Simultaneous multi-segment (SMSeg) EPI over multiple focal regions

Objective. This study aimed at developing a simultaneous multi-segment (SMSeg) imaging technique using a two-dimensional (2D) RF pulse in conjunction with echo planar imaging (EPI) to image multiple focal regions. Approach. The SMSeg technique leveraged periodic replicates of the excitation profile of a 2D RF pulse to simultaneously excite multiple focal regions at different locations. These locations were controlled by rotating and scaling transmit k-space trajectories. The resulting multiple isolated focal regions were projected into a composite ‘slice’ for display. GRAPPA-based parallel imaging was incorporated into SMSeg by taking advantage of coil sensitivity variations in both the phase-encoded and slice-selection directions. The SMSeg technique was implemented at 3 T in a single-shot gradient-echo EPI sequence and demonstrated in a phantom and human brains for both anatomic imaging and functional imaging. Main results. In both the phantom and the human brain, SMSeg images from three focal regions were simultaneously acquired. SMSeg imaging enabled up to a six-fold acceleration in parallel imaging without causing appreciable residual aliasing artifacts when compared with a conventional gradient-echo EPI sequence with the same acceleration factor. In the functional imaging experiment, BOLD activations associated with a visuomotor task were simultaneously detected in two non-coplanar segments (each with a size of 240 × 30 mm2), corresponding to visual and motor cortices, respectively. Significance. Our study has demonstrated that SMSeg imaging can be a viable method for studying multiple focal regions simultaneously.


Introduction
Single-shot gradient-echo echo-planar imaging (ss-GRE-EPI) pulse sequence has been widely used in perfusion imaging and functional neuroimaging due to its fast acquisition speed, robustness against motion, and relatively high signal-to-noise ratio efficiency (Kwong et al 1992, Bernstein et al 2004, Feinberg and Yacoub 2012. ss-GRE-EPI is typically performed with a full field-of-view (FOV) covering the entire anatomy such as the brain. To achieve an adequate spatial resolution, a long readout echo train is typically employed, which is subject to T2 * decay. Moreover, since ss-GRE-EPI has a low bandwidth along the phase-encoded direction, off-resonance effects caused by main magnetic field inhomogeneities, magnetic susceptibility differences among tissues, concomitant magnetic fields, and other factors can lead to severe in-plane geometric distortion (Bernstein et al 2004). When a focal region within the imaged organ is of interest, reducing the FOV is an effective strategy to alleviate the aforementioned problems, and consequently has been increasingly used in several applications (Alley et al 1997, Rieseberg et al 2002, Saritas et al 2008, Finsterbusch 2010, Finsterbusch 2013, Islam and Glover 2016, Zhong et al 2019. A common approach to reducing the FOV is to employ a two-dimensional (2D) RF pulse so that the spatial excitation extent along the phase-encoded direction can be limited in a selected slice. The reduced FOV (rFOV) allows a shorter readout echo train to be used, which decreases geometric distortion, or uses the same readout echo train to achieve a higher spatial resolution. Existing rFOV techniques are limited to a specific focal region. In some applications, more than one focal region are needed (Christensen et al 2007). For example, to study visuomotor interactions, functional activations in the visual and motor cortices need to be evaluated simultaneously. Although the conventional rFOV approach can be used successively by focusing on one region at a time (Finsterbusch 2013, Islam andGlover 2016), its time efficiency is poor. To address this issue, Finsterbusch proposed a method to simultaneously excite multiple isolated brain regions using a 2D RF pulse, which was designed by applying a Fourier transform of the prescribed regions of interest (ROIs) (Finsterbusch 2015). While this approach offers flexibility in selecting the targeted focal regions, it requires the unwanted outer segments be placed outside the object to avoid perturbance. Herein we propose an alternative method, coined simultaneous multi-segment (SMSeg) imaging, where the side excitation bands are exploited to simultaneously excite multiple focal ROIs within a volume. These focal ROIs or segments, each having the benefits of rFOV imaging, can be separated using a parallel image reconstruction algorithm and visualized in a composite 'slice'. The proposed approach does not need to re-design the RF pulse, while still achieving multifocal coverage. We have validated the SMSeg technique on a phantom and illustrated its applications for anatomic imaging and fMRI of the human brain.

SMSeg
Unlike conventional rFOV imaging with a 2D RF pulse (figure 1(A)), where the side bands (i.e. replicates) in the excitation profile are problematic, SMSeg makes use of the side bands to achieve SMSeg excitation. The segments, which are designed to be obliquely distributed within a volume, can cover isolated regions in different slices. The signals from different segments, which do not overlap in the phase-encoded direction, are projected into one composite 'slice', resulting in efficient acquisition of multiple focal regions using a single excitation. Further, the composite 'slice' acquisition can be accelerated by parallel imaging in a manner analogous to GRAPPA (Griswold et al 2002). Under-sampling of the composite 'slice' along the phase-encoded direction results in aliasing of the simultaneously excited segments. The tilted spatial distribution of the segments (figure 1(C)) enables the use of coil spatial sensitivity variations in both phase-encoded and slice-selection directions to resolve the aliasing and increase the in-plane acceleration factor for the composite 'slice' (see the Image Reconstruction subsection below). SMSeg combined with parallel imaging can lead to high sampling efficiency for simultaneously imaging different focal regions, while reducing image distortion.
RF pulse and pulse sequence design of SMSeg 2D RF pulse design is essential to implementing the SMSeg technique. A 2D RF pulse was designed by employing a fly-back EPI-like excitation k-space trajectory to avoid Nyquist ghosts (Finsterbusch 2010, 2013, Zhou and Sui 2017, Zhong et al 2019, Sun et al 2022a, 2022b) ( figure 1(A)). Both the sub-pulses and the envelope pulse were designed with a linear phase using a Shinnar-Le-Roux algorithm (Pauly et al 1991). Discrete sampling of excitation k-space along the blipped gradient direction resulted in periodic replicates (or side segments) of the excitation profile (main segment; figure 1). To flexibly control the locations of the segments, an excitation k-space sampling trajectory was designed by rotating a 2D square excitation k-space raster with an angle of α ( figure 1(B), left) (Sui et al 2014, Zhou andSui 2017), followed by scaling the relative amplitude of the phaseencoding and slice-selection gradients ( figure 1(B), right). The original unrotated and unscaled k-space yielded a set of excitation bands (or segments) along the phase-encoded direction as shown on the left of figure 1(C). After applying k-space rotation with an angle α, the corresponding excitation bands are rotated by the same angle as governed by the rotation theorem of Fourier transform (see the middle of figure 1(C)). Note that although the bands are rotated about the center of the FOV, each band itself is not rotated and remains parallel to the y and z axes. The subsequent scaling of gradients changes the aspect ratio of the excitation profile, producing a final tilt angle β as shown on the right of figure 1(C).
The total extent of the excitation k-space traversed along the phase-encoded and slice-selection directions, which is determined by the width of each segment (Bernstein et al 2004), is denoted as K y and K , z respectively ( figure 1(B)). After rotating and scaling the gradient, the final tilt angle β is related to the original rotation angle α by equation (1) The perpendicular distance ΔK between two neighboring excitation k-space lines is related to the original k-space interval ΔK orig by equation ( The rotated and scaled excitation segments in the image domain were simulated by Bloch equations (figure 1(C)). The center-to-center separation between the main segment and the first side segment is determined by the inverse of ΔK: where N is the number of sub-pulses. By controlling the two angles α and β, the main and the side segments closer to the FOV center can be positioned at the targeted focal regions, and the outer side segments can be A 2D RF pulse that produces a central spatial response and multiple replicates in the plane defined by the slice-selection (i.e. z-direction) and phase-encoding (i.e. y-direction) axes. (B): the excitation k-space trajectory was first designed as a square in the plane defined by the phase-encoded (k y ) and slice-selection (k z ) directions, and then rotated by α degree in the k y -k z plane, followed by scaling the relative amplitude of the phase-encoding and slice-selection gradients to achieve a desired excitation k-space coverage. After scaling, the rotation angle became β, and the perpendicular distance between two neighboring excitation k-space lines changed from ΔK orig to ΔK. (C): the corresponding spatial profiles simulated using the Bloch equations. According to the properties of Fourier transform, k-space rotation by an α angle results in the excitation profile to rotate by the same angle (from the image on the left to the image in the middle). A subsequent gradient scaling produces a final tilt angle β. The center-to-center distance between two adjacent segments is determined by 1/ΔK. placed outside the object to avoid perturbance. The individual segments at the targeted regions can be projected onto a composite slice for display (figure 2(A)). The 2D RF pulse was incorporated into a commercial ss-GRE-EPI pulse sequence, producing an SMSeg sequence for experimental demonstrations. In our implementation, a total of three segments were used, although more segments could also be incorporated into the SMSeg sequence.

Image reconstruction
For SMSeg EPI, the composite slice with full k-space sampling was reconstructed by a Fourier transform. When combined with parallel imaging with sparse k-space sampling, the aliased signals caused by under-sampling in the phase-encoded direction were resolved by a GRAPPA algorithm (Griswold et al 2002). The RF pulse spatial profile in SMSeg EPI was employed to improve the unaliasing results. Let the spatial sensitivity at a specific pixel in the main segment be C (x 0 , y 0 , z 0 ), the corresponding pixels in the two nearest side segments can be expressed as C (x 0 , y 0 -S · cosβ, z 0 + S · sinβ) and C (x 0 , y 0 + S · cosβ, z 0 -S · sinβ), respectively. Without losing generality, let us consider an under-sampling factor of two for the composite slice acquisition with a two-coil receiver array. Each aliased pixel in the component coil images (I 1 and I 2 ) can be expressed as a superimposition of spin densities (ρ 1 and ρ 2 ) weighted by their spatial sensitivities (C 11 and C 12 , C 21 and  (4) illustrates that sensitivity variations in both the phase-encoded and slice-selection directions can be exploited in reconstructing SMSeg images. The above description can be readily extended to SMSeg imaging with a larger acceleration factor and/or a larger number of receiver coil elements. The image reconstruction algorithm was implemented in MATLAB (The MathWorks, Inc., Natick, Massachusetts, USA) using custom programs.

Experiments
The SMSeg pulse sequence based on ss-GRE-EPI was implemented on a GE MR750 3 T scanner (GE Healthcare, Waukesha, Wisconsin, USA) with a maximal gradient strength of 50 mT m −1 and a maximal slew rate of 200 T m −1 s −1 . Phantom and in vivo human experiments were carried out using a 32-channel head coil (Nova Medical, Inc., Wilmington, MA, USA). All human scans were conducted on healthy volunteers under approval by the Institutional Review Board (IRB) with written informed consent. The first experiment was performed on a 17 cm (inner diameter) phantom filled with silicone to validate the pulse sequence and its associated image reconstruction algorithm. For the RF excitation pulse design, eleven sub-pulses, each with a pulse width of 1 ms and a time-bandwidth product (TBP) of 3.0, were modulated by an envelope pulse with a pulse width of 21 ms and a TBP of 4.2 ( figure 1(A)). The excitation k-space was sampled by a tilted trajectory using a rotation angle α of 50.0°and a resultant β of 8.6°after gradient scaling. The spatial response of the 2D RF pulse (figure 1(C)) was simulated using Bloch equations with customized programs written in MATLAB. Three obliquely distributed segments in different axial slices (figure 2(A)) were simultaneously excited and acquired with the following parameters: repetition/echo time (TR/TE) = 2000/ 30 ms, flip angle = 90°, segment size = 44 × 44 × 4 mm 3 , acquisition FOV of the composite slice = 220 × 220 mm 2 , matrix size = 192 × 192, in-plane spatial resolution = 1.15 × 1.15 mm 2 , slice thickness/spacing = 4/ 1 mm, number of composite slices = 10, and scan time = 2 s (i.e. one TR). Further, SMSeg EPI was combined with parallel imaging with acceleration factors of 2, 4, and 6 (i.e. acquisition FOV = 220 × 110 mm 2 , 220 × 55 mm 2 , and 220 × 37 mm 2 ), respectively. To compare the image quality among different acceleration factors, conventional ss-GRE-EPI acquisitions were performed at the central segment location with the same parameters as in the SMSeg EPI acquisitions.
The second experiment was to illustrate the ability of SMSeg EPI to acquire anatomic images from multiple focal regions of the brain. The phantom imaging protocol of SMSeg EPI was employed except for: segment size = 48 × 48 × 4 mm 3 , FOV = 240 × 240 mm 2 , matrix size = 120 × 120, and in-plane spatial resolution = 2 × 2 mm 2 . Figure 3(A) shows the locations of the segments for multi-slice acquisition, with different colors corresponding to individual composite slices (only three composite slices are shown for simplicity, although ten composite slices were acquired). The three simultaneously acquired segments in each composite slice were positioned at the frontal, temporal, and occipital lobes, respectively. Similar to the phantom experiment, SMSeg EPI with 2-, 4-, and 6-fold accelerations was successively applied to the human brain.
The third experiment was designed to evaluate the feasibility of using SMSeg EPI to acquire BOLD images from multiple isolated brain regions. A SMSeg EPI sequence was designed to simultaneously acquire three axial segments covering visual and motor cortices (figure 4(A)) at different axial-slice locations, with a high temporal resolution of 80 ms. Note that the FOV of the composite slice was reduced by a factor of two in the phaseencoded direction because one half of the full FOV (240 × 120 mm 2 ) was sufficient to cover the targeted visual and motor cortices. Such FOV reduction can decrease the image distortion (Sun et al 2022a). Twenty-one subpulses (0.7 ms each) with a TBP of 3.01 were modulated by an envelope pulse whose TBP was 5.74 and pulse width was 28.7 ms. The excitation k-space was sampled by a tilted trajectory with a tilt angle α of 70.0°(β = 34.9°a ccording to equation (1)). The other imaging parameters were: TR/TE = 80/32 ms, flip angle = 30°, segment size = 30 × 30 × 4 mm 3 , acquisition FOV = 240 × 60 mm 2 (i.e. the total acceleration factor = 4), matrix size = 120 × 30, in-plane spatial resolution = 2 × 2 mm 2 , and slice thickness = 4 mm. : two representative slices of SMSeg echo-planar images (selected from a total of 10 slices) from a healthy human brain (24 year-old male) with different acceleration factors (AF). The three segments were positioned at the frontal, temporal, and occipital lobes, respectively, as shown in (A). A relatively large field of view (FOV, 240 × 240 mm 2 ) was used to cover the three prescribed segments in different axial planes. As the AF increased, geometric distortion decreased as expected, especially at the frontal lobe.
Visual stimulation in the third experiment was delivered using a commercial system (SensaVue, Invivo Corporation, Gainesville, Florida, USA) with a dark-gray and light-gray checkboard pattern flashing at 8 Hz. Our block-design paradigm contained four 48 s blocks, each with a 24 s rest period and a 24 s stimulation. The total acquisition time was 3 min and 12 s. A total of three healthy human subjects (age range: 32-58 years) underwent the fMRI scan using SMSeg EPI. All subjects were instructed to fixate on the cross-hair presented at the center of visual field while performing hand open-close movement simultaneously during the stimulus epoch. All data were analyzed using SPM8 on MATLAB. Motion correction and spatial smoothing (full width at half maximum (FWHM) = 6 mm) were applied to the magnitude SMSeg images, followed by statistical analyses using a general linear model for activation detection with an FWE-corrected p-value threshold of <0.001 and a spatial cluster size of at least 30 pixels. Figure 2 shows a set of representative images obtained using SMSeg EPI without and with acceleration, together with those from conventional ss-GRE-EPI for comparison. In SMSeg imaging, three segments from different axial slices were simultaneously acquired and displayed as one composite slice ( figure 2(A)). The corresponding composite slices with acceleration factors of 1, 2, 4, and 6 are displayed in figures 2(B)-(E), respectively. With GRAPPA reconstruction, SMSeg EPI successfully resolved aliasing up to an acceleration factor of 6. In contrast, aliasing artifacts on conventional ss-GRE-EPI images were visible with a four-fold acceleration (figure 2(H)), and became unacceptable with an acceleration factor of 6 (figure 2(I)). Our results indicated that the SMSeg excitation pattern helped support a higher acceleration factor in parallel imaging. The SNRs of the SMSeg images with acceleration factors of 1, 2, 4, and 6 at the central segment were 96.2, 83.9, 62.2, and 59.4, respectively, which were lower than those of conventional ss- 90.5,69.5,and 66.0, respectively) by approximately 11.6% on the average.

In vivo results
In vivo anatomic brain imaging results from a representative healthy human subject (24 year-old male) are displayed in figure 3. Figure 3(B) shows that the three simultaneously acquired segments covered multiple brain regions (i.e. the frontal, temporal, and occipital lobes) at different axial slice locations. As the acceleration factor increased, no obvious aliasing artifacts were observed even with an aggressive six-fold acceleration. The increased acceleration factor led to a progressive reduction in geometric distortion, especially at the frontal lobe, as expected. Figure 4 shows the SMSeg fMRI results from the visual and motor cortices of a representative subject. As illustrated in figure 4(B), three non-coplanar regions in the occipital and parietal lobes were covered by an axial composite slice with a small FOV (240 × 120 mm 2 ), demonstrating the flexibility in FOV selection of the SMSeg sequence. fMRI activations associated with the visuomotor task were detected in the visual cortex, primary motor cortex, and supplementary motor area, as shown in the bottom and the top segments ( figure 4(C)). The absence of BOLD signal was also observed in the middle segment (i.e. the parietooccipital sulcus region) where functional activation was not expected. To illustrate the reproducibility of the fMRI results, activation maps from two additional subjects are shown in the supplementary figure S1. All three subjects in this study produced consistent activation results, demonstrating that SMSeg EPI was capable of focusing on multiple isolated brain activation regions using a single composite slice.

Discussion
Our study illustrated a technique-SMSeg in which multiple segments at different slice locations were simultaneously excited to image multiple focal regions. A number of ROIs in the volume could be scanned with a high temporal resolution and displayed in a single composite slice. By taking advantage of coil sensitivity variations in both the phase-encoded and slice-selection directions, SMSeg was able to accommodate a greater acceleration factor than conventional parallel imaging techniques. rFOV imaging has been increasingly used to achieve high spatial resolution in a focal region and/or to reduce image distortion (Finsterbusch 2013, Islam and Glover 2016, Sun et al 2022a. However, the limited spatial coverage hampered the applications where multiple focal regions are desirable. The issue can be addressed by sequentially acquiring several isolated segments, each corresponding to one focal region. A major disadvantage, however, is the substantially increased scan times. To overcome the limitation, Finsterbusch proposed that multiple distributed regions can be excited by a tailored 2D RF pulse in which the envelope was calculated by taking the Fourier transformation of the target excitation profile (Finsterbusch 2015). This approach offers great flexibility in selecting the targeted focal regions, but requires RF pulse redesign should the targets change. Different from that approach, the SMSeg technique described herein takes advantage of periodic replicates of the excitation profile of a 2D RF pulse to simultaneously excite multiple isolated regions, which are subsequently reconstructed and displayed in one composite slice. Although three segments (one main segment and two side segments on each side) were employed in the experiments, additional side segments can be used to excite more regions of interest simultaneously by controlling the tilt angle and the distance between two neighboring segments, as described by equations (1)-(3). The different targeted focal regions are excited by rotating and scaling the gradients that are concurrent with the RF pulse waveform, thus it does not involve RF pulse waveform redesign. In addition, multi-slice acquisition is compatible with SMSeg imaging as shown in our in vivo studies. Similar to conventional 2D EPI, multiple composite slices can be acquired in sequential or interleaved manner to cover a focal volume. One limitation of the SMSeg technique is that the selected multiple segments are along an oblique line ( figure 1(C)). In principle, these linearly arranged segments may be covered in part using an oblique slice. However, oblique imaging planes, particularly those with a large tilt angle from the axial slice, are not typically used in single-shot EPI acquisitions because of distortion caused by concomitant magnetic fields (Zhou et al 1998, Du et al 2002, additional ghosts produced by gradient anisotropy Maier 1996, Zhou et al 1997), and increased likelihood of peripheral nerve stimulation.
Parallel imaging is a common imaging acceleration method for EPI to reduce geometric distortion. The acceleration factor, however, is limited by the elevated residual aliasing and other considerations such as the SNR. Compared to conventional in-plane parallel imaging, the excitation pattern in the SMSeg technique can exploit sensitivity variations along both the phase-encoded and slice-selection directions, allowing a more aggressive acceleration factor (e.g. 6) to be used. In addition, dark bands between neighboring segments are not excited, leading to reduced contribution to the folded image. The excitation pattern of the 2D RF pulse in SMSeg provides an additional way to abate aliasing, contributing further to a higher acceleration factor.
A popular technique to acquire multiple slices with a single excitation is simultaneous multi-slice (SMS) imaging (Breuer et al 2005). A major distinction between SMSeg and SMS resides in the zooming capability within the selected slices. Another distinction lies in the different performance in parallel acquisitions. To illustrate this, we carried out a comparison between SMSeg and SMS on the silicone phantom with a throughplane acceleration factor of 3 and in-plane acceleration factors of 1, 2, and 4, respectively. With an in-plane acceleration factor of 2 and 4, SMS exhibited substantial residual aliasing artifacts (see figure S2 in supplementary materials). In contrast, the aliasing artifacts were not seen in the SMSeg images (figures 2 and 3). When in-plane acceleration was not employed (i.e. acceleration factor = 1), quantitative SNR measurement in the phantom showed an approximately 6.6% reduction in the SMSeg image (SNR = 96.2) compared with that in the SMS image (SNR = 103.0).
Due to the finite length of the 2D RF pulse, each segment in SMSeg is subject to non-ideal excitation profile in the phase-encoded direction, which manifests itself as extended spatial excitation of magnetization. As the FWHM was used to define the width of the segment, a slowly rolling transition zone leads to a reduced average signal intensity in the entire segment and consequently a lower SNR. When compared with the full-FOV excitation, our simulation showed 6.3% reduction in SNR, which was less than the experimentally observed SNR reduction ( ). The additional SNR loss could be attributed to the nonuniform excitation of the 2D RF pulse due to gradient waveform infidelity caused by eddy currents and other adverse effects (Bernstein et al 2004). The excitation profile can be improved at the expense of a prolonged pulse duration. A lengthy 2D RF pulse, however, will limit the minimum repetition and echo times for the SMSeg EPI sequence, and increase the sensitivity to T2 * relaxation and flow effects. For example, a minimum TE of 32 ms was used in the third experiment to achieve an adequate BOLD contrast at an in-plane spatial resolution of 2 × 2 mm 2 . The minimum TE can be reduced to 29.3 ms and 27.5 ms when lower in-plane spatial resolutions of 2.5 × 2.5 mm 2 and 3 × 3 mm 2 are employed, respectively. On the other hand, a higher in-plane spatial resolution would lead to an increased TE and decreased SNR, which may compromise BOLD detectability.

Conclusion
We have demonstrated a novel SMSeg technique that excites multiple tilted segments simultaneously, while leveraging the coil sensitivity variations in both the phase-encoded and slice-selection directions for GRAPPAbased image reconstruction. SMSeg combined with parallel imaging has been demonstrated in anatomic and functional imaging of multiple focal brain regions, where a high acceleration factor up to six was achieved together with reduced geometric distortion. By incorporating other SMSeg techniques, such as the one described by (Finsterbusch 2015), the SMSeg technique has the potential to become a viable method for a broader range of applications, including anatomic, diffusion, perfusion, and functional MRI in which multiple focal regions can be studied simultaneously.