Whole-Brain Myelin Imaging Using 3D Double-Echo Sliding Inversion Recovery Ultrashort Echo Time (DESIRE UTE) MRI

Abstract

Background

Signal contamination from long T2 water is a major challenge in direct imaging of myelin with MRI. Nulling of the unwanted long T2 signals can be achieved with an inversion recovery (IR) preparation pulse to null long T2 white matter within the brain. The remaining ultrashort T2 signal from myelin can be detected with an ultrashort echo time (UTE) sequence.

Purpose

To develop patient-specific whole-brain myelin imaging with a three-dimensional double-echo sliding inversion recovery (DESIRE) UTE sequence.

Materials and Methods

The DESIRE UTE sequence generates a series of IR images with different inversion times during a single scan. The optimal inversion time for nulling long T2 signal is determined by finding minimal signal on the second echo. Myelin images are generated by subtracting the second echo image from the first UTE image. To validate this method, a prospective study was performed in phantoms, cadaveric brain specimens, healthy volunteers, and patients with multiple sclerosis (MS). A total of 20 healthy volunteers (mean age, 40 years ± 13 [standard deviation], 10 women) and 20 patients with MS (mean age, 58 years ± 8; 15 women) who underwent MRI between November 2017 and February 2019 were prospectively included. Analysis of variance was performed to evaluate the signal difference between MS lesions and normal-appearing white matter in patients with MS.

Results

High signal intensity and corresponding T2* and T1 of the extracted myelin vesicles provided evidence for direct imaging of ultrashort-T2 myelin protons using the UTE sequence. Gadobenate dimeglumine phantoms with a wide range of T1 values were selectively suppressed with DESIRE UTE. In the ex vivo brain study of MS lesions, signal loss was observed in MS lesions and was conformed with histologic analysis. In the human study, there was a significant reduction in normalized signal intensity in MS lesions compared with that in normal-appearing white matter (0.19 ± 0.10 vs 0.76 ± 0.11, respectively; P < .001).

Conclusion

The double-echo sliding inversion recovery ultrashort echo time sequence can generate whole-brain myelin images specifically with a clinical 3-T scanner.

© RSNA, 2019

Online supplemental material is available for this article.

See also the editorial by Port in this issue.

Summary

Three-dimensional double-echo sliding inversion recovery ultrashort echo time MRI enabled robust direct imaging of myelin confirmed via phantom, ex vivo, and in vivo studies; obvious myelin signal loss was seen in multiple sclerosis lesions.

Key Results

• ■ Three-dimensional double-echo sliding inversion recovery ultrashort echo time MRI yielded robust long T2 white matter signal suppression by using a range of inversion times to null signal from tissue components with different T1 values, coupled with early detection of short T2 signals to enable direct visualization of myelin in the brain.

• ■ In ex vivo brain myelin imaging, signal loss was seen in multiple sclerosis (MS) lesions and was associated with myelin loss seen at histology.

• ■ In patients, a reduction in normalized myelin image signal intensity was seen in MS lesions when compared with normal-appearing white matter (0.19 vs 0.76, P < .001).

Introduction

Loss of myelin is a hallmark of numerous inflammatory and neurodegenerative disorders of the central nervous system, including multiple sclerosis (MS), in which characteristic foci of demyelination are seen. These are associated with a variety of neurologic and neuropsychological problems (1). Direct in vivo evaluation of myelin could be of major importance in the recognition and characterization of demyelination and subsequent remyelination, both spontaneously and in response to treatment in patients with MS or other diseases (2).

Although MRI is widely used in the noninvasive evaluation of MS (3,4), direct imaging of myelin with this technique has proven difficult due to the ultrashort T2 (T2 <1 msec) of myelin lipids and proteins (5–7). The usual clinical approach uses T1- or T2-weighted fast spin-echo or gradient-echo sequences, which only depict the longer T2 components in white and gray matter. The changes seen with these sequences are not specific to myelin, and there is substantial overlap between the MRI appearance of abnormalities in demyelinating diseases and those seen in other diseases of the central nervous system (2). Magnetization transfer imaging has been investigated as a method to show changes in myelin (8), but a recent large-scale multicenter study reported poor correlation between disability assessed with the expanded disability status scale and the magnetization transfer ratio of brain tissue (r = −0.16) in MS (9). In recent years, several other techniques have been developed to achieve more specific imaging of myelin in patients with MS, including multicomponent T2 analysis (10), multicomponent-driven equilibrium single-component observation of T1 and T2, or mcDESPOT, (11), and myelin-specific imaging (12). However, these techniques all use conventional sequences with echo times of several milliseconds or more. These can be used to detect signal from water associated with myelin that has T2* values of 5 msec or longer (13), but they cannot be used to detect significant signal from myelin protons that have much shorter T2* values that are less than 1 msec (5–7). Ultrashort echo time (UTE) MRI sequences with echo times 100–1000 times shorter than those of conventional sequences can be used to directly detect signal from myelin protons; thus, they could enable a more specific evaluation of the changes in myelin that occur in demyelination and remyelination (5–7).

Two-dimensional adiabatic inversion recovery UTE MRI sequences have been developed to enable direct detection of myelin protons in white matter with clinical MRI scanners (5,14–17). The inversion recovery preparation suppresses signal from the long T2 components in white matter (WM_L, mainly water) by inverting and nulling their longitudinal magnetization using an appropriate inversion time. The longitudinal magnetization of myelin is not inverted using conventional MRI due to its extremely short T2. Instead, it is partially saturated; however, it can be detected with subsequent use of UTE data acquisition. An accurate inversion time is critical for nulling the signal from WM_L because its signal may be 10 or more times greater than that from myelin. Slightly longer or shorter inversion times than the optimum for nulling may lead to water signal contamination and may compromise the accuracy and specificity of myelin imaging. However, the optimal nulling point (TI_null) for WM_L may be difficult to determine. The T1 of normal WM_L varies in different regions of the brain and between individuals (18,19). A wide range of T1 values (ie, 690–1100 msec) have been reported for white matter at 3 T (20,21). In addition, the T1 of WM_L may change in patients with overt disease and those with occult disease within normal-appearing white matter. T1 measurement is time consuming, which is a limitation in clinical applications. In addition, the commonly used two-dimensional inversion recovery UTE sequence is sensitive to eddy currents and suffers from out-of-slice signal excitation, especially in off-center slices. Furthermore, two-dimensional techniques of this type cannot easily be used to perform single-acquisition whole-brain imaging (15,22).

To address these limitations, we have developed a three-dimensional (3D) double-echo sliding inversion recovery (DESIRE) UTE sequence, which provides nulling that accommodates the differences in the T1 of long T2 white matter seen between human participants. In a single scan, this technique yields a series of images with a wide range of equally spaced inversion times covering all likely nulling points for WM_L due to normal variation with age or sex and disease. As a result, the 3D DESIRE UTE sequence can yield robust suppression of WM_L and therefore can enable more specific and accurate imaging of myelin. To validate this technique for whole-brain myelin imaging with a clinical 3-T scanner, we performed studies using phantoms, cadaveric brain specimens, healthy volunteers, and patients with multiple sclerosis.

Materials and Methods

We received local institutional review board approval for this prospective study, and written informed consent was obtained from all participants.

All MRI examinations were performed with a 3-T clinical scanner (MR750; GE Healthcare Technologies, Milwaukee, Wis). A 12-channel receive-only head coil was used for signal reception. A myelin phantom study was designed to validate the signal source in 3D DESIRE UTE imaging, and a water phantom with different T1 values was used to assess the efficiency of water signal suppression. This was followed by ex vivo and in vivo brain studies.

Phantom Preparation

Extracted myelin from the porcine brain was used to validate direct imaging of myelin using UTE sequences. White matter isolated from the porcine brain was mechanically homogenized with a Dounce homogenizer (Thomas Scientific, Swedesboro, NJ). Myelin vesicles were purified by using discontinuous sucrose gradient ultracentrifugation and osmotic shocking according to published methods (23), with the procedure repeated three times to achieve a high degree of purity. After thoroughly washing out any residual sucrose, the myelin vesicles were resuspended twice in a deuterated hydroxymethyl aminomethane hydrochloride buffer to remove residual water (Fig E1 [online]) and then placed in a 1.0-mL syringe.

To test the effectiveness of the DESIRE sequence for suppressing water signals, six myelin and water phantoms with different T1 values were prepared. These phantoms were doped with gadobenate dimeglumine (MultiHance; Spectrum Chemical, Los Angeles, Calif) at concentrations of 0.075, 0.125, 0.175, 0.225, 0.275, and 0.325 mmol/L. The six phantoms were then placed in parallel in a cylindrical container filled with 1% agarose gel; in effect, this created a seventh control phantom without gadobenate dimeglumine.

Cadaveric Human Brain Study

Four cadaveric human heads from four donors (two healthy donors, an 82-year-old man and an 87-year-old woman; two donors with MS, a 45-year-old man and a 56-year-old woman) were obtained from a nonprofit whole-body donation company (United Tissue Network, Phoenix, Ariz). The cadaveric heads were fresh frozen and had been stored in a freezer at −80°C. Each specimen was placed in a 4°C refrigerator for 12 hours then removed and stored at room temperature for 24 hours to thaw before the MRI studies. After imaging, each specimen was refrozen at −80°C and cut into 1-cm axial sections. Biopsies were performed in regions of interest identified at image review, and tissue samples were fixed in zinc formalin (Anatech, Battle Creek, Mich) for 1 week. After paraffin embedding, 5-μm sections were cut and stained overnight in alcoholic 0.1% Luxol fast blue at 60°C, counterstained with neutral red, dehydrated, and mounted (Permount; Fisher Scientific, Pittsburgh, Pa).

In Vivo Brain Study

A total of 20 healthy volunteers and 20 patients with MS were scanned between November 2017 and February 2019.

MRI Protocol

A 3D adiabatic inversion recovery–prepared UTE-Cones sequence was used in this study (Fig 1) (24,25). The adiabatic inversion pulse (Silver-Hoult sequence; duration, 6.048 msec; bandwidth, 1.643 kHz; maximum B1 amplitude, 17 µT) can be used to uniformly invert the longitudinal magnetizations of long T2 components in both white matter and gray matter and is insensitive to B1 inhomogeneity (Fig 1a) (26). The longitudinal magnetization of myelin with its extremely short T2* of approximately 0.3 msec is not inverted by this pulse but is partially saturated. At inversion time, a train of dual-echo (UTE and gradient-echo) Cones spokes (total number of spokes, N_sp) are used for data acquisition. In each spoke, the UTE data acquisition starts as soon as possible after the radiofrequency excitation, with a minimal nominal echo time of 32 µsec (Fig 1b).

Figure 1a:

Three-dimensional DESIRE (double-echo sliding inversion recovery) ultrashort echo time (UTE) sequence. (a) This sequence uses an adiabatic inversion pulse for selective long T2 signal suppression followed by a train of data acquisition spokes. (b) In each spoke, a short rectangular pulse is used for signal excitation, followed by dual-echo spiral sampling with a minimal nominal echo time of 32 µsec. After the inversion recovery (IR) pulse, the magnetizations of long T2 components in white matter (WM_L) and gray matter (GM_L) are inverted and recover via longitudinal relaxation. However, myelin magnetization is partially saturated by the relatively long adiabatic IR pulse because of its extremely short T2. It recovers during inversion time. To obtain images with inversion times at or close to the signal nulling points of long T2 components, a sliding-window technique is used to generate a series of images with gradually increasing inversion times. Spokes within a window with a size of N_w are used to generate one image. The window starts at the beginning and slides forward one spoke each time, thus generating a new image. The total number of images is equal to N_sp−N_w + 1, where N_sp is the number of spokes and N_w is the window size. (c) All spokes are randomly ordered for data sampling to ensure each group of spokes is uniformly distributed in k-space. The k-space trajectory has been normalized from −1 to 1 for all axes. DAW = data acquisition window, FID = free induction decay, G_x,y,z = gradients in x, y, and z axes, RF = radiofrequency, TE = echo time, TR = repetition time.

Figure 1b:

Three-dimensional DESIRE (double-echo sliding inversion recovery) ultrashort echo time (UTE) sequence. (a) This sequence uses an adiabatic inversion pulse for selective long T2 signal suppression followed by a train of data acquisition spokes. (b) In each spoke, a short rectangular pulse is used for signal excitation, followed by dual-echo spiral sampling with a minimal nominal echo time of 32 µsec. After the inversion recovery (IR) pulse, the magnetizations of long T2 components in white matter (WM_L) and gray matter (GM_L) are inverted and recover via longitudinal relaxation. However, myelin magnetization is partially saturated by the relatively long adiabatic IR pulse because of its extremely short T2. It recovers during inversion time. To obtain images with inversion times at or close to the signal nulling points of long T2 components, a sliding-window technique is used to generate a series of images with gradually increasing inversion times. Spokes within a window with a size of N_w are used to generate one image. The window starts at the beginning and slides forward one spoke each time, thus generating a new image. The total number of images is equal to N_sp−N_w + 1, where N_sp is the number of spokes and N_w is the window size. (c) All spokes are randomly ordered for data sampling to ensure each group of spokes is uniformly distributed in k-space. The k-space trajectory has been normalized from −1 to 1 for all axes. DAW = data acquisition window, FID = free induction decay, G_x,y,z = gradients in x, y, and z axes, RF = radiofrequency, TE = echo time, TR = repetition time.

When UTE acquisitions start at the TI_null of WM_L, the myelin signal in white matter is selectively detected and further highlighted by the echo subtraction (second echo from the first), which reduces the residual signal from gray matter. To find TI_null, a series of images with different TIs are generated with the DESIRE UTE sequence. Acquisition spokes are grouped in a window with a fixed size of N_w, and these spokes are subsequently used to generate an image (Fig 1a). The window starts sliding after the very first N_w spokes with the shortest inversion time, then slides one spoke at a time to generate a new image with a slightly longer inversion time. The total number of groups or images is equal to N_sp − N_w + 1. With this sliding window technique, a series of images with different T1-dependent contrasts is generated. Individual images can then be used to determine the inversion time that best nulls WM_L. This TI_null is found when the signal in white matter in the second echo image reaches a minimum (the noise level). This is where the WM_L signal is nulled and the short T2 myelin signal has fully decayed. Different subjects may have different TI_null values due to the T1 variations among volunteers and patients.

Five different windows (N_w = 10, 20, 30, 40, and 50) were used to investigate whether different windows would affect the optimal TI_null for WM_L and thus the effectiveness of myelin imaging.

We used random view ordering with a bit-reversed algorithm for data sampling. With this method, groups of spokes in the same window were randomly distributed in the k-space (Fig 1c); this ordering can be effectively reconstructed by using combined parallel imaging with compressed sensing (PI-CS) reconstruction.

Figure 1c:

Three-dimensional DESIRE (double-echo sliding inversion recovery) ultrashort echo time (UTE) sequence. (a) This sequence uses an adiabatic inversion pulse for selective long T2 signal suppression followed by a train of data acquisition spokes. (b) In each spoke, a short rectangular pulse is used for signal excitation, followed by dual-echo spiral sampling with a minimal nominal echo time of 32 µsec. After the inversion recovery (IR) pulse, the magnetizations of long T2 components in white matter (WM_L) and gray matter (GM_L) are inverted and recover via longitudinal relaxation. However, myelin magnetization is partially saturated by the relatively long adiabatic IR pulse because of its extremely short T2. It recovers during inversion time. To obtain images with inversion times at or close to the signal nulling points of long T2 components, a sliding-window technique is used to generate a series of images with gradually increasing inversion times. Spokes within a window with a size of N_w are used to generate one image. The window starts at the beginning and slides forward one spoke each time, thus generating a new image. The total number of images is equal to N_sp−N_w + 1, where N_sp is the number of spokes and N_w is the window size. (c) All spokes are randomly ordered for data sampling to ensure each group of spokes is uniformly distributed in k-space. The k-space trajectory has been normalized from −1 to 1 for all axes. DAW = data acquisition window, FID = free induction decay, G_x,y,z = gradients in x, y, and z axes, RF = radiofrequency, TE = echo time, TR = repetition time.

The sequence parameters for the phantom study and ex vivo and in vivo (in healthy volunteers and patients with MS) brain studies are shown in the Table. The total scanning time for the DESIRE UTE sequence in the in vivo study was 10 minutes.

Sequence Parameters for the Phantom and ex Vivo and in Vivo Brain Studies

Image Reconstruction

Each group of raw data was reconstructed independently with PI-CS. The sensitivity encoding–based PI and CS are incorporated into a single optimization framework (27,28):

where f is the unknown vector defining the desired 3D full–field-of-view image to be computed; E is the encoding matrix, which includes both regridding and sensitivity-encoding operators; d is the vector-formed k-space data from all receive channels; is a wavelet transform; is the l₁ norm; is the l₂ norm; and is the total variation term; λ and α are the two regularization parameters used to control the weights of l₁ and the total variation terms, respectively.

The coil sensitivities are derived by dividing highly smoothed images from all the channels by their sum-of-squares combination. The images used to calculate coil sensitivity are obtained from the final group of spokes with the longest inversion time.

Regridded images were used as the initial image f estimation in the PI-CS computation. We set λ and α to the same value, which was equal to a constant C multiplying the maximum value of the initial images after regridding. The constant C was experimentally determined by solving the Equation with different C values and choosing one so that Ef − d₂ < 0.2 (27,28). We found that good imaging quality (a relatively high signal-to-noise ratio [SNR] with no apparent artifact and preservation of image detail) could be achieved with C of 0.004 for all the data.

In-house implementation of PI-CS reconstruction was written in Matlab 2016b (MathWorks, Natick, Mass) at a workstation with an Intel i7–4790 CPU, 3.6-GHz processor base frequency, and 32-GB RAM. A 3D nonuniform fast-Fourier transform Compute Unified Device Architecture, or CUDA, implementation with Matlab mex interface was used for fast 3D regridding (29).

Statistical Analysis

The contrast-to-noise ratios between MS lesions and normal white matter were calculated as their signal difference divided by background noise. Mean ± standard deviation for the contrast-to-noise ratios was reported. The measured signal intensities from the DESIRE UTE myelin images were normalized using the equation S_n = (S-S_min)/(S_max-S_min) before statistical analysis. S, S_min, S_max, and S_n are measured signal intensity, minimal signal intensity, maximal signal intensity, and normalized signal intensity, respectively. Analysis of variance was performed to evaluate the signal difference between MS lesions and normal-appearing white matter. All data processing was performed with in-house code written in Matlab.

Results

Phantom Studies

In the myelin phantom, 3D UTE images showed high signal from the intact myelin vesicles, which had a short T1 of 367 msec ± 4 (mean ± standard deviation) and a T2* of 225 µsec ± 7 (Fig E2 [online]).

As mentioned previously, a major challenge for direct imaging of myelin in vivo is the high signal from WM_L. Figure 2 shows that the DESIRE UTE sequence can be used to null signals from gadobenate dimeglumine–doped water containing agarose gel phantoms with T1 values of 513, 565, 678, 854, 1069, 1484, and 2880 msec, as measured with a 3D UTE actual flip angle and variable flip angle technique (30,31). Seventy-six images with different inversion times were generated from a single DESIRE UTE data acquisition. The signal intensity changes with inversion time for each phantom are shown in Figure 2b. The optimal TI_null for each phantom is found at the lowest point of each signal intensity curve, and, as expected, the longer the T1, the longer the optimal TI_null. Thus, one DESIRE UTE acquisition can generate images, with at least one of them showing near-perfect nulling despite their very different T1 values. The reconstructed image numbers (N_imgs) range from 1 to 74 depending on their inversion time. The N_imgs for nulling are listed here and are shown in Figure 2c (phantom 1, N_img = 28; phantom 2, N_img = 31; phantom 3, N_img = 34; phantom 4, N_img = 37; phantom 5, N_img = 41; phantom 6, N_img = 46; and agarose phantom, N_img = 51).

Figure 2a:

Three-dimensional (3D) double-echo sliding inversion recovery (DESIRE) ultrashort echo time (UTE) imaging of gadobenate dimeglumine phantoms. (a) The gray scale of the left image was adjusted to show image details. The T1s of the six water phantoms doped with different concentrations of gadobenate dimeglumine and agarose gel (Agr) were 513, 565, 678, 854, 1069, 1484, and 2880 msec, respectively, as measured with the 3D UTE-Cones actual flip angle and variable flip angle methods. The unit for the color legend is milliseconds. (b) Signal curves of the 76 images obtained with the sliding-window method for all phantoms. (c) DESIRE UTE images were reconstructed at the optimal nulling point for each phantom. N_img = image number.

Figure 2b:

Three-dimensional (3D) double-echo sliding inversion recovery (DESIRE) ultrashort echo time (UTE) imaging of gadobenate dimeglumine phantoms. (a) The gray scale of the left image was adjusted to show image details. The T1s of the six water phantoms doped with different concentrations of gadobenate dimeglumine and agarose gel (Agr) were 513, 565, 678, 854, 1069, 1484, and 2880 msec, respectively, as measured with the 3D UTE-Cones actual flip angle and variable flip angle methods. The unit for the color legend is milliseconds. (b) Signal curves of the 76 images obtained with the sliding-window method for all phantoms. (c) DESIRE UTE images were reconstructed at the optimal nulling point for each phantom. N_img = image number.

Figure 2c:

Three-dimensional (3D) double-echo sliding inversion recovery (DESIRE) ultrashort echo time (UTE) imaging of gadobenate dimeglumine phantoms. (a) The gray scale of the left image was adjusted to show image details. The T1s of the six water phantoms doped with different concentrations of gadobenate dimeglumine and agarose gel (Agr) were 513, 565, 678, 854, 1069, 1484, and 2880 msec, respectively, as measured with the 3D UTE-Cones actual flip angle and variable flip angle methods. The unit for the color legend is milliseconds. (b) Signal curves of the 76 images obtained with the sliding-window method for all phantoms. (c) DESIRE UTE images were reconstructed at the optimal nulling point for each phantom. N_img = image number.

Human Studies

A total of 20 healthy volunteers (mean age, 40 years ± 13; 10 women) and 20 patients with MS (mean age, 58 years ± 8; 15 women) were recruited. Figure 3a shows representative DESIRE UTE images corresponding to five different inversion times (N_img 4, 18, 27, 36, and 60) from a total of 61 inversion times and demonstrates variation in image contrast with inversion time in one brain scan. The corresponding dual-echo images reconstructed with PI-CS are shown in Figure 3b (27,32). Differences in T1-dependent contrast are seen in the five dual-echo images: In N_img 4, white and gray matter have similar signal intensities. In N_img 18, white matter signal intensity is lower than gray matter signal intensity. N_img 27 is close to the nulling point of WM_L and highlights gray matter. N_img is close to the nulling point of long T2 components in gray matter (GM_L) and highlights white matter. In N_img 60, white matter has a higher signal intensity than gray matter. Signal intensity and contrast differences between the two different echo time images at the same inversion time mainly result from differences in T2-dependent signal decay after excitation and before acquisition of the second echo time image. Ultrashort T2 signals (from myelin protons) in the first echo image are always positive because their signals recover quickly starting immediately after the adiabatic inversion recovery pulse, which saturates rather than inverts the longitudinal magnetization. However, WM_L or GM_L signals are inverted by the inversion recovery pulse, and the MRI signal subsequently detected from them is negative at short inversion times (ie, before N_img 27 for WM_L or before N_img 36 for GM_L, respectively). In this situation, white matter and gray matter signals in the second echo image are higher than those in the first echo image, as seen in N_img 18. After N_img 36, both WM_L and GM_L signals are positive. Thus, as shown in image 60, the first echo image has a higher signal intensity than the second echo image because of the additive positive myelin and water signals in the first echo. For image 27, the WM_L signal is almost nulled, as demonstrated by the near-zero signal for WM_L in the second echo image. The first echo image mainly shows the short T2 myelin signal in white matter. If echo subtraction is performed as for image 27 (ie, UTE₁–TE₂), myelin signals in white matter are highlighted because the signal in white matter is positive, while the signal in gray matter is negative.

Figure 3a:

Three-dimensional double-echo sliding inversion recovery ultrashort echo time brain imaging in a representative healthy 37-year-old male volunteer. (a) Five different time points (4, 18, 27, 36, and 60) were selected from a total of 61 time points to demonstrate variation in image contrast. (b) The corresponding dual-echo images were reconstructed by using parallel imaging–compressed sensing reconstruction. Contrast changes can be seen clearly in these five dual-echo images. IR = inversion recovery, TR = repetition time.

Figure 3b:

Three-dimensional double-echo sliding inversion recovery ultrashort echo time brain imaging in a representative healthy 37-year-old male volunteer. (a) Five different time points (4, 18, 27, 36, and 60) were selected from a total of 61 time points to demonstrate variation in image contrast. (b) The corresponding dual-echo images were reconstructed by using parallel imaging–compressed sensing reconstruction. Contrast changes can be seen clearly in these five dual-echo images. IR = inversion recovery, TR = repetition time.

The normalized signal intensity curves for a region of interest drawn in a white matter region are displayed for the first and second echo images (Fig 4a, 4b). The TI_null for WM_L can be determined accurately by choosing the signal nulling point for the second echo (N_img 26), as shown by the arrow in Figure 3a. Figure 3b shows the dual-echo images and the corresponding subtraction images close to the WM_L nulling point. The highest myelin contrast with the most complete water suppression was achieved for N_img 26. Lower myelin signal was observed with a shorter inversion time (N_img <26) because of the cancelation of positive magnetization from myelin and negative magnetization from WM_L. Higher myelin signal was observed with a longer inversion time (N_img >26) because of the residual positive magnetization from WM_L, which adds to that of myelin, leading to a higher signal level.

Figure 4a:

Three-dimensional double-echo sliding inversion recovery ultrashort echo time brain MRI with inversion times close to the long T2 components in white matter (WM_L) nulling point (TI_null). (a) Normalized signal intensity curves for a region of interest in a white matter region (yellow circle in b) for the first (Echo 1, echo time = 0.032 msec) and second (Echo 2, echo time = 2.2 msec) echo images. TI_null for WM_L corresponds to the signal TI_null for Echo 2 (arrow). (b) Dual-echo images and corresponding subtraction images close to TI_null for WM_L. Myelin images are obtained by subtracting the first echo from the second echo, N_img = image number.

Figure 4b:

Three-dimensional double-echo sliding inversion recovery ultrashort echo time brain MRI with inversion times close to the long T2 components in white matter (WM_L) nulling point (TI_null). (a) Normalized signal intensity curves for a region of interest in a white matter region (yellow circle in b) for the first (Echo 1, echo time = 0.032 msec) and second (Echo 2, echo time = 2.2 msec) echo images. TI_null for WM_L corresponds to the signal TI_null for Echo 2 (arrow). (b) Dual-echo images and corresponding subtraction images close to TI_null for WM_L. Myelin images are obtained by subtracting the first echo from the second echo, N_img = image number.

Images at the optimal nulling point for all five different window sizes (N_w = 10, 20, 30, 40, and 60) are shown in Figure E3 (online). As can be seen, white matter signal in the second echo was almost completely nulled with all five window sizes, and the corresponding subtraction images all show high myelin contrast. This result shows that an optimal TI_null can be found for each of the different window sizes (ie, TI_null can be adjusted so that long T2 water magnetizations acquired before TI_null cancels that acquired after TI_null, leading to efficient water signal suppression). The SNR of myelin images increased from 9.4 for a window size of 10 to 16.4 for a window size of 40 and then dropped slightly to 15.6 for a window size of 50. The higher SNR with increased window size (from 10 to 40) is reasonable since more data are used for reconstruction. On the other hand, if window size is too large, there is more signal variation between the spokes (a large window size means more negative long T2 water magnetizations for spokes acquired before TI_null and more positive long T2 water magnetizations for spokes acquired after TI_null), which may lead to increased image artifacts and pseudonoise (and thus a decrease in SNR). The SNR of the myelin image with window size of 50 was slightly decreased, which suggested the existence of signal fluctuations even though there were no apparent artifacts (the PI-CS reconstruction framework helps in reducing these artifacts). As a result, a window size of less than 40 was preferred with the current DESIRE UTE protocol. In this study, a fixed window (N_w = 30) was used for all patient data reconstruction. It yielded reasonable image SNR without apparent image artifacts. With the fixed window, the myelin images could be obtained at the optimal nulling point in all patients. Representative whole-brain myelin images in a healthy volunteer with a window size of 30 are shown in Figure 5. The TI_null used in volunteers and patients ranged from 309 msec to 324 msec.

Figure 5:

Selected whole-brain myelin images generated with double-echo sliding inversion recovery ultrashort echo time imaging with a window of 30 and image number 17.

Figure 6 shows four representative imaging slices obtained from the brain of an individual with MS by using DESIRE UTE and clinical T1- and T2-weighted sequences. MS lesions identified on the clinical images are clearly depicted on the 3D DESIRE UTE images and are shown as areas of signal reduction corresponding to loss of myelin as subsequently found at histology. Similar findings were observed in all the other MS brain specimens.

Figure 6a:

Ex vivo images of the brain in a 45-year-old man with multiple sclerosis (MS). (a) MS lesions (arrows and circles) on clinical T1-weighted (T_1w) magnetization-prepared rapid acquisition gradient-echo (repetition time msec/echo time msec, 7/3) (top row) and T2-weighted (T_2w) fast spin-echo (1500/60) (middle row) images are shown as signal voids on three-dimensional double-echo sliding inversion recovery (DESIRE) ultrashort echo time images (bottom row), corresponding to marked myelin loss. Note the location of the lesion sampled for histologic analysis (arrowheads). (b) The 5-μm sections of the brain sample were cut and stained overnight in alcoholic 0.1% Luxol fast blue at 60°C and were counterstained with neutral red, dehydrated, and mounted. Histologic analysis of normal-appearing white matter (top), lesion edge (middle), and lesion (bottom) show specific loss of myelin staining in the MS lesion.

Figure 6b:

Ex vivo images of the brain in a 45-year-old man with multiple sclerosis (MS). (a) MS lesions (arrows and circles) on clinical T1-weighted (T_1w) magnetization-prepared rapid acquisition gradient-echo (repetition time msec/echo time msec, 7/3) (top row) and T2-weighted (T_2w) fast spin-echo (1500/60) (middle row) images are shown as signal voids on three-dimensional double-echo sliding inversion recovery (DESIRE) ultrashort echo time images (bottom row), corresponding to marked myelin loss. Note the location of the lesion sampled for histologic analysis (arrowheads). (b) The 5-μm sections of the brain sample were cut and stained overnight in alcoholic 0.1% Luxol fast blue at 60°C and were counterstained with neutral red, dehydrated, and mounted. Histologic analysis of normal-appearing white matter (top), lesion edge (middle), and lesion (bottom) show specific loss of myelin staining in the MS lesion.

As with the MS brain specimen study, myelin loss in patients was identified on DESIRE UTE images in MS lesions shown with clinical T1- and T2-weighted sequences (Figs 7, 8). The mean contrast-to-noise ratios between lesions and normal-appearing white matter measured for representative patients in Figures 7 and 8 were 15.8 ± 2.2 and 15.2 ± 1.9, respectively. In all the patients involved in this study, there were significant differences in the normalized signal intensities between MS lesions (0.19 ± 0.10) and normal-appearing white matter (0.76 ± 0.11) with a P value less than .001. This result demonstrated the clinical feasibility of the 3D DESIRE UTE technique for volumetric imaging of myelin in the brain.

Figure 7:

Images in the top, middle, and bottom rows are from different brain sites in a 59-year-old woman with multiple sclerosis (MS). MS lesions (arrows and circles) on the clinical T1-weighted (T_1w) magnetization-prepared rapid acquisition gradient-echo (repetition time msec/echo time msec, 8.2/3.2) (left) and T2-weighted (T_2w) fluid-attenuated inversion recovery (7600/117) (middle) images are shown as signal voids, corresponding to marked myelin loss on the three-dimensional double-echo sliding inversion recovery (DESIRE) ultrashort echo time images (right).

Figure 8:

Images in the top and bottom rows are from different brain sites in a 45-year-old woman with multiple sclerosis (MS). MS lesions (arrows and circles) on the clinical T1-weighted (T_1w) magnetization-prepared rapid acquisition gradient-echo (repetition time msec/echo time msec, 8.2/3.2) (left) and T2-weighted (T_2w) fluid-attenuated inversion recovery (7600/117) (middle) images are shown as signal voids, corresponding to marked myelin loss on the three-dimensional double-echo sliding inversion recovery (DESIRE) ultrashort echo time images (right).

Discussion

In this study, we have shown that the three-dimensional (3D) double-echo sliding inversion recovery (DESIRE) ultrashort echo time (UTE) sequence can generate patient-specific whole-brain myelin images using a clinical 3-T scanner. The high MRI signal from myelin vesicles extracted from the porcine brain provided evidence that myelin can be directly imaged with 3D UTE sequences. The DESIRE UTE method enables inversion and nulling of long T2 signals in white matter regardless of their variation in T1 and enables selective imaging of myelin in white matter of the brain. It avoids trying to preselect the correct inversion time for long T2 white matter nulling for each volunteer and patient. The water phantom study showed the robustness of signal nulling provided by this technique. The study of the cadaveric human brain in individuals who had multiple sclerosis (MS) further validated direct myelin imaging using the 3D DESIRE UTE sequence. Finally, the whole-brain myelin images in healthy volunteers and patients with MS showed reduction in mean normalized signal intensity in MS lesions compared with normal-appearing white matter (0.19 ± 0.10 vs 0.76 ± 0.11, P < .001). MRI showed obvious myelin loss in MS lesions identified with conventional clinical MRI sequences, as well as the feasibility and potential value of the 3D DESIRE UTE technique for volumetric imaging of myelin in vivo. The technique also has potential in diseases other than MS, as loss of myelin is also seen in numerous inflammatory and neurodegenerative disorders, including different forms of leukodystrophy (33).

We used a relatively low excitation flip angle of 10° to reduce the likelihood of contrast changes between adjacent DESIRE UTE images and to help achieve the best nulling point for selective myelin imaging. The contrast variation between adjacent DESIRE UTE images can also be reduced by using a shorter spoke interval. We used the shortest possible spoke intervals in the DESIRE UTE acquisitions. However, the degree to which spoke interval can be shortened is limited by echo spacing and the use of a dual-echo acquisition. To further reduce the spoke interval, an imaging system with a higher maximum gradient amplitude could be used. In addition, flexibility in setting the window size allows greater sharing of high-frequency projection data to improve myelin SNR while reducing streak artifacts.

The 3D DESIRE UTE technique provides a series of morphologic images with different T1-dependent contrast in a single scan compared with the single T1-dependent contrast obtained with conventional clinical T1- or PD-weighted MRI sequences. Moreover, we showed that whole-brain T1 mapping can be performed by fitting the second-echo images with a simple model, similar to the method described by Kecskemeti et al (34). Thus, in addition to whole-brain myelin imaging, the proposed 3D DESIRE UTE technique yields comprehensive morphologic images with different T1 weightings. It can also produce quantitative images, such as volumetric T1 maps, although more work still needs to be done to estimate the myelin T1 from the series of dual-echo DESIRE UTE images, including consideration of T2* decay of the short and long T2 components.

While we only investigated 3D DESIRE UTE Cones imaging, the DESIRE contrast mechanism could, in principle, be combined with other UTE imaging techniques, such as radial UTE, zero echo time, or ZTE, (35,36), and sweep imaging with Fourier transformation, or SWIFT, (37). Furthermore, the DESIRE technique may be combined with conventional gradient-echo or spin-echo acquisitions to yield images with selective nulling of tissues or tissue components with specific T1 values using a range of TI values.

There were several limitations to this study. First, the total data acquisition time for high-resolution volumetric myelin imaging is relatively long. Apparent motion artifacts were observed in four of 20 patients during scanning. Use of a dedicated head coil with more receive channels and higher SNR performance would permit greater undersampling and would reduce the total scan time and any patient discomfort associated with this technique. Second, for this study, we used a conventional hardware image processing workstation that required extensive data processing time (8 hours for one subject); use of a graphics processing unit–based PI-CS reconstruction or a more advanced workstation would greatly increase computational efficiency. Third, both regularization parameters used to control the weights of l₁ and the total variation terms (λ and α, respectively) were experimentally determined similar to previous studies (27,28). A more comprehensive analysis (comparison between prepost SNR, image artifacts, and blurriness) would be warranted in future studies. Fourth, we are currently conducting a larger study to compare DESIRE UTE with conventional clinical sequences to detect lesions and monitor response to initiation of therapy in patients with MS. Results of this study will be reported. In this study, we only focused on morphologic detection of MS lesions. Quantitative developments and measurements will be conducted soon. Fifth, we did not seek to image myelin in gray matter in this study; a similar strategy to the one used here for white matter myelin imaging could be used for gray matter myelin imaging, although this would be more challenging because gray matter has a considerably lower myelin content than white matter. The technique could also be used for volumetric myelin imaging in the spinal cord and peripheral nervous system. Sixth, quantitative imaging of myelin, including T1, T2*, and myelin proton density mapping is still under development and will need validation. Lastly, no comparative study was performed to see if DESIRE UTE would provide different lesion findings in diseases without demyelination (eg, migraine) compared with MS (38). In the future, we will further develop quantitative DESIRE UTE techniques and apply them to different neurologic diseases to investigate the potential of the sequence to improve differential diagnosis.

In summary, the three-dimensional (3D) double-echo sliding inversion recovery (DESIRE) ultrashort echo time (UTE) sequence is a robust technique for accurate whole-brain myelin imaging, as demonstrated in studies of phantoms, brain specimens, volunteers, and patients with multiple sclerosis (MS). The extracted myelin vesicle study provides clear evidence that myelin can be directly imaged with UTE sequences using a clinical 3-T scanner, and the water phantom study showed that the long T2 signal suppression was effective for different T1 values. The studies of healthy volunteers and patients with MS demonstrate the potential of 3D DESIRE UTE in the morphologic evaluation of demyelination and remyelination in vivo. This may improve the accuracy of diagnosis and monitoring of disease progression and remission in patients with MS or other neurologic diseases.