High-resolution DTI of a localized volume using 3D single-shot diffusion-weighted STimulated echo-planar imaging (3D ss-DWSTEPI)


  • Eun-Kee Jeong,

    Corresponding author
    1. Utah Center for Advanced Imaging Research, Department of Radiology, University of Utah, Salt Lake City, Utah, USA
    • Utah Center for Advanced Imaging Research, Department of Radiology, University of Utah, 729 Arapeen Dr., Salt Lake City, UT 84108
    Search for more papers by this author
  • Seong-Eun Kim,

    1. Utah Center for Advanced Imaging Research, Department of Radiology, University of Utah, Salt Lake City, Utah, USA
    Search for more papers by this author
  • Eugene G. Kholmovski,

    1. Utah Center for Advanced Imaging Research, Department of Radiology, University of Utah, Salt Lake City, Utah, USA
    Search for more papers by this author
  • Dennis L. Parker

    1. Utah Center for Advanced Imaging Research, Department of Radiology, University of Utah, Salt Lake City, Utah, USA
    Search for more papers by this author


Diffusion tensor MRI (DTI) using conventional single-shot (SS) 2D diffusion-weighted (DW)-EPI is subject to severe susceptibility artifacts. Multishot DW imaging (DWI) techniques can reduce these distortions, but they generally suffer from artifacts caused by motion-induced phase errors. Parallel imaging can also reduce the distortions if the sensitivity profiles of the receiver coils allow a sufficiently high reduction factor for the desired field of view (FOV). A novel 3D DTI technique, termed 3D single-shot STimulated EPI (3D ss-STEPI), was developed to acquire high-resolution DW images of a localized region. The new technique completes k-space acquisition of a limited 3D volume after a single diffusion preparation. Because the DW magnetization is stored in the longitudinal direction until readout, it undergoes T1 rather than T2 decay. Inner volume imaging (IVI) is used to limit the imaging volume. This reduces the time required for EPI readout of each complete kx-ky plane, and hence reduces Tmath image decay during the readout and T1 decay between the readout of each kz. 3D ss-STEPI images appear to be free of severe susceptibility and motion artifacts. 3D ss-STEPI allows high-resolution DTI of limited volumes of interest, such as localized brain regions, cervical spinal cord, optic nerve, and other extracranial organs. Magn Reson Med, 2006. © 2006 Wiley-Liss, Inc.

Diffusion-weighted MR imaging (DWI) is a well established tool for detecting abnormal water diffusion in the brain (e.g., ischemic stroke) (1). The directional information obtained using diffusion tensor MRI (DTI) is valuable for understanding and evaluating white matter (WM) abnormalities in neurological diseases, such as Alzheimer disease, schizophrenia, multiple sclerosis, and neurofibromatosis (2). DWI and DTI may also give useful information about the development and disorders of ordered structures in extracranial organs, such as the heart, kidney, breast, and prostate (3–5).

Although DTI can provide useful information about WM diseases in the brain, high-resolution DTI of small neural structures (e.g., the spinal cord or optic nerve), extracranial organs in vivo, and brain regions near the temporal bone or sinuses has been difficult to achieve using conventional 2D single-shot DW EPI (2D ss-DWEPI) techniques. This is because strong, nonuniform local magnetic fields are created by magnetic susceptibility changes at tissue/bone or tissue/air interfaces, which typically induce severe distortion on the resultant ss-DWEPI images. The amount of susceptibility-induced geometric distortion is proportional to the total sampling time in EPI. Typically, an increase in spatial resolution requires an increase in the duration of the data acquisition window, which in turn increases the distortion from off-resonance effects. As a result, the spatial resolution obtained using conventional 2D ss-EPI is generally much lower than that obtainable with conventional multishot MRI, which results in decreased resolution for measurements of interest, such as WM tract anatomy and nerve fiber anatomy. For these reasons, 2D ss-DWEPI has been clinically useful only for moderately low-resolution intracranial applications. EPI with parallel imaging has been successfully applied to high-resolution brain DWI and DTI studies, and resulted in substantial improvements in image quality (6, 7).

There are several non-EPI SS-DWI techniques that complete the total data acquisition following a single diffusion weighting, including multiple spin-echo (SE) sequences such as SS fast SE (ss-FSE) (8) and gradient SE (GRASE) (9), stimulated echo acquisition mode (STEAM) (10), and fast gradient-echo (FGRE) sequences (11). These 2D sequences typically acquire slightly more than half of the ky encodings in about 500 ms after a single DW preparation and employ relatively thick slices to overcome their intrinsic low signal-to-noise ratio (SNR) (8–11).

Multishot imaging techniques may be used to increase SNR, improve spatial resolution, and reduce susceptibility-induced artifacts (12–15). However, the majority of multishot DWI acquisition techniques suffer from instability of phase errors between shots due to global or localized motions during application of the large diffusion gradients. Reasonable success has been achieved with techniques that use navigator echoes to detect and correct phase errors (12, 14, 16, 17), and non-SS-EPI approaches that are less sensitive to phase errors. Because most of these are 2D acquisition techniques, however, they produce relatively poor resolution along the slice direction.

In this paper, 3D single-shot DW STimulated EPI (3D ss-DWSTEPI) is presented as a novel technique to perform 3D SS DWI and DTI of a restricted 3D volume. 3D ss-DWSTEPI acquires the entire 3D k-space data from a limited 3D volume after a single diffusion-prepared driven-equilibrium (DPDE) preparation by short EPI readouts of several stimulated echoes. The EPI readout time is shortened by using an inner volume imaging (IVI) technique along the phase-encoding direction (18, 19). To our knowledge, this is the first report of a 3D SS-DWI technique that acquires the complete k-space in a single diffusion preparation. This novel approach may be advantageous for high-resolution DTI and DWI because it is a 3D acquisition technique that provides increased SNR and contiguous thin slices. Even though spatial coverage is limited to suppress susceptibility artifacts, localized 3D ss-DWSTEPI may allow the acquisition of high-resolution DTI data from nearly any localized region of the body, and thus overcome some major limitations of conventional 2D ss-DWEPI.


Pulse Sequence Description

The 3D ss-DWSTEPI pulse sequence was developed from the multishot SE-EPI sequence using the IDEA pulse sequence development environment (Siemens Medical Solutions, Erlangen, Germany). The diagram of the technique is shown in Fig. 1, where ACQ represents the digital sampling of the MR signal. The pulse sequence consists of two main sections: DPDE preparation and 3D data acquisition.

Figure 1.

Diagram of the 3D ss-DWSTEPI pulse sequence. DPDE includes 90° excitation, double inversion for IVI, three reference gradient echoes for EPI correction, diffusion weighting, dephasing gradients, 90° tip-up, and spoiling gradients. A complete set of ky views are acquired after each α RF pulse. a: The horizontal arrows → and ← represent the dephasing and rephasing crusher gradients, respectively. The vertical arrows indicate the slice-selection gradient. b: The imaging echo train is repeated for all actual slice-encodings (#SE). The stimulated echo occurs at ky = 0 of each echo train.

DPDE (90°–[180°–180°]IVI–GD–180°–GD–90°tip-up) preparation precedes the stimulated-echo imaging sequence. The first two 180° RF pulses (the pulses enclosed in the dotted box in Fig. 1a) following the 90° RF excitation pulse determine the localized volume for interleaved multiple IVI. The first inversion of the double inversion is used to invert all magnetization and ultimately eliminate unwanted signal from the out-of-volume magnetization. The second inversion is used to restore the magnetization in other slabs to be imaged and allows time-efficient interleaved acquisition of multiple slabs (18, 19). The earliest group of ACQs between the double inversion and the first diffusion gradient collects three reference echoes (two odd and one even) for EPI phase correction. Before it is tipped up to the longitudinal direction, the diffusion-prepared transverse magnetization in each voxel is dephased more than 2π by a dephasing gradient (indicated by right arrow in Fig. 1a) to remove the image intensity dependence on the tip-up RF pulse phase (20). The residual transverse magnetization is suppressed by a spoiler gradient applied after the tip-up pulse. The slice-selection gradient (indicated by the vertical arrow in Fig. 1a) is applied in the slice-encoding direction for all RF pulses in DPDE preparation, except for the two IVI refocusing/inversion RF pulses, where the first pulse is spatially nonselective and the gradient for the second 180° pulse is applied along the phase-encoding direction to define the reduced phase FOV.

The data acquisition part of the pulse sequence consists of multiple segments (Fig. 1b). Each segment includes an excitation RF pulse (creating a single stimulated echo), rephasing crusher gradient, EPI readout, and rewinding gradients. For each segment the flip angle of the imaging RF pulses is gradually increased to reduce the T1 decay-related blurring in the slice direction. The flip angle for the last segment is 90° to consume all remaining longitudinal DW magnetization. The rephasing crusher gradient (indicated by the left arrow in Fig. 1), which is applied immediately after the slice-selection gradient of excitation RF pulse, α, rephases the phase accumulated during the dephasing crusher gradient (→) prior to the tip-up 90° RF pulse. The echo-train length (ETL) of the EPI readout in each segment is chosen to be the same as the number of acquired ky phase-encodings (e.g., 31 for an imaging matrix with 48 ky views). This number is kept small to reduce susceptibility artifacts. The phase-encoding order is increased linearly in each segment, and a center-out slice encoding order is used to improve the SNR by placing the center of slice-encoding at the earliest echo-train acquisition. Asymmetric sampling was used in the phase- and slice-encoding directions, and a reduced sampling with 62.5% was used to reduce the ETL and the length of the total data sampling time (i.e., 10 slice encodings for 16 slices). The data were zero-filled and reconstructed using the reconstruction program supplied by the manufacturer. After completion of each EPI echo train, the remaining transverse magnetization is either completely spoiled by a spoiler gradient or rewound to preserve the transverse coherence and maintain some level of steady-state transverse magnetization. Diffusion weighting in 3D ss-DWSTEPI is implemented as shown in Fig. 2. An extra delay is inserted between the DW SE position and the 90° tip-up RF pulse, which is equal to the time interval between the center of the imaging RF pulse and the stimulated-echo position. Diffusion weighting is achieved by applying the Stejskal-Tanner diffusion-weighting gradient on both sides of the third 180° RF pulse (21) (Fig. 1a) and additional bipolar gradients during the delay to maximize the diffusion weighting for a given echo time (TE). Neglecting gradient ramping up/down time, the b-value for the diffusion-weighting scheme is given by

equation image(1)

Figure 3 describes the evolution of spins to form the diffusion-encoded stimulated echoes in STEPI. DW magnetization Mmath image( equation image) is refocused at the SE position. Then, Mmath image(equation image) is dephased more than 2π by the gradient Gcr1, and the 90°-x RF pulse tips half of the dephased magnetization to the longitudinal direction, leaving the other half in the transverse plane. The transverse component is spoiled by the gradient Gsp. As a result the diffusion-prepared magnetization with dephasing is aligned to the longitudinal direction before the imaging RF pulses. The imaging RF pulse αn tips a fraction of the magnetization into the transverse plane, the gradient Gcr2 rephases the dephasing caused by Gcr1, and a stimulated echo is formed at the position “STE”, at the position where ky = 0 in the EPI readout.

Figure 2.

Diffusion weighting in DPDE preparation. The symbols α and Δ represent the duration of the diffusion gradient and the spacing between the two main diffusion gradients, respectively.

Figure 3.

Formation of the DW stimulated echo in DWSTEPI. The labels “spin-echo” and “stim-echo” indicate the expected position of the spin-echo and the stimulated echo, respectively. Gradients Gcr and Gsp are crusher and spoiler gradients, respectively.

Magnetization Evolution in 3D ss-DWSTEPI

Equation [2] describes the longitudinal magnetization just before the nth imaging RF pulse with respect to the previous longitudinal magnetization value Mmath image (22). Here, αn is the flip angle of the nth imaging RF pulse, and τ is the duration of each data acquisition segment:

equation image(2)

The two terms in Eq. [2] are the freshly recovered and diffusion-prepared magnetization, respectively. Signal from the first term, which is not diffusion weighted, is spoiled after each excitation by the rephasing crusher gradient (indicated by the left arrow in Fig. 1). As a result, the detected MR signal reflects only the DW magnetization, yielding a straightforward single exponential dependence on the applied b-value. The diffusion-prepared longitudinal magnetization decreases along the slice-encoding direction due to the repeated RF pulses and T1 decay.

Neglecting the steady-state transverse magnetization, the DW transverse magnetization after the nth imaging RF pulse αn can be described as:

equation image(3)

where the diffusion-prepared magnetization is defined by:

equation image(4)

For a diffusion weighting of b. The effective TE (TE = TE1 + TE2) is the sum of TEs indicated in Fig. 1, TD is the time delay between the tip-up RF pulse and the first imaging RF pulse (α1), and TACQ is the total pulse sequence duration, which includes the diffusion preparation and complete 3D data readout. The factor ½ arises from applying the pretip-up dephasing gradient to remove the signal dependency on the relative phase between the tip-up RF pulse and DW magnetization (20).

As shown by the second term in Eq. [2], the measured signal experiences T1 rather than T2 decay along the slice-encoding direction. This is very advantageous because T1 is typically an order of magnitude longer than T2 in most tissues. Blurring in the slice-encoding direction, which may arise from T1 decay during the long data acquisition, may be reduced by using variable (ramped) flip angles (23). The transverse magnetizations Mmath image)( equation image,t) and Mmath image(equation image,t) after two consecutive RF pulses (αn−1 and αn) are:

equation image(5)

To achieve equal signal amplitude (Mmath image( equation image,t) = Mmath image(equation image,t)), the relationship between the flip angles of two adjacent RF pulses should satisfy

equation image(6)

The flip angle for the last RF pulse is set to 90° to consume all remaining longitudinal magnetization, and the flip angles of the proceeding RF pulses can be calculated using the relation in Eq. [6]. Typical values of τ and T1 are about 40 ms for 31 ETL with a receiver bandwidth of 1.086 kHz/pixel, and T1 = 1.0 s for WM at 3T. In general, the use of this equation resulted in very small initial flip angles and correspondingly low-SNR images. For this reason we made a compromise between T1 decay-related blurring in the slice-encoding direction and image SNR by using a ramped variable flip-angle scheme with a larger starting angle and smaller increases, ending again with a 90° pulse to consume all remaining longitudinal magnetization.

Because the central planes of k-space are acquired during the first few stimulated echoes, with rewinding or spoiling, the DWI signal intensity undergoes simple exponential decay with respect to the b value, as

equation image(7)

MRI Experiments

MRI studies were performed on a Siemens Trio 3 Tesla MRI system (Siemens Medical Solutions, Erlangen, Germany) with Sonata gradients (40 mT/m strength and 150 T/m/s slew rate). The common imaging parameters for all experiments were the receiver bandwidth of 1.086 kHz/pixel and 31 echoes per EPI echo train. Other parameters typically were slice thickness =1.25 mm, TR = 4000 ms, and effective TE = 75–85 ms, which includes the time for the DPDE preparation and the TE for the EPI echo train (see Fig. 1a). A ramped flip angle scheme was used with a 30° RF pulse for the first echo train. The typical imaging matrix was 192 × 48 × (12–24) with 5/8 partial Fourier acquisition in both phase- and slice-encoding directions, covering 60 mm and 15–30 mm in the phase- and slice-encoding directions, respectively, for 1.25-mm isotropic resolution. Slice oversampling by one or two additional slices on each edge of the imaged volume was used to reduce the slab boundary effect due to imperfect RF pulse profiles. When imaging with multiple averages was used, the resulting image was constructed using magnitude averaging to avoid the phase instability between the averages that can deteriorate both the image quality and the accuracy of the DTI measurement.

DTI acquisition was accomplished with the diffusion encodings applied along seven noncollinear directions: (1,0,0), (0,1,0), (0,0,1), (1,1,1)/ equation image, (−1,−1,1)/equation image (1,−1,−1)/ equation image (−1,1,−1)/ equation image in physical gradient coordinates (Gy, Gx, Gz (representing the vertical, horizontal, and magnet-bore directions, respectively)) or (anterior–posterior (A/P), right–left (R/L), and superior–inferior (S/I)) anatomic coordinates.

DW images were postprocessed using DTI analysis software written in IDL (Research Systems Inc., Boulder, CO, USA) to 1) calculate apparent diffusion coefficient (ADC) maps using Eq. [7], 2) extract the six independent elements of the diffusion tensor matrix using singular-value decomposition, 3) diagonalize the diffusion tensor matrix, and 4) visualize the results using either fractional anisotropy (FA) or a three-color (RGB) representation of the principal eigenvector. The FA was obtained from the three eigenvalues (λ1, λ2, λ3) using (24):

equation image(8)

Multiple imaging studies were done to test different aspects of the new technique, including reduced-FOV preparation (interleaved multiple IVI), blurring (resolution loss) in the slice-encoding direction, and applicability to DWI and DTI.

The quality of the images acquired by 3D ss-DWSTEPI is strongly dependent on the reduced-FOV preparation. A cylindrical agar phantom with a T1 of about 2.0 s and containing numerous tubes of various diameters running parallel to the cylinder axis was imaged to test the reduced-FOV technique incorporated in 3D ss-DWSTEPI. The phantom was positioned parallel to the magnet-bore direction, and an oblique imaging volume with 24 slices and 17% slice oversampling was acquired.

The standard reduced-FOV (inner volume) preparation (18, 19), which uses a single 180° inversion RF pulse with the slice-selection gradient applied in the phase-encoding direction, cannot be used to acquire multiple slices or slabs in a time-efficient interleaved mode. Our implementation of inner volume preparation with double inversion is applicable for interleaved multiple volume imaging. To compare these techniques, 3D ss-DWSTEPI with the standard and the new reduced-FOV preparation were used to image various numbers of localized volumes in one interleaved acquisition. The scan parameters for the experiment were as follows: imaging matrix = 192 × 48 × 8, TR = 2000 ms, and TE = 75 ms. The total data acquisition duration for each slab was about 300 ms, which limited the number of interleaved slabs to six for the given TR. The mean signal intensity was measured in a circular ROI selected in a central slice of the imaged slab located at the center of the phantom.

Additionally, four slabs were imaged to demonstrate the feasibility of interleaved multiple IVI to acquire contiguous volumes in an interleaved mode. Diffusion weighting was applied along the magnet-bore direction with b = 500 s/mm2. The imaging matrix was 192 × 48 × 12 with 17% slice oversampling, TR = 4000 ms, and TE = 75 ms. Two separate acquisitions (passes) were used to image slabs 1 and 3, and slabs 2 and 4, respectively.

To study resolution loss in the slice direction due to T1 decay and variable flip-angle excitations, the stimulated-echo trains were measured with phase- and slice-encoding gradients off and on using the same agar phantom as in the previous experiments. The object was scanned in an oblique imaging plane with a 192 × 48 × 24 imaging matrix and 17% slice oversampling.

To demonstrate that the new technique produces high-quality DTI measurements, a freshly excised animal heart with T1 and T2 relaxation times similar to those of human tissues was imaged using a 192 × 48 × 16 acquisition matrix with 13% slice oversampling, TR = 4000 ms, TE = 75 ms, eight averages, and diffusion encodings with b = 0 and 750 s/mm2 along seven noncollinear directions. The spatial resolution of the study was 1.0 × 1.0 × 1.5 mm3. A transmit-receive wrist coil (MRI Devices, Waukesha, WI, USA) was used in this study.

To display the in vivo DTI capability of the new acquisition technique, 3D ss-DWSTEPI was applied to obtain the DW images of the midbrain of a healthy volunteer using an eight-channel receive-only head coil (MRI Devices, Waukesha, WI, USA) and the following scan parameters: 192 × 48 × 16 imaging matrix with 13% slice oversampling, 1.25-mm isotropic spatial resolution, TR = 4000 ms, and TE = 75 ms. Diffusion encoding was accomplished with b = 0 and 400 s/mm2 along seven diffusion-encoding directions. The procedure was approved by the institutional review board, and the volunteer gave informed consent. The imaging time was 4 min 20 s for eight averages (magnitude averaging). Further investigation is necessary to study the potential effect of various motions, as might be observed in in vivo applications to other organs, on the accuracy of DTI measurement.


The images shown in Fig. 4a were obtained by 3D ss-DWSTEPI from the volume selected by our technique for reduced-phase FOV preparation. The prescribed volume is indicated by the dotted box in Fig. 4b. The resulting images are of good quality and lack aliasing along the phase-encoding direction from the phantom regions external to the prescribed volume. This result demonstrates the applicability of our reduced-FOV preparation scheme to limit FOV in the phase-encoding direction. Such a restricted in-plane FOV can be sampled by a short EPI readout resulting in significantly reduced image distortion due to local magnetic field susceptibility. The degree of distortion in the images acquired by 3D ss-DWSTEPI was comparable to that of 2D ss-DWEPI with analogous EPI readout duration. Note that the 3D ss-DWSTEPI data acquisition was accomplished with 15 applications of the excitation RF pulse followed by the EPI acquisition of 31 gradient echoes. The duration of each segment, including the RF pulse and complete ky acquisition of 31 gradient echoes, was about 38 ms. The total duration of the 3D data acquisition was around 560 ms for the 15 actual slice-encodings required to reconstruct a 24-slice volume.

Figure 4.

DW images of a limited volume acquired using interleaved multiple inner volume 3D ss-DWSTEPI from an agar phantom in an oblique plane with b = 100 s/mm2 in the S/I direction. The horizontal dotted line in a demonstrates the angulation of the holes passing through the slab. The dotted box in b shows the orientation of the imaged slab with annotated encoding directions.

As shown in Fig. 5, the signal loss in interleaved multislab imaging with reduced-FOV preparation was substantially reduced using the new technique with double inversion (□) compared to the rapid signal decay for the standard method with a single inversion pulse (Δ). Adiabatic RF pulses with 5.12-ms duration were used to implement double inversion reduced-FOV preparation. The separation between the inversion RF pulses in the new technique was around 6.0 ms.

Figure 5.

Signal intensity vs. the number of interleaved slabs for 3D ss-STEPI with the standard (Δ) and new (□) reduced-FOV preparations. Vertical axis represents the normalized signal intensity. The dotted lines are approximations of the data: a linear trend for the new method, and exponential decay for the standard single inversion technique can be observed.

Figure 6 illustrates two sets of 12 slices from four contiguous 12-slice slabs. Two separate acquisitions (passes) were used to image slabs 1 and 3 in the first acquisition, and slabs 2 and 4 in the second. The signal loss evident on the edge slices of each slab is due to the RF profile variation and is a common problem in most 3D imaging methods.

Figure 6.

DW images of slabs 2 and 4 of interleaved multislab 3D ss-DWSTEPI for four slabs, in two acquisitions with b = 500 s/mm2.

The peak amplitudes of the stimulated echoes are plotted in Fig. 7a with respect to their time of occurrence relative to the excitation RF (t = 0) in DPDE preparation. The amplitude of the later echoes of the 3D data readout was about 40%, compared to the first echo (kz = 0). The corresponding point-spread function (PSF) is shown in Fig. 7b. The full width at half maximum (FWHM) for the PSF was about 1.8 pixels, which indicates that image blurring in the slice direction was mild. The 3D interleaved multiple inner volume ss-DWSTEPI images shown in Fig. 4a (xy-plane) and Fig. 7c (xz-plane) demonstrate high resolution without any noticeable blurring in the slice-encoding direction.

Figure 7.

a: Plot of the peak magnitude of the 15 stimulated echo trains of 3D ss-DWSTEPI. b: The corresponding PSF. c: The xz-plane images (the corresponding xy-plane images are shown in Fig. 4a).

The results from the DTI study of a canine heart ex vivo are shown in Fig. 8. There are some residual aliasing artifacts in the left portions of the images along the phase-encoding direction. The helical structure of the myocardial muscle is well presented in the color map, in agreement with previous results from excised animal hearts (25, 26).

Figure 8.

DW images of a fresh excised canine heart: (a) eight central slices from 16 contiguous slices covering 24 mm, (b) DWI (b = 0 and 750 s/mm2 with diffusion encodings along seven noncollinear directions), and (c) FA and RGB representation of the principal eigenvector. Phase-encoding was along the up–down direction. There was an air pocket inside the left ventricular cavity (arrow). Color coding: red = horizontal, green = vertical, blue = in-out of the image plane.

Images from the DTI study of the midbrain of a healthy volunteer are presented in Figs. 9–11. Note that the images were acquired using a head coil, in which the signal reception sensitivity rapidly drops near the mid level of the cervical spinal cord. Figure 9 shows nine central slices from 16 contiguous slices covering a 20-mm-thick slab. The bright signal indicated by the arrow appears to be a susceptibility-induced artifact.

Figure 9.

Nine contiguous DW images acquired by 3D ss-DWSTEPI with b = 400 s/mm2 along the slice-encoding direction. The arrow indicates a susceptibility artifact induced by an air/tissue interface.

Figure 10.

DW images of a slice with b = 0 and 400 s/mm2 in seven encoding directions. The third image demonstrates the largest signal loss in the pons and the corpus callosum (as indicated by arrows) that occurred when the diffusion gradient was applied parallel to the WM fibers.

Figure 11.

DTI measurements of a healthy human midbrain: (a) FA and (b) principal eigenvector maps. The corpus callosum and cervical spinal cord demonstrate high FA values. Color coding: red = L/R, blue = S/I, green = A/P.

DW images of the central slice are shown in Fig. 10 for b = 0 s/mm2 and 400 s/mm2 for seven noncollinear directions. DW images were processed to estimate DTI parameters, such as eigenvectors, eigenvalues, and FA values. The resultant FA maps and RGB colored maps of the principal eigenvector are presented in Fig. 11 for the central nine slices, which completely cover the cervical spinal cord in the transverse direction. These results are very promising for in vivo human applications of 3D ss-DWSTEPI for high-resolution DTI.


Imaging techniques that acquire the complete 2D k-space after a single DPDE preparation, such as DPDE gradient echo or DPDE stimulated-echo acquisition mode (STEAM), suffer from low SNR because the frequent application of RF pulses causes a rapid decay of the diffusion-prepared longitudinal magnetization (10). The newly developed 3D ss-DWSTEPI technique requires much fewer RF pulses to acquire the complete 3D k-space by multiple short EPI readouts. Because of the relatively long duration of the 3D data acquisition, there was concern about potential resolution loss in the slice direction. However, the diffusion-prepared longitudinal magnetization in 3D ss-DWSTEPI experiences T1 decay, as indicated in Eq. [2], which is typically on the order of 1.0 s at 3T for most biological tissues. We chose to use a greater flip angle near the center of k-space and still increase the flip angle toward the edge of k-space (somewhere between the constant flip angle and the optimum variable flip angle). This results in an increase in signal near the center of k-space and a decrease for the outer planes of kz. Although some blurring occurs, there is still excellent resolution in the slice direction, as indicated in Fig. 4. Further investigation may be necessary to study the relationship among the choice of flip angles, the SNR, and the spatial resolution in slice direction on the resultant images.

Given multiple receiver coils with appropriate sensitivity profiles, it is possible that parallel imaging could be used to increase the FOV in the phase- or slice-encoding directions or to obtain a further decrease in data sampling time. The successful application of parallel imaging will depend directly on the relative variation of the receiver sensitivities over the desired FOV.

As shown in Fig. 7a and b, the blurring in the slice direction can be reduced by decreasing the number of slice encodings. It can also be improved by using a ramped flip-angle scheme with an optimal starting flip angle. In that case, however, the number of averages should be increased to obtain an acceptable SNR.

The image distortion observed in STEPI is a function of the number of echoes in the EPI echo train. Because there are typically more phase encodings than slice encodings, the number of echoes in the EPI echo train can be reduced by interchanging phase and slice encoding. With this switch, slice encoding is performed in conjunction with the EPI readout, and one phase encoding is applied for each EPI segment.

The 3D ss-DWSTEPI pulse sequence was able to acquire accurate DTI measurements of a localized volume without severe susceptibility or motion-related artifacts because of the short EPI echo train length and the immunity of the SS acquisition to the motion-related phase errors. It was demonstrated that localized 3D ss-DWSTEPI may be useful in DTI of any local anatomic region (as in Figs. 8 and 9) and DTI visualization (Fig. 11). Interleaved imaging of multiple subvolumes was successful. As is common in 3D MRI, the edge slices of each slab showed decreased signal intensity due to the spatial profile of the RF pulses.

A reduced EPI ETL and improved SNR will be crucial if EPI is to be used to perform high-resolution DTI. The geometric distortions in 2D ss-DWEPI induced by Bo field inhomogeneity can be reduced by shortening the EPI readout. If the number of phase encodings is reduced to shorten the EPI readout, the imaging FOV must be reduced to avoid aliasing artifacts. Parallel imaging and IVI have been used to resolve this problem while at the same time providing high resolution and strongly reducing artifacts caused by Bo field inhomogeneity. EPI with parallel imaging has been successfully applied to high-resolution brain DWI and DTI studies and resulted in substantial improvements in image quality (6, 7). EPI with parallel imaging has been used for cervical spinal cord DWI at 1.5 Tesla and yielded good results with minimal geometric distortion (27). The reduction factor in the study was limited to 2, resulting in relatively low in-plane resolution (2 mm). To achieve higher spatial resolution with the same EPI readout duration, the parallel imaging reduction factor should be increased. However, this approach requires highly specialized phased-array coils, in which sensitivities change substantially and differently along the phase-encoding direction in the imaged FOV to allow reliable reconstruction.

IVI techniques limit the excited FOV in the phase-encoding direction to include only the anatomy of interest (18, 19). The main disadvantage of standard IVI techniques is the low time efficiency due to the inability to perform interleaved multislice data acquisition. The ZOnally Oblique Multislice (ZOOM) EPI method was proposed to resolve this problem by obliquely applying phase-FOV (28, 29); however, it is mainly useful when the imaging plane is orthogonal to the main orientation of a predominantly 1D anatomical structure (e.g., coronal for optic nerve imaging, axial for spinal cord studies). This imaging orientation is not optimal because it requires a large number of slices and a long scan time to cover the structure of interest. Recently, 2D ss-DWEPI with reduced-phase FOV was reported for high-resolution DTI of a limited FOV with reduced ETL, which utilized an effective interleaved multislice imaging technique by incorporating double inversion (19). This technique allows time-efficient interleaved multivolume IVI because the two refocusing pulses with slice selection along the phase-encoding direction that are used to create the limited FOV also return most of the out-of-slab magnetization to the longitudinal direction. In the present study, the signal loss in interleaved multiple IVI was further reduced by applying the double inversion pulses immediately after the initial excitation 90° pulse with a very short interpulse delay. In the future it may be possible to reduce the out-of-slab signal loss by using 2D volume-selective excitation (30).


Three-dimensional ss-DWSTEPI can acquire the DW magnetization of a localized volume after a single diffusion preparation. Even though spatial coverage is limited in the phase and slice directions, the FOV in the readout direction is totally arbitrary and is limited only by the desired image dimensions and the sensitivity volume of the receiver coils. This new technique not only reduces susceptibility artifacts by using significantly shortened EPI readouts, it also freezes most of the physiologic motion by using a single, short data acquisition. Three-dimensional ss-DWSTEPI can be useful for high-resolution 3D DTI of limited volumes of interest, such as localized brain regions, cervical spinal cord, optic nerve, heart, or other extracranial organs.