b{1}_{+rms}$ used as a patient independent measure of SAR. (e) Optimized variable flip angle (VFA) scheme based on a low CRLB and $b{1}_{+rms}$ and high AUDC and relative lesion-to-liver contrast. The constant flip angle (CFA) scheme is also shown in the plot.

Simulation Results

The results of the Monte Carlo simulations analyzing the effect of T1 and B1 on T2 estimation error are shown in Supporting Table 1. As previously shown for the constant flip angle (24), assuming a constant T1 does not affect the T2 estimation performance of the VFA technique. Estimating T2 with a 3-parameter fit (T2, B1, T1) resulted in a higher standard deviation to the increased number of parameters. Based on these results, all T2 estimation was performed assuming T1=2000 ms in the signal model.

Phantom Results

T2 estimates with RADTSE-VFA and RADTSE-CFA in the agarose gel phantoms compared against reference single-echo spin-echo measurements are shown in Table 1. Note that the T2 values from RADTSE-VFA show good concordance with those from RADTSE-CFA and both methods yield T2 estimates that are not significantly different from the reference single-echo spin-echo method.

Table 1:

T2 estimation on physical phantom comparing the RADTSE-VFA to RADTSE-CFA (using a train of 150° refocusing pulses).

RADTSE-VFA		RADTSE-CFA (150°)		Single-echo spin-echo
Estimated T2 (ms)	Relative error	Estimated T2 (ms)	Relative error	Reference T2 (ms)
131.4 ± 3.9	3.4%	132.5 ± 3.8	4.2%	127.1 ± 2.4
75.2 ± 1.7	2.4%	75.5 ± 2.2	2.7%	73.5 ± 0.9
57.6 ± 2.8	1.5%	58.3 ± 4.1	2.7%	56.8 ± 0.6
38.9 ± 1.6	3.6%	38.6 ± 1.4	2.6%	37.6 ± 0.5

RADTSE-VFA: Variable Flip Angle Radial Turbo Spin Echo

RADTSE-CFA: Constant Flip Angle Radial Turbo Spin Echo

Figures 2A and 2B show the decay curves for RADTSE-VFA and RADTSE-CFA, respectively, for phantoms vials with T2=38 ms (representing normal liver) and T2=74 ms (representing malignancies). Note that the VFA and the constant FA schemes provide a similar relative contrast at TE_eff = 90ms (the contrast typically used by radiologists to detect focal liver lesions). The contrast afforded by the VFA scheme for all phantoms relative to the phantom with T2=38 ms was not significantly different than the CFA scheme (p=0.10).

Figure 2:

Comparison of relative contrast between the (A) variable flip angle (VFA) and (B) constant flip angle (CFA) schemes using phantom data with T2 values of 38 ms (representing normal liver) and 74 ms (representing malignancies). Dotted blue line indicates echo number corresponding to TE_eff = 90 ms which was calculated based on the definition in (29). (C) Comparison of relative contrast at TE_eff = 90 ms using phantom vials with T2 = 57, 74, and 127 ms versus the T2=38 ms vial (reference).

Figures 3A and 3B show the signal evolution for the T2=38 ms and T2=74 ms phantom vials for the VFA and CFA schemes. The plots show higher signal intensities for the VFA scheme along the decay curve. As a result, the AUDC (used here as a surrogate measure of SNR for T2 estimation) is larger for the VFA scheme as shown in Figure 3C for all the phantoms.

Figure 3:

Comparison of the signal decay between the variable flip angle (VFA) and constant flip angle (CFA) schemes for data acquired on phantoms with (A) T2=38 ms and (B) T2=74 ms showing the higher signal intensity of the VFA scheme throughout the echo train. This results in a larger Area Under the Decay Curve (AUDC) for the VFA scheme as shown in (C) for all the phantoms.

Results of Monte Carlo simulations of the acquired phantom data are shown in Figure 4 for no added noise (i.e., noise standard deviation, σ_n=0) and added noise with σ_n=0.3 and σ_n=0.5. As shown in Figure 4A, adding noise increases the signal intensity in the late echoes. This is more accentuated for the CFA data and for species with shorter T2 values (T2=38 ms and T2=56 ms). This is due to the signal in the late echoes dropping below the noise level and thus, noise rather than signal is being fitted. As a result, T2 is overestimated with a more pronounced effect for the CFA scheme and for short T2 species as indicated in the Figure 4B. The VFA scheme is less affected by this effect due to the higher signal in the late echoes.

Figure 4:

Effect of noise on T2 estimation. (A) Signal decay curves comparing physical phantom data without the addition of extra noise (σ_n= 0) to data where noise was added corresponding to standard deviation (σ_n) of 0.3 and 0.5. (B) T2 estimates and relative error with respect to σ_n= 0. Inset shows enlargement of signal along the Y axis from latter part of the decay curve (echoes 25 – 32).

In vivo Results

Figures 5–6 compare RADTSE-VFA to the CFA counterpart for subjects with different abdominal pathologies. The figures show T2w images at TE_eff = 50 ms, 90 ms, 130 ms and the corresponding T2 maps. Figure 5 shows images for two subjects with focal liver lesions, a hemangioma (dashed arrow in Figure 5A) and liver metastases (dashed arrows Figure 5B). Both VFA and CFA schemes demonstrate that the T2 maps show the expected higher T2 for the hemangioma (a benign lesion) compared to the metastatic lesions and that both types of lesions have higher T2 values than adjacent liver (5,6,10,12,14,20). Note that the liver T2 is higher in the CFA T2 map compared to the VFA, which is more noticeable in the central part of the liver compared to the VFA (solid arrows) where the coil sensitivity is lower. The higher T2 values from RADTSE-CFA are most likely due to noise, consistent with the results presented in Figure 4.

Figure 5:

Abdominal T2-weighted images for different TEs and corresponding T2 maps obtained from data acquired with the RADTSE-VFA and RADTSE-CFA pulse sequences for (A) a subject with a benign hepatic hemangioma (dashed arrow) and (B) a subject with liver metastases (dashed arrows). Note that the T2 of liver is higher in the CFA T2 map as indicated in the central part of the normal liver (solid arrows in A and B) compared to the VFA counterpart.

Figure 6:

T2-weighted images for different TEs and T2 maps from data acquired with the RADTSE-VFA and RADTSE pulse sequences for (A) a subject with hepatocellular carcinoma (dashed arrows), and (B) a subject with mixed sclerosing cholangitis and central cholangiocarcinoma (tumor not shown). T2 maps for both subjects show regions of severe fibrosis distinctly as areas with elevated T2 values (black arrows in A and B) and areas of less severe fibrosis (solid white arrows in B).

Figure 6 shows images for two subjects with advanced chronic liver disease (CLD) and hepatic fibrosis (HF). According to the radiological report, the subject in Figure 6A has hepatocellular carcinoma (HCC) as indicated by the dashed arrows and a common pattern of CLD and HF involving the liver diffusely but with more advanced disease in the peripheral mid-liver region. The subject in Figure 6B has mixed sclerosing cholangitis and central cholangiocarcinoma (tumor not shown). This subject has regions of highly progressed focal confluent fibrosis. T2 maps for both subjects show regions of advanced severe fibrosis distinctly as areas with higher T2 values (black arrows in A and B) compared to areas of less severe fibrosis (solid white arrows in B). In addition, the subject in Figure 6B has an enlarged spleen, which is a characteristic feature of chronic advanced portal hypertension that results from HF. Note that the T2 values for the spleen for VFA and CFA RADTSE for the subjects in Figure 6A (T2_VFA=111.7±7.1 ms; T2_CFA=113.0±6.2 ms) and Figure 6B (T2_VFA=128.2±11.2 ms; T2_CFA=117.3±9.3 ms) are higher than the subject in Figure 5A (T2_VFA=94.5±4.6 ms; T2_CFA=97.4±4.7 ms) who does not have CLD and has a normal spleen. All subjects with CLD in this study (n=7) had high spleen T2 values ranging from 107.2–128.2 ms for RADTSE-VFA and 105.8–126.8 for RADTSE-CFA.

The AUDC maps corresponding to the data from Figures 5 and 6 are shown in Figure S1 included as Supplementary Material. As expected and consistent with our phantom results (Figure 3), AUDCs are larger for RADTSE-VFA.

A summary of in vivo data acquired with the RADTSE-VFA and RADTSE-CFA is presented in Table 2. No statistically significant differences between the two methods were found for (i) the mean spleen-to-liver and mean kidney (medulla)-to-liver relative contrasts at TE_eff = 90 ms and (ii) the mean T2 values for spleen and kidney medulla. On the other hand, (iii) the mean liver T2 was significantly higher for RADTSE-CFA compared to RADTSE-VFA, which is consistent with the results presented in Figure 4 showing that species with shorter T2s can be overestimated with RADTSE-CFA. Table 2 also shows (iv) the difference in SAR between the two methods. The mean SAR for RADTSE-VFA was significantly lower than the mean SAR for RADTSE-CFA.

Table 2:

Relative contrast, T2, and SAR values in vivo comparing RADTSE-VFA to RADTSE-CFA.

	RADTSE-VFA (mean± stdev)	RADTSE-CFA (150°) (mean± stdev)	p-value
Relative contrast (n=10)
Spleen-to-liver	0.58 ± 0.35	0.64 ± 0.32	0.62
Kidney-to-liver	1.10 ± 0.35	1.00 ± 0.27	0.49
T2, ms (n=10)
Liver	52.7 ± 6.6	60.4 ± 8.0	0.02*
Spleen	109.2 ± 12.3	110.7 ± 11.1	0.78
Kidney (medulla)	113.0 ± 8.7	114.0 ± 8.6	0.79
SAR (n=12)	0.75 ± 0.09	1.31 ± 0.15	<0.001*

RADTSE-VFA: Variable Flip Angle Radial Turbo Spin Echo

RADTSE-CFA: Constant Flip Angle Radial Turbo Spin Echo

A comparison of lesion characterization based on quantitative T2 measurements is shown in Figure 7 for 24 focal liver lesions. The T2 distributions show no overlap between malignant (22 metastases, 2 HCC) and benign (10 cysts, 1 hemangioma) lesions for both RADTSE-VFA and RADTSE-CFA methods. Mean T2 values and standard deviation for RADTSE-VFA are 89.3 ± 17.1 (malignancies) and 211.3 ± 55.4 ms (benign). Mean T2 and standard deviation for RADTSE-CFA are 89.2 ± 16.2 ms (malignancies) and 220.0 ± 53.4 ms (benign). No significant differences between the VFA and CFA methods were found for the malignant (p=1.0) and benign (p=0.39) T2 distributions.

Figure 7:

T2 distributions of malignant (22 metastases, 2 HCC) and benign (10 cysts, 1 hemangioma) lesions for RADTSE-VFA and RADTSE-CFA.

DISCUSSION

We have implemented and tested a variable refocusing flip angle radial TSE pulse sequence for abdominal T2 mapping and T2w imaging based on a long train of refocusing pulses. The refocusing flip angle scheme was designed to minimize the T2 estimation error and SAR while maximizing the relative contrast between liver and malignant lesions. Longer ETLs in TSE pulse sequences are more efficient for a given scan time (e.g. breath hold); they reduce the number of shots to acquire the desired number of k-space lines and increase the TR to include more slices. However, with longer ETLs the slice efficiency is limited by RF power deposition. The refocusing flip angle design used in this work reduced the SAR associated with the long echo train (ETL=32) thereby increasing the overall slice efficiency compared to a constant flip angle refocusing scheme by 1.5 (e.g., 11 slices with RADTSE-VFA vs 7 slice with RADTSE-CFA in an 18 second breath hold at 1.5T). Data for each slice cover 32 TEs spaced by ~ 7ms providing high temporal sampling for T2 encoding allowing for a better representation of the signal evolution. In fact, our implementation of the signal model accounts for the effect of indirect echoes from RF refocusing pulses that are < 180°.

In addition to the improved slice efficiency and T2 encoding, the VFA design is more robust to noise in terms of T2 estimation compared to the constant flip angle counterpart. As shown in the phantom experiments, adding noise to the data causes overestimation of the T2 values due to the signal in the late echoes falling below the noise level and thus, noise rather than signal being fitted. This effect is more pronounced for the CFA scheme and for species with short T2s. The VFA scheme is less affected by this effect due to the higher signal in the late echoes. In vivo experiments were consistent with phantom results showing a significantly higher mean T2 value for liver with CFA whereas organs with higher T2s, such as kidney and spleen, have similar mean T2s for VFA and CFA schemes.

Other techniques have been used for T2 mapping of the abdomen. Recent reports on abdominal T2 mapping used the T2-prep bSSFP pulse sequence which has limited temporal resolution (3 TE points) and acquires only once slice per breath hold (13). The limited number of TE points does not allow full modeling of the signal evolution and the limited number of slices makes it impractical for applications where good coverage of the abdomen is needed. A technique based on an accelerated Cartesian fast spin-echo pulse sequence has also been presented recently for abdominal T2 mapping (11). This technique yields adequate temporal resolution and slice coverage but scan times exceed a comfortable breath hold duration, requiring respiratory triggering and making the technique more sensitive to motion. In a recent report (8), the technique was used in combination with prospective respiratory triggering with scan times ranging from 2 min to 6 min (depending on the patient’s breathing rhythm and trigger efficiency) for 16 TEs and 15 slices. In that study, the effect of peristaltic motion was mitigated by using an injection of scopolamine.

The characterization of tissue via quantitative imaging is a growing area and T2 mapping of the abdomen could bring new ways of assessing disease should the technique be practical for clinical applications. For instance, the characterization of abdominal lesions can be objective and less dependent on radiological interpretation. T2 mapping can also provide a new way to assess changes in the liver and abdominal organs due to chronic liver disease, where an increase in T2 has been associated with hepatic fibrosis both in humans and rodents (38). Assessing changes in tumor characteristics in response to treatment is another potential application, an area that has been explored previously with diffusion-weighted MRI (12,39). T2 mapping with the proposed RADTSE-VFA technique could be complementary to diffusion-weighted imaging and may provide better information for small tumors due to its higher spatial resolution and insensitivity to motion and magnetic susceptibility effects.

Limitations

The VFA scheme used in this study was limited to ETL=32. Longer ETLs could further improve slice coverage and future studies should be explored to assess the best compromise between slice coverage and T2 mapping.

In this study, the VFA scheme was designed based on an optimally defined contrast between species with T2=80 ms (representing malignant lesions) and T2=40 ms (representing normal liver) at TE_eff = 90 ms. This was chosen based on the images typically used in radiological practice for detecting malignancies in the liver. For other applications, a different objective function may need to be derived a priori.

RADTSE-VFA was demonstrated here only for breath-held abdominal imaging not taking full advantage of the motion insensitivity of radial sampling. Extending RADTSE-VFA for free-breathing is a future extension of the technique. We expect the efficiency improvements afforded by the flip angle design to translate to free breathing imaging.

Parameter estimation presented in this work did not account for magnetization transfer effects arising from the multi-slice TSE acquisition. These effects have been described recently by Malik et al (40), who also proposed an EPG based analytical model for magnetization transfer exchange. Study of magnetization transfer in the presence of variable refocusing flip angles will be a topic for future work.

The current reconstruction scheme is based on an iterative algorithm and, as implemented for this study, is not practical for clinical applications due to computation time limitations. It should be noted that newer scanner models have the computational power for handling iterative reconstructions within clinically acceptable times. Furthermore, a convolutional neural network-based approach has been recently proposed to reconstruct data from radial TSE acquisitions (23). The technique was shown to generate T2-weighted images and T2 maps with comparable image quality and accuracy as iterative algorithms with significantly reduced reconstruction times (~20 sec per slice). Thus, computational power and/or deep learning based reconstruction approaches should enable the translation of RADTSE-VFA to the clinic.

The current study was focused on the technology and testing was performed in a limited number of subjects. A larger patient study needs to be undertaken to fully assess the technique including a systematic subjective evaluation by experts which was not performed in this study.

Conclusion

A variable refocusing flip angle design framework was introduced for slice efficient breath-held abdominal T2-weighted imaging and T2 mapping with a radial turbo spin echo technique. The proposed variable flip angle design reduces T2 estimation error while minimizing SAR, improving T2 mapping and slice efficiency compared to a constant flip angle scheme. Phantom and in vivo results show that the proposed technique provides noise robust T2 estimation compared to the constant flip angle radial TSE while generating comparable T2-weighted contrast. Further, the utility of the technique for T2 based characterization of common abdominal neoplasms has also been demonstrated.

Supplementary Material

supinfo

Figure S1: Area under the decay curve (AUDC) maps corresponding, from left to right, to the subjects in Figures 5A, 5B, 6A, and 6B, respectively. The higher AUDC values for the VFA data are due to the slower T2 signal decay of the VFA scheme compared to the CFA counterpart.