Optimization of Look-Locker Turbo-Field Echo-Planar Imaging and Evaluation of Its Accuracy in Head and Neck 3D T1 Mapping

Purpose: We present a sequence for T1 relaxation-time mapping that enables a rapid and accurate measuring. The sequence is based on the Look-Locker method by employing turbo-field echo-planar imaging (TFEPI) acquisitions and time to free relaxation after constant application of the radiofrequency (RF) pulses. We optimized the sequence, and then evaluated the accuracy of the method in imaging of head and neck. Materials and Methods: The method was implemented on a standard clinical scanner, and the accuracy of the T1 value was evaluated against that with the two-dimensional (2D) inversion recovery method. Results: The percentage errors of the T1 value, as validated by phantom imaging measurements, were 3.1% for slow-relaxing compartments (T1 = 2736 msec) and 1.1% for fast-relaxing compartments (T1 = 264.2 msec). Conclusion: We demonstrated a fast 3D sequence to obtain multiple slices, based on the Look-Locker method for T1 measurement, which provided a rapid and accurate way of measuring the spin-lattice relaxation time. An acquisition time of approximately 5 min was achieved for T1 mapping; in principle, this can provide head and neck coverage with 15 slices.


Introduction
Several of the new magnetic resonance (MR) applications require quantitative measurement of the spin-lattice relaxation time (T 1 ) in three dimensions (3Ds) with relatively short acquisition times. For example, quantitative tracer kinetic studies, in which vascular parameters such as blood volume and capillary permeability are calculated from dynamic contrast-enhanced MR data, require fast and accurate measurement of tissue T 1 values. Before application of a tracer kinetic model, tissue enhancement following contrast agent administration must be converted into contrast agent concentration, and it can be shown that this calibration depends strongly on the pre-contrast tissue T 1 value. [1][2][3][4][5] Ideally, tracer kinetic studies should be done in 3D for complete characterization of a lesion, or to locate small lesions not apparent in pre-contrast scans, or to examine multiple lesions in one study. Moreover, tissue T 1 measurement in the head and neck should be done in less than 5 min for studies to have clinically realistic acquisition times.
One of the major problems encountered in making accurate T 1 maps is the long imaging time required. For good accuracy over a wide range of T 1 values, multiple points on the T 1 recovery curve must be sampled. If a conventional two-dimensional (2D) inversion recovery sequence is used, data acquisition for each slice can take a few hours. Several schemes have been developed for rapid T 1 mapping in 2D in which multiple points on the recovery curve are sampled. 6 These techniques include methods based on the work of Look-Locker, [7][8][9][10][11][12][13][14][15][16][17][18][19] snapshot-fast low angle shot (snapshot-FLASH), [20][21][22][23] a mixed sequence method, 24 and a variable flip angle (VFA) method. 25,26 The major advantages of using a sequence based on the Look-Locker method and employing turbo-field echo-planar imaging (TFEPI) acquisitions are that the acquisition time is short and the T 1 relaxation behavior has been well-characterized. TFEPI combines turbo-field echo (TFE) and echo-planar imaging (EPI). A theoretical analysis has shown that if the Look-Locker technique is used, rather than the more time-consuming conventional inversion recovery method, T 1 values can be measured quickly with no penalty to the signalto-noise ratio of the calculated T 1 map. Other authors have demonstrated a high degree of accuracy and precision achievable experimentally with this technique. However, it is difficult to apply the Look-Locker technique of multiple slices clinically, because this technique is a continuous application of the radiofrequency (RF) pulses used for imaging, and if inversion recovery pulse interval is same, sampling time becomes insufficient.
Here, we introduce a fast 3D technique for rapidly acquiring data of multiple slices for accurate T 1 mapping. This technique is based on the principles of the 2D Look-Locker T 1 measurement scheme, and employs TFEPI acquisitions and time to free relaxation after constant application of the RF pulses. The acquisition time needed for volumetric T 1 mapping has been shortened considerably by segmenting the acquisition of the k y phase encode lines. This technique is similar to that which might be obtained by modifying snapshot-FLASH based T 1 measurement schemes for 3D. However, there is no delay between acquisition of successive volumes and T 1 relaxation is governed by the principles introduced by Look and Locker. We optimized the sequence on the basis of the Look-Locker method, and then evaluated the accuracy of the method in imaging of head and neck.

Theory
In this method, the relaxation process is influenced by constant application of the RF pulses. The effective longitudinal relaxation is determined by the effective longitudinal relaxation time T 1 * , which is smaller than T 1 . The longitudinal magnetization M(t) approaches a saturation value, M 0 * , which is smaller than the equilibrium value M 0 (Fig. 1). Thus, after inversion of the completely relaxed spin system, the relaxation process is described by the formula The effective longitudinal relaxation time T 1 * is given by where α is the flip angle.
This method is applicable only if the condition TR < T 1 * holds. Thus, Eq. (3) may be simplified to For the evaluation of T 1 a three-parameter fit of the image signal intensities is performed pixelwise according to the equation Comparison with the above equations yields Thus, T 1 may be calculated directly from the fit parameters The results of the pixelwise calculation can be displayed in a quantitative T 1 map. Of note, the method does not require knowledge of the flip angle α.

Materials and Methods
We tested the method on a phantom and a human in a 1.5-T whole-body scanner (Intera Achieva 1.5-T Nova, Philips, The Netherlands) with a maximum gradient strength of 66 mT/m and a gradient slew rate of 160 mT/m/msec using a Philips 8-channel SENSE head coil.
We measured T 1 values using magnitude image. The maps of T 1 * , M 0 * , and M 0 were obtained by fitting the image signal intensities on pixelwise basis to Eq. (5) using the nonlinear least-squares method by MATLAB R2014b (MathWorks, Natick, MA, USA). Moreover, T 1 map was calculated by using Eqs. (6) and (7) with maps of T 1 * , M 0 * , and M 0 .

Phantom study
To validate the T 1 values resulting from the sequence to be based on the Look-Locker method as described above, we applied standard inversion recovery method to a multi-compartment phantom. The phantom consists of six cylindrical sample bottles filled with water and different concentrations of gadoteridol (Gd-HP-DO3A, ProHance, Eisai Co. Ltd., Tokyo, Japan). The gadoteridol concentrations are 0.05, 0.1, 0.2, 0.5, and 1.0 mmol/L, respectively.
The 2D inversion recovery sequence is a simple acquisition with an inversion recovery pulse before an excitation pulse. The imaging parameters for acquisition of a single image are as follows and shown in Table 1: repetition time (TR) 15000 msec; echo time (TE) 20 msec; field of view (FOV) 230 × 196 mm; acquisition matrix 256 × 218; acquisition pixel size 0.9 × 0.9 mm; recon matrix 256 × 218; recon pixel size 0.9 × 0.9 mm; 1 slice with a thickness of 5 mm; band width 64.3 Hz; and sampling points at 50, 100, 200, 500, 1000, 2000, 3000, and 5000 msec. The acquisition time for this sequence is 7 h 14 min. Each measurement was performed three times, and the averaged values were used for the comparison.
To verify that we had properly chosen the EPI factor which is the number of k-space profiles collected per excitation, TFE factor which is the number of k-space profiles collected per sampling point, and recovery period (t r ) which is time to free relaxation after constant application of the RF pulses, for the phantom measurements using the Look-Locker sequence, the experiments were repeated with all parameters held constant except for EPI factor, TFE factor, and t r . Three values of EPI factor (1, 3, and 11) were chosen. Because each acquisition time was the same time, three values of TFE factor (33, 11, and 3) were chosen. Furthermore, at the end of the train of α-pulses, an undisturbed recovery period was optionally inserted to allow the recovery of longitudinal magnetization before the next inversion pulse. Two values of t r (3136 and 4993 msec) were chosen.
The Look-Locker sequence was used to obtain images to perform fitting of the T 1 relaxation of the doped water in the different bottles. The measurement parameters are as follows and shown in Table 1: TR shortest (6.5, 11, and 22 msec); TE shortest (3.2, 4.8, and 11 msec); FOV 230 × 196 mm; acquisition matrix 192 × 127; acquisition pixel size 1.2 × 1.54 mm; recon matrix 256 × 218; recon pixel size 0.9 × 0.9 mm; 15 slices with a thickness of 5 mm; band width 229.2, 151.2, and 54.7 Hz; flip angle 10°; inversion recovery pulse interval 7000 msec; and sampling points is shown in Table 1. The acquisition time for this sequence is 5 min 3 sec. To demonstrate the accuracy of the T 1 values obtained by using the Look-Locker method, we compared the resultant T 1 values with those obtained from 2D inversion recovery method measurements and calculated the percentage errors. In each bottle a circular region of interest (ROI) comprising 15 × 15 pixels was used to calculate a mean T 1 value.

Volunteer study
We performed a study on a single healthy volunteer by using the 2D turbo inversion recovery and optimized 3D Look-Locker methods after obtaining their informed consent as required by our institutional review board. The 2D turbo inversion recovery 3D     The 3D Look-Locker sequence parameters that we chose were those optimized in the phantom study. We compared the T 1 values of the cerebrospinal fluid (CSF), sternocleidomastoid muscle (SCM), and parotid gland (PG) of the healthy volunteer with those obtained by using 2D turbo inversion recovery measurements and calculated the percentage errors.

Results
We calculated T 1 maps of the multi-compartment phantom by using data acquired with the Look-Locker method (Fig. 2). Tables 2 and 3 give the T 1 values ( Table  2) and percentage errors (Table 3) for the phantom measurements resulting from the 2D inversion recovery, 2D turbo inversion recovery, and 3D Look-Locker measurements. The 2D inversion recovery measurements served as a reference for all other experiments. The locations of "bottle 1," "bottle 2," "bottle 3," "bottle 4," "bottle 5,"and "bottle 6" correspond to water and the nominal gadoteridol concentrations of 0.05, 0.1, 0.2, 0.5, and 1.0 mmol/L, respectively (Fig. 2). There was good agreement between the T 1 values from the 2D inversion recovery measurements and those from the 2D turbo inversion recovery measurements.
With t r = 3136 msec, there was not good agreement between the 2D inversion recovery measurements and the 3D Look-Locker measurements in water and gadoteridol at the nominal concentrations of 0.05 mmol/L. With EPI factor = 1, TFE factor = 33, and EPI factor = 11, TFE factor = 3, there was not agreement in gadoteridol at the nominal concentrations of 0.5 and 1.0 mmol/L. However, with EPI factor = 3, TFE factor = 11, and t r = 4993 msec, there was good agreement. These parameters gave the best agreement. We then compared the T 1 values for the phantom measurements sequence parameters for acquisition of a single image are as follows and shown in  resulting from the 2D inversion recovery, the 2D turbo inversion recovery, and the optimized 3D Look-Locker measurements (Fig. 3). A high correlation was observed between the results obtained with the three methods.
We next calculated the T 1 map of a healthy volunteer from data obtained by using the optimized 3D Look-Locker method (Fig. 4). T 1 measurements made by using the 2D turbo inversion recovery and optimized 3D Look-Locker sequences are compared  in Table 4 and Fig. 5 for ROIs in CSF, sternocleidomastoid muscle, and parotid gland. There was a good correlation between the T 1 measurements made by using the two methods.

Discussion
There is great interest in fast T 1 mapping sequences, particularly for the diagnosis of different diseases, 27 MR temperature monitoring, 28,29 studies of intra-and extracellular water discrimination, 30 and quantification of regional blood flow. 31 All these measurements require highly accurate T 1 values obtained with clinically acceptable acquisition times and with high in-planar resolution. The snapshot-FLASH sequence provides accurate and precise T 1 mapping with  This method calculates value from two spoiled gradient echo volumes acquired at two different flip angles. The advantages of the VFA method are that it is easier to implement and mapping can be calculated more rapidly, as a linear regression can be used for the fit. On the other hand, for the small TR necessary for short 3D acquisition times, it is impossible to properly optimize the choice of the two flip angles to allow accurate T 1 measurement over a wide range of T 1 values. When applied to 3D data, using a short TR, the VFA method suffers from poor accuracy and precision. Brookes et al. found that they could measure T 1 values accurately only for T 1 < 900 msec. 32 In general, sensitivity to pulse sequence settings, such as flip angle or inversion time, is a weakness of any 2-point T 1 measurement technique. The Look-Locker sequence has made it possible to measure T 1 values in a 3D volume in approximately 5 min with less than 3.1% error, in the case of T 1 values between 264.2 and 2736 msec. We have found the performance of the sequence to be relatively sensitive to pulse sequence parameters. We recommend using t r = 4993 msec, EPI factor = 3, and TFE factor = 11 for optimum accuracy of the T 1 measurements. In the case of fast-relaxing compartments, the accuracy of T 1 values declined using TFE factor = 33, because sampling intervals of data acquisitions after inversion recovery pulse are long for fast-relaxing compartments, and the accuracy declined using EPI factor = 11, because TR (22 msec) is long for short T 1 * compartments. Moreover, EPI is highly sensitive to static magnetic field inhomogeneities, and increases chemical shift artifact (Fig. 6). This is caused by the accumulation of a phase shift. The EPI factor is one of the causes of this accumulation. 33 In the imaging of high in-planar resolution and with useful acquisition times. 15 However, for wide-range coverage of T 1 mapping, snapshot-FLASH acquisition must be repeated, and the high in-planar resolution requirement reduces the number of time points sampled on the recovery curve. An alternative approach employs a mixed sequence to calculate images of ρ, T 1 , and T 2 by MR imaging, although a long imaging time is Vol. 15 No. 3,2016 phantom study, there was ghost artifact (Fig. 2). This is also dependent on EPI factor, but this did not have a serious problem of imaging to measure T 1 values in the imaging of volunteer study (Fig. 4). Although, T 1 values of parotid gland in the volunteer study were shorter than those of previous study. The reason may be that errors due to the examination of a single volunteer, imaging distortion due to the EPI factor, and changes in the signal intensity due to the phase difference between water and fat by the setting of TE had an influence on the resultant T 1 values. In the case of slow-relaxing compartments, the accuracy declined using t r = 3136 msec, because the longitudinal magnetization M(t) did not sufficiently recover to the equilibrium value M 0 . The longitudinal magnetization M(t) should recover not to the saturation value M 0 * but to the equilibrium value M 0, because fitting of the recovery curve is done by using Eq. (1). In the case of optimized parameters, we thought the longitudinal magnetization M(t) of slow-relaxing compartments recovers to the equilibrium value M 0 sufficiently, because T 1 * value which is changed by scan parameters was very short (T 1 * value of water = 541.2 msec), and time (t r ) to free relaxation after constant application of the RF pulses was long enough. Therefore, the accurate T 1 mapping obtained by using the 3D Look-Locker method needs optimal sampling points and a long enough time (t r ) to free relaxation. Moreover, the degree of local imaging distortion, chemical shift artifact, and ghost artifact is reduced sufficiently by using EPI factor = 3 in the imaging of volunteer study (Figs. 4,7). In this study, we did not examine errors due to flip angle and RF inhomogeneities in detail because that these errors do not affect T 1 values is implicated by the theoretical formula, and the difference of T 1 values in each slice was less than 1.1% for bottle phantom (T 1 = 396.6 msec). Because accuracy, scan time, and artifact are changed by scan parameters, these parameters should be chosen to suit the clinical situation.

Conclusion
A fast 3D sequence to obtain multiple slices, based on the Look-Locker method for T 1 measurement, provided a rapid and accurate way of measuring the spin-lattice relaxation time. The percentage errors of the T 1 values validated by phantom imaging measurements   6. The imaging of a healthy volunteer obtained by using the optimized three-dimensional (3D) Look-Locker sequence and the following parameters: repetition time (TR) 22 msec; echo time (TE) 11 msec; field of view (FOV) 230 × 196 mm; acquisition matrix 192 × 127; acquisition pixel size 1.2 × 1.54 mm; recon matrix 256 × 218; recon pixel size 0.9 × 0.9 mm; 15 slices with a thickness of 5 mm; echo-planar imaging (EPI) factor 11; turbo-field echo (TFE) factor 3; band width 54.7 Hz; flip angle 10°; inversion recovery (IR) pulse interval 7000 msec; recovery period (t r ) 4993 msec; sampling points at 1908 msec. Chemical shift artifact was caused significantly. were 3.1% for slow-relaxing compartments (water, T 1 = 2736 msec) and 1.1% for fast-relaxing compartments (Gd-1.0 mmol/L, T 1 = 264.2 msec). An acquisition time of approximately 5 min was achieved for T 1 mapping; in principle, this can provide head and neck coverage with 15 slices.