Dictionary-based T2-mapping with multi-echo turbo-spin echo

doi:10.21203/rs.3.rs-4528325/v1

Download PDF

Research Article

Dictionary-based T2-mapping with multi-echo turbo-spin echo

https://doi.org/10.21203/rs.3.rs-4528325/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Multi-echo turbo-spin echo (ME-TSE) is a pulse sequence commonly used for T2 mapping in MRI. As compared to other pulse sequences such as MESE, ME-TSE is largely faster. It has been previously shown that dictionary-based T2-mapping can be used to provide accurate T2 values from MESE datasets but the corresponding accuracy on ME-TSE datasets has never been assessed. In the present study, we aimed to investigate the impact of combining effective echo signals in a ME-TSE pulse sequence on the accuracy of T2 mapping using a dictionary-based reconstruction method.

We initially compared three different combinations of echo signals on phantom scans and identified the corresponding differences. We then determined the best combination among them. In the pulse sequence with an echo train length (ETL) of 3 and 3 contrasts, we found that the most accurate and homogeneous combination was achieved when the first echo was assigned to the center in the first contrast, the second echo in the second contrast, and the third echo in the third contrast.

Next, we compared the use of dictionaries generated for a single slice versus multiple slices (N = 5) in the reconstruction process for this specific combination. The dictionaries were generated using a Bloch simulator, taking into account the saturation effect between slices during dictionary generation for the multi-slice case.

Our results show that using single-slice dictionaries for reconstructing T2 phantom maps with no gap between slices can lead to variations in T2 values between slices. However, this variation can be reduced when using a multi-slice dictionary so that more accurate T2 values can be obtained.

While MRI is widely employed for clinical assessment of multiple diseases, quantitative MRI (qMRI) has been developed so as to offer biomarkers sensitive to disease severity and progression. Unlike qualitative MRI which provides details related to contrast, qMRI intends to map real physical or functional parameters of tissues, such as relaxation times [1], diffusion and perfusion coefficients [2], fat fraction [3], etc...

Among others, the spin-spin relaxation time of water protons - T2 has been used as a biomarker of oedema/inflammation. As part of a macroscopic approximation, T2 refers to the exponential attenuation of the transverse magnetization component [4]. T2 mapping methods have been widely used for the diagnosis of diseases associated with inflammation and hydration processes in tissues [5]. This biomarker can be mapped in vivo using qMRI. One of the most frequently used technique is based on acquisition of a set of images using multi-echo spin-echo (MESE) pulse sequence [6, 7]. This sequence can be used to follow the time evolution of the transverse component of the magnetization vector which has been described as an echo-modulated curve (EMC). However, acquisition time can be a critical factor in some applications which needs such breath-hold imaging (heart, abdomen, etc …). In that context, MESE-based technique are too long and accelerated pulse sequences which combine several echoes from the echo train using a multi-echo turbo-spin echo sequence (ME-TSE) are required [8‒10]. However, similarly to what has been described for MESE sequences, ME-TSE provide stimulated echo signals that contribute to the EMC. As previously described, EMCs strongly deviate from the expected exponential decay so that exponential approximation are providing incorrect estimates of T2 relaxation times [11]. In addition, in multi-slice scanning, direct saturation between neighboring slices can also bias T2 values [12]. These issues have been elegantly addressed using a dictionary-based method aiming at mapping T2 from multi-slice MESE images [13, 14].

In the present work, we aimed at creating an accurate and fast method so as to obtain multi-slice T2 maps from ME-TSE MR images. More specifically, we investigated how the choice of effective echo times could affect T2 accuracy when a dictionary-based fitting method was used. We have also compared T2 values among slices when dictionaries including a single-slice and multi-slices were used for the generation of T2 values from a multi-slice dataset.

Measurements were conducted at 3T (MAGNETOM Vida, Siemens Healthcare, Erlangen, Germany). All images were acquired for a set of 5 calibrated agarose-based phantoms (Diagnostic Sonar, Livingston, UK) with known T1 and T2 values (Table 1). Values were provided by the manufacturer with a 3% error. In order to double-check the T2 values, a “gold-standard” T2 single slice mapping was computed from MR images acquired using a spin echo pulse sequence with the following parameters: TR/TEs = 5.5 s / 10-30-50-100-150-200-300 ms, voxel size = 1.17x1.17x2 mm. In order to assess the effect of a multi-slice ME-TSE acquisition on the inter-slice T2 values homogeneity, single–slice and multi–slice images of the phantoms were obtained using an ME-TSE sequence with the following parameters: TR = 2 s, first TE = 10.3 ms, inter-echo time = 10.3 ms, turbo factor = 3, number of contrasts = 3, slice thickness = 2 mm, number of slices = 5, inter-slice gap = 0 and 100% (100% gap was used for reference T2 maps, as this gap provides negligible direct saturation between slices[13]).

Table 1. T1 and T2 values in phantoms
Phantom #	1	2	3	4	5
T2, ms	97.2	118.5	170.5	158.7	131.6
T1, ms	336.1	672.5	1235.2	1288.9	1407.7

For a ME-TSE acquisition, the k-space for each image is filled with data from different echo times and the number of such echoes is called turbo-factor. Figure 1 illustrates echo modulation curves of an ME-TSE sequence with a turbo factor = 3. The central part of k-space of the first image in a series (first contrast) is filled with the data from echo #1 (and thus, in this case, the first echo time is considered as an effective - TE_eff), the outer parts of this k-space are filled with data from the echo #2 and #3. The k-spaces for the remaining contrasts are filled in the same way i.e. with echoes #4–6 and #7–9 for contrasts 2 and 3, respectively.

The effective echo time is defined as the echo time when the k-space center of a specific contrast is acquired. A centric phase encoding scheme is used to maximize the signal intensity in the center of k-space. For the following simulations it was assumed that the image contrast was only determined by central k-space lines that were acquired at the effective echo time TE_eff). In our work, we used three variants of TE_eff: combinations: first-first-first (FFF), first-second-third (FST) and first-third-third (FTT).

The proposed T2-mapping pipeline is illustrated in Fig. 2. Experimental EMCs were initially obtained from the MR images. Then, dictionaries of echo modulation curves, containing 9 echo signals, were generated using the ME-TSE pulse sequence diagram with the relevant parameters and taking into account the inter-slice interactions as previously described [13].

The magnetization time-dependent changes throughout the ME-TSE pulse sequence was numerically simulated in Matlab (The MathWorks, Natick, 2016). Given that the existing tool for EMC dictionary generation [14] did not allow to model the behavior of magnetization in a multi-slice mode throughout the ME-TSE pulse sequence, we modified an initial script version designed to turbo spin echo response modeling [15]. As previously described, the modified script intended to solve Bloch equations in a matrix form using rotation and relaxation matrices [4, 16]. The parameters and shapes of radio frequency and gradient pulses were included in order to take into account the characteristics of the actual ME-TSE sequence provided by the vendor. The script was modified so that the multi-slice acquisition was taken into account, with an interleaved slice order (1-3-5-2-4) similarly to the experiment. This code allowed us to generate EMC dictionaries in both single-slice (without slice cross-talk, as originally proposed [8]) and multi-slice modes. Parameters of the dictionaries were as follows: B₁⁺ = [80:1:120] %; T2 = [50:1:210] ms.

After that, experimental EMC curves were matched with the single-slice and multi-slice dictionaries. Similarly to what we proposed in our seminal paper [13, 14], the matching criteria were based on the minimal L2-norm difference between the experimental curve and curve from dictionary. Each experimental ME-TSE EMC contained only 3 points. Thus, they were matched only with the corresponding points in the generated EMCs. Multi-slice T2 maps based on zero gap ME-TSE were obtained for each phantom: when multi-slice data were fitted with the single-slice dictionary (MS), and when multi-slice data were fitted with the multi-slice dictionary (MM).

Figure 2 displays T2 maps of phantom #4 (T2ref = 158.73 ms) generated by aligning experimental data (gap 100%) with the single-slice dictionary. The results are presented for three combinations of effective echo times: FFF, FST, FTT. A 100% gap between slices was chosen for this comparison to minimize the saturation effect and ensure uniformity of T2 values across slices during dictionary-based fitting.

The worst fitting result was observed for the FTT case: T2 maps were inhomogeneous and biased compared to T2_ref. T2 values inside the phantoms were much better in terms of uniformity of the T2 map for the FFF combination. The most homogeneous maps with the smallest bias in T2 estimation were obtained for the FST case. On that basis, only FST data were further used to analyze the effect of the multi-slice dictionary on inter-slice T2 variation. As expected, when scanning with a 100% gap, no inter-slice T2 variation was observed.

Figure 4-A illustrates T2 maps of the 4th phantom for 5 slices scanned using the Teff combination FST and fitted using multi-slice and single-slice dictionaries. Visually, one could observe that multi-slice fitting was characterized by T2 variation between slices, whereas the use of a multi-slice dictionary made T2 more homogeneous between slices. In order to numerically estimate the T2 variation between the slices, coefficients of variation were calculated (Figure 4-B). With a 100% slice gap, the coefficients of variation were lower than 1.5% for all phantoms. Decreasing the gap between slices to 0 but fitting with a single-slice dictionary increased the mean T2 variations between slices to 3.14% (ranging from 0.98 to 4.70%). Additionally, a strong dependence on T1_ref was observed i.e. the coefficient of variation increased with T1_ref. The coefficient of variation (1.21% mean value ranging from 0.80 to 1.66%) was strongly reduced when using a multi-slice dictionary approach.

Distributions of the measured T2 values within 5 slices for all phantoms are presented in Figure 3. Thin dashed horizontal lines indicated reference T2 values. For a 100% slice gap, calculated T2 values were homogeneous between slices, as expected. When the data obtained with a zero gap were fitted with single-slice dictionaries, the largest and lowest T2 values were consistently obtained for slices #1 or #5 (edge) and slices #2-4 (central), respectively. A visible improvement in inter-slice reproducibility was clearly observed in the majority of cases for MM. Relative T2 errors with respect to the reference T2 values are depicted in Figure 5-B as a function of T1 values. In the MS case, an increased error for higher T1 values was observed. These errors decreased in the MM case.

In the present study, an improved EMC-dictionary method for ME-TSE-based T2 mapping has been proposed. The combination of effective echo times (TE_eff) that provided the lowest T2 estimation error was identified. The use of the improved EMC method, which considers direct saturation effects for T2 maps reconstruction from multi-slice data, resulted in a reduction of both T2 variation between slices and relative error of T2 measurements.

The comparison of different combinations of effective echo times (FFF, FST, FTT) indicated that the FST combination provided the most homogeneous T2 maps with the smallest bias in T2 estimation. A previous work has already considered the k-space filling trajectory in T2 maps reconstruction and suggested that phase encoding profile effects should be considered for T2 estimations. However, the authors did not compare different combinations and only presented results for 2 contrasts [17]. We have to acknowledge that many other combinations could have been used i.e. increasing the number of contrasts and the number of ETL. However, in the first contrast, the first echo in k-space should always remain unchanged because the whole set of curves was normalized to the first value. In the present study, we clearly showed that the choice of effective echo time combination was important when fitting with dictionaries. One could hypothesize that for a seemingly logical method like FFF, this combination might not necessarily be the best solution. Future studies could consider the k-space filling trajectory when generating dictionaries in order to better understand why this difference occurred for different combinations.

In addition, we assessed the influence of multi-slice mode on T2 variation among slices. Such a variation has been previously investigated for T2 maps reconstructed from MESE images using dictionary fitting [13]. Accordingly, we showed that T2 inter-slice variation was small when 100% slice gaps were utilized (CV < 1%). When using a single-slice dictionary, the largest variation was observed with a small TR = 2s and zero slice gap. These parameters were also relevant for ME-TSE i.e. a short repetition time reduced the scanning time, which is an advantage of turbo spin-echo; a small slice gap reduced information loss and allowed the detecting of structures close to the slice thickness. For these reasons, experiments were conducted with these pulse sequence parameters. Using a single-slice dictionary, a significant coefficient of variation between slices, especially for phantoms with high T1 values (CV = 4.70% for phantom #5) was found. Multi-slice dictionary fitting provided a comparably small coefficient of variation (1.66%). A slightly better reduction was observed when increasing the slice gap up to 100% (0.80%).

Overall, the present results provided evidence for the advantages of employing multi-slice dictionaries in T2 mapping MRI using ME-TSE pulse sequence. These results are important for improving the reliability of qMRI techniques and may contribute to improving the accuracy and reproducibility of T2 mapping in various clinical applications.

Ethical Approval

Not applicable.

Funding

The work was supported by state assignment No. FSER-2022–0010 within the framework of the national project “Science and Universities”.

Author Contribution

Z.B. and K.B. planned experiments, conducted all analyses, performed post-processing, and wrote the main manuscript text. C.-A. M. wrote the main script for dictionary generation. S.R. and T.T. assisted with MRI scanning. D.B. supervised the research. All authors reviewed the manuscript.

S.C. Deoni, S.C. Williams, P. Jezzard, J. Suckling, D.G. Murphy, D.K. Jones, Neuroimage. (2008) https://doi.org/10.1016/j.neuroimage.2007.11.052
C. Federau, P. Maeder, K. O’Brien, P. Browaeys, R. Meuli, P. Hagmann, Radiology. (2012) https://doi.org/10.1148/radiol.12120584
T. J. Bray, M.D. Chouhan, S. Punwani, A. Bainbridge, M.A. Hall-Craggs, Brit J Radiol. (2017) https://doi.org/10.1259/bjr.20170344
R.W. Brown, Y.C. Cheng, E.M. Haacke, M.R. Thompson, R. Venkatesan, Magnetic resonance imaging: physical principles and sequence design. (John Wiley & Sons, 2014)
H.K. Kim, T. Laor, P.S. Horn, J.M. Racadio, B. Wong, B.J. Dardzinski, Radiology. (2010) https://doi.org/10.1148/radiol.10091547
P.J. van Houdt, H.K. Agarwal, L.D. van Buuren, S.W. Heijmink, S. Haack, H.G. van der Poel, U.A. van der Heid, Radiology. (2018) https://doi.org/10.1002/mrm.26816
Y. Fatemi, H. Danyali, M.S. Helfroush, H. Amiri, Magn Reson Med. (2020) https://doi.org/10.1002/mrm.28309
D.H. Hoffman, A. Ayoola, D. Nickel, F. Han, H. Chandarana, K.P. Shanbhogue, Abdom Radiol (2020). https://doi.org/10.1007/s00261-019-02382-9
B.M. Ellingson, A. Lai, H.N. Nguyen, P.L. Nghiemphu, W.B. Pope, T.F. Cloughesy, Clin Cancer Res. (2015) https://doi.org/10.1002/mrm.28309
L. Feng, R. Otazo, H. Jung, J.H. Jensen, J.C. Ye, D.K. Sodickson, D. Kim, Magn Reson Med. (2011) https://doi.org/10.1002/mrm.22756
S. Majumdar, S.C. Orphanoudakis, A. Gmitro, M. O'donnell, J.C. Gore, Magn Reson Med (1986). https://doi.org/10.1002/mrm.1910030305.
J.B. Kneeland, A. Shimakawa, F.W. Wehrli, Radiology. (1986) https://doi.org/10.1148/radiology.158.3.3945757
E.A. Brui, Z. Badrieva, C.A. de Mayenne, S. Rapacchi, T. Troalen, D. Bendahan, Magn Reson Med. (2024) https://doi.org/10.1002/mrm.29987
N. Ben‐Eliezer D.K. Sodickson K.T. Block, Magn Reson Med. (2015) https://doi.org/10.1002/mrm.25156
E.A. Brui, S. Rapacchi, D. Bendahan A.E. Andreychenko, Magn Reson Mater Phy. (2021) https://doi.org/10.1007/s10334-021-00937-w
Hargreaves B. Bloch Equation Simulator. http://mrsrl.stanford.edu/~brian/blochsim/. Accessed May 30, 2024.
S.W. Anderson, O. Sakai, J.A. Soto, H. Jara, Magn Reson Med. (2013) https://doi.org/10.1002/mrm.24213

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Dictionary-based T2-mapping with multi-echo turbo-spin echo

Status:

Version 1

Abstract

Figures

Introduction

Methods

Results

Discussion and conclusions

Declarations

Ethical Approval

Funding

Author Contribution

References

Additional Declarations

Status:

Version 1