Three‐dimensional EPI with shot‐selective CAIPIRIHANA for rapid high‐resolution quantitative susceptibility mapping at 3 T

QSM provides insight into healthy brain aging and neuropathologies such as multiple sclerosis (MS), traumatic brain injuries, brain tumors, and neurodegenerative diseases. Phase data for QSM are usually acquired from 3D gradient‐echo (3D GRE) scans with long acquisition times that are detrimental to patient comfort and susceptible to patient motion. This is particularly true for scans requiring whole‐brain coverage and submillimeter resolutions. In this work, we use a multishot 3D echo plannar imaging (3D EPI) sequence with shot‐selective 2D CAIPIRIHANA to acquire high‐resolution, whole‐brain data for QSM with minimal distortion and blurring.


INTRODUCTION
QSM uses phase data from a T 2 *-weighted acquisition to determine relative magnetic susceptibility of tissues within a region of interest (ROI).In the brain, the standard MRI pulse sequence for acquiring data for QSM is a 3D gradient echo (3D GRE).For whole-brain coverage and an isotropic resolution of 1 mm, the resulting scan time can be in excess of 5 min.This increases to 8-15 min for whole-brain, submillimeter acquisitions that have demonstrated clinical applications in imaging multiple sclerosis (MS), 1 cerebral microbleeds, 2 and in presurgical mapping for deep brain stimulation. 3,4Long scans are more susceptible to physiological fluctuations and motion artifacts, which affects image quality and susceptibility values. 5otion in the scanner tends to increase with age in adults 6 and is an important consideration when imaging clinical populations, such as Alzheimer's disease, Parkinson's diseases, and MS, where involuntary motion is more common than in healthy control participants.
Both 2D 7 and 3D EPI 8 have been used to reduce acquisition times for QSM.Three-dimensional EPI is generally more SNR-efficient than 2D EPI, achieves thinner slices with better slice profiles, and has the advantage of being able to accelerate in both phase-encoding directions without complex multiband RF pulses.][8][9][10][11][12][13][14][15] Three-dimensional EPI has also been shown to be useful in clinical applications.Sati et al. used 3D EPI with 2D SENSE to image MS patients at 3 T and 0.55-mm isotropic resolution in just under 4 min. 16n MS imaging, T 2 *-weighted images allow for easy visualization of both the central vein sign 16 and the iron-rich paramagnetic rim that surrounds some lesions. 1,16,17 Wicaksono et al. found that an unaccelerated 1-mm isotropic 3D EPI QSM with 2 minute acquisition time was just as reliable as multi-echo 3D GRE QSM in detecting cerebral microbleeds. 9ecent technical advances include the use of multishot 3D EPI sequences with 2D CAIPRIHANA (short: CAIPI) acceleration to further accelerate acquisition times while keeping the geometry-dependent noise amplification factor (g-factor) and geometric distortions to a minimum.Multishot, interleaved 3D EPI with shot-selective CAIPI achieves a 2D-CAIPI pattern by selectively acquiring shots, rather than introducing additional k z -blips.This has been used to acquire high-resolution functional MRI data at 7 T 11 and T 1 -weighted images at 3 T. 13 Skipped CAIPI extends shot-selective CAIPI by adding kz-blips and has been used recently in multi-parametric mapping at 3 T and 7 T, 14 and submillimeter QSM at 7 T. 15 In this work, we implemented a 3D EPI with shot-selective 2D-CAIPIRIHANA acceleration for rapid, high-resolution (≤ 1 mm 3 ) QSM at 3 T.The sequence was tested in two parts: First, clinical viability was tested on a group of MS patients who were imaged at 1-mm isotropic resolution with a protocol optimized for acquisition time.The appearance of the susceptibility maps, the number of detectable lesions in the maps, and susceptibility values for different brain regions are compared between the 3D EPI sequence and multi-echo (ME) 3D GRE.Second, the protocol was optimized for submillimeter imaging in healthy subjects.We acquired whole-brain images at 3 T in healthy participants with 1-mm, 0.78-mm, and 0.65-mm isotropic resolution and compared the calculated susceptibility maps to ME 3D GRE.The trade-off between acquisition time and resolution, the number of EPI shots, and the effect of distortion on the calculated susceptibility maps is discussed.

METHODS
The 3D EPI sequence acquires each partition using multiple interleaved shots.Navigators are acquired for each shot, providing the prerequisite for odd-even echo corrections.Two-dimensional CAIPIRINHA 18 is implemented in a shot-selective 11 fashion, which can also be thought of as a simplified skipped-CAIPI without additional gradient blips. 14For a 3D EPI sequence with a k y in-plane phase direction and k z partition direction, the desired CAIPI pattern is achieved by skipping in-plane shots to create the desired in-plane acceleration, and offsetting the k y starting position of the acquired shots in each k z partition to achieve the CAIPI shift.In this way, the CAIPI pattern is achieved without the use of additional k z blips, unlike the 2D-CAIPI implementations in single-shot and k z -segmented 3D EPI sequences.However, this does place some restrictions on the available undersampling patterns described further in the Supporting Information and Figure S1.In the application of high-resolution scanning, in which short echo-train lengths and high shot numbers are generally used, these restrictions are rarely a limitation.Autocalibration lines are acquired to perform in-line GRAPPA image reconstruction 19 as part of the 3D EPI sequence using separate GRE acquisition.The time taken to do this is included in the total acquisition time (TA) listed for the protocols found subsequently.

MS cohort
As part of a larger study, 20

Submillimeter scans
To optimize the 3D EPI sequence for minimal geometric distortions, 4 healthy participants were scanned using the prototype 3D EPI sequence and a ME 3D GRE sequence at 1-mm and 0.78-mm isotropic resolutions.Images were acquired on a second MAGNETOM PrismaFit 3T MR scanner (Siemens Healthcare) located at a different site to the MS study, with a 64-channel head/neck coil.Scans were approved by the local ethics committee, with written, informed consent given by participants.For the 3D EPI sequence, a slab-selective binomial 1-2-1 water-excitation pulse with FA = 12 • and transmitter BW = 24 was used.
The CAIPIRINHA acceleration was a 2 × 2 z1 ≡ 4 × 1 y2 pattern.A complete list of parameters for these scans is given in Table 1.Due to the shot-selective CAIPI implementation, the number of acquired shots changes with acceleration factor.For completeness, the ETL, number of total shots, and number of acquired shots are all reported in Table 1.For example, the 3D EPI 1-mm isotropic resolution scan with an ETL = 3 requires 76 shots in an unaccelerated acquisition, and 19 shots with the 2 × 2 z1 CAIPI encoding.Three different ETLs were acquired for each resolution (Table 1).Partial Fourier in the k y direction was used in the 0.78-mm 3D EPI scans with ETL = 24 (Table 1) to allow the same TE = 25 ms to be used.For comparison, a ME 3D GRE was acquired at 1-mm and 0.78-mm isotropic resolutions, using a slab-selective binomial 1-2-1 water-excitation pulse with FA = 12 • and transmitter BW = 24, and 2×2 GRAPPA acceleration.For all the ME 3D GRE and 3D EPI images, the in-plane phase encoding direction was left-right, and the partition direction was head-foot.A T 1 -weighted MPRAGE was acquired at 0.78-mm isotropic resolution for image segmentation.
A comparison between GRAPPA and CAIPIRIHANA undersampling for 3D EPI at 1-mm and 0.78-mm for two different acceleration factors was also carried out.The detailed protocols for these are found in the Supporting Information Table S1.
A fifth participant was scanned at 0.65-mm isotropic resolution using the 3D EPI sequence FA = 16 • , Imaging parameters for the prototype 3D EPI and multi-echo 3D gradient-echo scans for the 4 healthy participants.

Image processing and analysis
Reconstruction of all images was performed on each scanner's image reconstruction computer.The 3D EPI multicoil data were combined using optimum phase combine, 21 and the ME 3D GRE phase data were combined over coils inline using ASPIRE. 22The resulting 3D EPI and 3D GRE magnitude and phase images were then processed offline with the QSMxT pipeline (version 1.3.5) to perform masking, susceptibility map calculation, registration, and image segmentation.To improve masking, a homogeneity correction 23 available in QSMxT was applied to the 3D EPI protocols on healthy subjects.The remaining protocols, including the 3D EPI MS patient scans, used prescan normalization 24 and did not need correction.The QSMxT two-pass masking method 25 was used to reduce streaking artifacts.Brain masks for each pass were determined by thresholding magnitude images, with threshold values optimized for each protocol by visual inspection of the masks.Susceptibility maps were produced using projection onto dipole fields 26 background-field removal and rapid two-step dipole inversion. 27For the 3D GRE, susceptibility maps were generated for each echo, and a combined map was computed as the average across all echoes.An affine registration of the T 1 -weighted image to 3D GRE or 3D EPI scans was performed in the QSMxT pipeline using Advanced Normalization Tools (ANTs). 28egmentation of the T 1 -weighted scan in QSMxT was done using FastSurfer. 29Finally, CSF from the lateral ventricles was used as a reference tissue. 30istortions for each 3D EPI scan were visually assessed by overlaying the edges of the last TE in the ME 3D GRE scan onto the 3D EPI image.Edges of the brain in the 3D GRE used for the overlays were found using the 3D Canny edge detection algorithm implemented in MATLAB.

MS cohort
The 3D EPI images for the MS patients were acquired in 28 s, compared with 5 min 3 s in 3D GRE scans.Figure 1 shows the distortions in the 3D EPI images from Patient 16 as a result of the longer ETL (57).These were as large as 9 voxels (9 mm) in the frontal cortex but were limited to the peripheral regions of the brain and did not appear to affect the position of deep brain nuclei.

F I G U R E 1
The 3D EPI magnitude images of an example multiple sclerosis patient acquired with echo train length (ETL) = 57.Overlaid in red are the edges of the 3D gradient-echo magnitude images at TE = 35 ms.White arrows indicate areas of distortion and signal dropout.The in-plane phase-encoding direction is anterior-posterior.TA, acquisition time.
Figure 2 demonstrates the performance of 3D EPI in comparison to the 3D GRE for imaging in MS, showing FLAIR, T 2 *-weighted images, and resultant susceptibility map for Patient 10.This subject had large lesions, easily visible at the 1-mm isotropic resolution, and lesions comparable in size to the imaging resolution.In cases such as this, submillimeter scanning could potentially improve lesion visualization.Images of other example patients can be found in the Supporting Information and Figures S9-S11.In Figure 2, most of the potential MS lesions easily seen on the FLAIR images were also observed as hypointense on the T 2 *-weighted images.However, not all of the lesions were seen in the susceptibility maps, as the appearance of MS lesions in QSM as hypointense, isointense, or hyperintense is dependent on the degree of demyelination, re-myelination, or iron accumulation, respectively. 31The 3D EPI susceptibility map is visually similar to the 3D GRE susceptibility map.All lesions that were visible on the 3D GRE susceptibility map were also visible on the 3D EPI susceptibility map.As a quantitative comparison between the 3D GRE and 3D EPI, the distribution of susceptibility values across the 19 MS volunteers for seven of the automatically segmented ROIs is shown in Figure 3.

The 1-mm and 0.78-mm isotropic resolution scans
No obvious distortions were seen in the 1-mm isotropic resolution ETL = 3 and ETL = 7 3D EPI scans, and a maximum displacement of approximately 3 voxels (∼3 mm) was observed in the ETL = 19 scan.These minor distortions in the longer ETL acquisition were limited to regions near the para-nasal sinuses and did not appear to affect the apparent position of the deep brain nuclei.Figure 4 shows the 3D EPI magnitude images after homogeneity correction at 0.78-mm isotropic resolution for Participant 1, acquired with ETLs of 3, 9, and 24.No obvious distortions were seen in the ETL = 3 scan, which was approximately twice as fast as the 3D GRE acquisition.Minor distortions were observed for both the ETL = 9 and ETL = 24 3D EPI acquisitions.The maximum displacement observed in the ETL = 24 scan was 6 voxels (∼4.5 mm) in the temporal lobe near the para-nasal sinuses.
A comparison of the 1-mm isotropic resolution 3D GRE and 3D EPI susceptibility maps for Participant 2 is shown in Figure 5.The QSMs produced from the 3D EPI scans are of comparable quality to those produced using the ME 3D GRE.Difference maps included in Figures S6  (A and S7 show the largest differences occurring in the veins and in the regions immediately around them. Examples of submillimeter QSMs for Participants 3 and 4 are shown in Figures 6 and 7, respectively.These figures compare the depiction of finer brain structures in the 3D GRE and 3D EPI acquisitions.At 0.78-mm isotropic resolution, the 3D EPI scans with ETL = 9 show comparable detail to the 3D GRE acquisition but with a 5.7-times reduction in acquisition time.Some blurring can be seen in the 3D EPI scans with ETL = 24 and 0.78-mm isotropic resolution as a result of the PF acquisition.However, detail that is not seen in the 1-mm resolution scans, such as the surface of the dentate nucleus (Figure 7), is still retained.
Figure 8 displays the distributions of susceptibility values across the 4 participants for the seven ROIs.In white matter, gray matter, ventricles, and the thalamus, susceptibility values from the ME 3D GRE and 3D EPI were comparable.For regions of higher susceptibility, the caudate, putamen, and pallidum susceptibility values from the 3D EPI QSMs is underestimated compared with the values from the ME 3D GRE QSMs.The variability of measured susceptibility values within a ROI was similar between the 3D GRE and 3D EPI scans.

F I G U R E 3
Combined susceptibility distributions of all 19 multiple sclerosis patients for seven automatically segmented regions of interest.Upper and lower whiskers and upper and lower box bounds represent the 95th, 5th, 25th, and 75th percentiles respectively.The black line represents the median susceptibility value.GRE, gradient echo; ME, multi-echo.

F I G U R E 4
The 3D EPI magnitude images after homogeneity correction of Participant 1 at 0.78-mm isotropic resolution acquired with echo train length (ETL) of 3 (left), 9 (middle), and 19 (right); overlayed in red are the edges of the 3D gradient-echo (GRE) magnitude images at TE = 25 ms.White arrows indicate areas of distortion and signal dropout.The in-plane phase-encoding direction is left-right.PF, partial Fourier; TA, acquisition time.

The 0.65-mm isotropic scan
The 3D EPI scans were acquired with an ETL of 13, and no distortions were observed when the corresponding 3D GRE scan was overlaid.Some residual phase errors as a result of the segmentation were observed in each of the 3D EPI acquisitions; these are shown in Figure S3. Figure 9 shows the calculated QSMs.The fully sampled and 1 × 2 z1 3D EPI scans were approximately 2 and 4 times faster than the comparison 3D GRE scan with 1 × 2 GRAPPA acceleration, respectively.Compared with the 3D GRE which used PF to decrease overall scan time, the 3D EPI QSMs appear sharper, and finer brain structures, such as in the dentate nucleus (top row, Figure 9), are clearly visible.Due to increased noise, some of this detail is lost in the fastest 3D EPI scan, which was 7.85 times faster than the 3D GRE.
However, in all 3D EPI scans, there is clear demarcation of the boundaries of the subthalamic nuclei from the substantia nigra.The susceptibility gradient, indicated by the white arrow in the 3D GRE scans in Figure 9, caused by a change in iron distribution along the length of the subthalamic nuclei, 4 is also clearly visible in all scans.To further demonstrate the depiction of finer brain structures' magnitude, 3D EPI images showing the swallow tail sign in healthy substantia nigra 32 are included in Figure S12.

DISCUSSION
Despite many applications, the long scan time associated with QSM, and in particular submillimeter QSM, is a major barrier to routine clinical use.In this study

F I G U R E 5
Multi-echo 3D gradient-echo and 3D EPI QSMs for Participant 2 at 1-mm isotropic resolution.Susceptibility values range from −0.15 to 0.15 ppm.Note that different masks were used, leading to different appearance at the tissue borders, such as near the sagittal sinus.ETL, echo train length.

F I G U R E 6
The 1-mm (top row) and 0.78-mm (bottom row) QSMs for Participant 3 acquired with multi-echo (ME) 3D gradient echo (left) and 3D EPI (middle, right).Susceptibility values range from −0.15 to 0.2 ppm.Note that different masks were used, leading to different appearance at the tissue borders.ETL, echo train length; PF, partial Fourier.

F I G U R E 7
The 1-mm (top row) and 0.78-mm (bottom row) QSMs for Participant 4 acquired with multi-echo (ME) 3D gradient-echo (left) and 3D EPI (middle, right).Susceptibility values range from −0.15 to 0.2 ppm.Note that different masks were used, leading to different appearance at the tissue borders.ETL, echo train length; PF, partial Fourier.

(A) (B) F I G U R E 8
Combined susceptibility distributions of all 4 participants for 1-mm (A) and 0.78-mm (B) isotropic resolutions.Upper and lower whiskers and upper and lower box bounds represent the 95th, 5th, 25th, and 75th percentiles, respectively.The black line represents the median susceptibility value.ETL, echo train length; GRE, gradient echo; ME, multi-echo.
we used an interleaved multishot 3D EPI sequence with a shot-selective 11 2D CAIPIRHIANA to achieve high-quality, submillimeter QSMs at 3 T in a fraction of the time required for ME 3D GRE.
The susceptibility maps produced from the 3D EPI scans were visually similar to those from the 3D GRE in both the MS patient cohort and in healthy participants.Differences between the ME 3D GRE and 3D EPI susceptibility maps in peripheral parts of the brain and the sagittal sinuses are due to the threshold-based masking method producing different masks for the 3D EPI and 3D GRE protocols.Additionally, some skull and more of the peripheral brain regions are included at shorter TEs in the 3D GRE masks.This is retained in the final, averaged susceptibility maps for the GRE acquisitions, and not observed in the 3D EPI susceptibly maps.The largest difference between the susceptibility maps derived from the ME 3D GRE and 3D EPI acquisition was the susceptibility of the vessels and surrounding regions.We also found a similar differences when comparing the susceptibility maps derived from the 3D GRE using the last echo only (TE = 25 ms) and 3D EPI.A number of factors may affect the appearance of vessels in susceptibility maps including: (1) different flow considerations in the 3D GRE and 3D EPI sequences; (2) poor phase unwrapping in the vessels in the 3D EPI acquisition and 3D GRE at later TEs due to low SNR, leading to unreliable susceptibility estimates; and (3) threshold-based masking techniques (such as the two-pass method), which might exclude vessels due to poor SNR in magnitude images.It was found that the difference between the vessels in the ME 3D GRE and 3D EPI susceptibility maps was reduced when using two-pass masking compared to masking with the Brain Extraction Tool 32 (Figure S8).This is because the two-pass The 3D EPI with an ETL of approximately 8 with CAIPI 2 × 2 z1 acceleration provided a good balance between acquisition time and distortion in the submillimeter scans.The scan time was reduced by a factor of 5 compared with the 3D GRE, with distortions of up to 1 voxel visible in some regions.Despite the use of TE shifting, 33 some residual segmentation artifacts could be seen in 3D EPI images.In the 1-mm and 0.78-mm scans, they tended to be more prominent when using longer ETLs (Figure S2) but were the worst in the 0.65-mm scans (Figure S3).The effect on the appearance of the susceptibility maps was minimal.Correction of the artifacts can be achieved using a dual-polarity readout at the cost of twice the scan time. 15At the 1-mm voxel size, there were no obvious differences in T 2 *-weighted images and susceptibility maps when using an acceleration pattern with or without a CAIPI shift.However, for submillimeter scans with acceleration at 2 × 2 and above, we found the CAIPI shift was necessary to preserve the appearance of finer structures in both the T 2 *-weighted images and the susceptibility maps (Figures S4 and S5).Increasing the ETL had no significant effect on the measured susceptibility values in the 3D EPI scans in the healthy participants but does result in (1) smaller available acceleration factors and less flexibility in the shot-selective CAIPI undersampling pattern and (2) increased distortion.For example, the four-shot (ETL = 57) acquisition used in the MS patient cohort scans resulted in distortions in cortical regions of the brain.MS lesions are found in all regions of the brain, including cortical gray matter, 34 and in cases where lesions are present in these regions, distortion can be corrected for using a number of techniques 35 or by using a smaller ETL with higher acceleration factor (to match acquisition times).
The use of long echo trains also sets a limit on minimum TE in the 3D EPI scans.A possible advantage of 3D GRE sequences for QSM over EPI sequences are ME acquisitions, particularly with short TEs and short echo spacing.ME-QSM acquisitions allow for echo-dependent susceptibility analysis and enable temporal phase unwrapping if the difference in consecutive TEs is short enough.Additionally, acquisition of T 2 * images with short TEs allows for more accurate susceptibility values in regions of tissue near strong susceptibility boundaries and high susceptibility sources.This is observed, for example, in the difference between the 3D GRE and 3D EPI susceptibility maps near veins in the current data sets.Short TEs may also be particularly important when imaging pathology such as microbleeds and in cases where tissues accumulate iron, resulting in higher susceptibility values.TEs of 5-10 ms in 3D EPI are limited to highly segmented protocols, with either longer TAs or higher acceleration factors.For example, recent multiparameter mapping work by Wang et al. acquired 1-mm isotropic resolution whole-brain images using a ME 3D EPI with skipped-CAIPI 13 using an ETL of 5. 14 The restriction shot-selective CAIPI places on available acceleration factors and patterns will be most apparent in long ETL sequences, where a small number of shots are used.For example, an acquisition in which all lines are acquired in a single shot cannot be accelerated using this implementation, and an acquisition in which two shots are used to acquire all phase-encoding lines can achieve a maximum acceleration factor of 2. Aside from this, the restriction on acceleration factors and patterns is also apparent in acquisitions if FOV and resolutions are not flexible parameters.For example, when comparing GRAPPA and CAIPI accelerations in the 0.78-mm scans (Table S1), the 2 × 2 z1 acquisition used an ETL of 9 and had a TA of 79 s.To match the FOV and resolution, the ETL in the equivalent 2 × 3 z1 acquisition had to be decreased to ETL of 7, such that the number of shots was a multiple of 6, and the resultant scan had a TA = 70 s.In this case, the need to decrease the ETL negated the time saved by increasing the acceleration factor.In practice, however, outside our stringent protocol comparison, minor adjustments of the FOV and resolution would be possible, allowing for flexible adaptations of the protocol design.Additional flexibility in the CAIPI pattern could be achieved using a skipped CAIPI acquisition, 13 which uses k z -blips to spread the EPI readout across multiple partitions.However, for submillimeter applications, emphasis on accurate visualization of brain structures and shorter ETL is more likely.Additionally, submillimeter acquisitions at 3 T are likely to be limited by SNR to a total acceleration factor of 4. As a result, shot-selective CAIPI should be sufficient to achieve most submillimeter QSM protocols at 3 T.

CONCLUSIONS
For routine QSM, and in particular submillimeter QSM at 3 T, the 3D EPI with shot-selective CAIPI provides excellent visualization of finer brain structures in a significantly shorter TA compared with the conventional 3D GRE.Optimization of the sequence for anatomical accuracy, speed, or a combination of both is a trade-off between a number of different considerations including image resolution, the number of shots, the resulting geometric distortions, and the desired TE for QSM calculation.

2
Fluid-attenuated inversion recovery (FLAIR; left), 3D gradient-echo (GRE) T 2 *-weighted (TE = 35 ms; top middle), 3D EPI T 2 *-weighted (TE = 37 ms; top right), 3D GRE susceptibility map (bottom middle), and 3D EPI susceptibility map (bottom right) for an example multiple sclerosis patient.All lesions visible in the 3D GRE susceptibility map are also visible in the 3D EPI susceptibility map and are indicated by orange arrows.ETL, echo train length.

F I G U R E 9
Isotropic 0.65-mm QSMs from multi-echo (ME) 3D gradient echo (GRE; left) and 3D EPI with varied acceleration patterns (right three rows).The top row shows clear delineation of the boundary of the dentate nucleus (DN), including the finer hyperintense structure at the DN surface.The left subthalamic nucleus (STN) is labeled in the 3D GRE scan (middle and bottom rows), and the white arrow indicates the susceptibility gradient along the length of the STN.Note that different masks were used, leading to different appearance at the brain borders.ETL, echo train length; TA, acquisition time.masking reduces streaking artifacts and errors surrounding high-susceptibility (low-signal) regions.The most likely cause of the discrepancy between vessel susceptibly values are unreliable susceptibility values in the vessels at the later TEs as a result of low SNR.In both the 3D EPI and 3D GRE TE = 25 ms echo, the susceptibly values in the vessels were inconsistent, containing regions of large negative directly adjacent to large positive values.This can be seen in the 3D EPI susceptibility maps in Figure 5 (Rows 1 and 2) and is in contrast to the smooth appearance of the susceptibility values in the larger vessels seen in the ME 3D GRE.

Figure S6 .
Difference maps between the 1-mm isotropic multi-echo gradient echo (GRE) and the 3D EPI scans for Participant 2. The slices are the same as those in Figure5.For clarity, only voxels that are present in both the 3D GRE and 3D EPI masks are included.FigureS7.Absolute difference maps between the 1-mm isotropic multi-echo gradient echo (GRE) and the 3D EPI scans for Participant 2. The slices are the same as those in Figure5.For clarity, only voxels that are present in both the GRE and 3D-EPI masks are included.FigureS8.Difference maps between the 3D EPI and multi-echo (ME) 3D gradient echo (GRE) using two-pass masking (left) and Brain Extraction Tool (BET) masking (right).The middle panel shows the difference between the 3D EPI susceptibility map and the susceptibility map derived from using the last echo only of the 3D GRE acquisition.Figure S9.Fluid-attenuated inversion recovery (FLAIR; left), 3D gradient echo (GRE) T 2 *-weighted (TE = 35 ms; top middle), 3D EPI T 2 *-weighted (TE = 37 ms; top right), 3D GRE QSM (bottom middle), and 3D EPI QSM (bottom right) for multiple sclerosis Volunteer 2. All lesions visible in the 3D GRE QSM are also visible in the 3D EPI QSM and are indicated by orange arrows.Figure S10.Fluid-attenuated inversion recovery (FLAIR; left), 3D gradient-echo (GRE) T 2 *-weighted (TE = 35 ms, top middle), 3D-EPI T 2 *-weighted (TE = 37 ms, top right), 3D GRE QSM (bottom middle), and 3D EPI QSM (bottom right) for multiple sclerosis Volunteer 14.None of the lesions are easily visible in the QSMs.

Figure S12 .
Figure S12.Swallow tail sign in 0.65-mm isotropic 3D EPI axial (top) and coronal (bottom) images with echo train length (ETL) 13 and varied acceleration factors.TableS1.Imaging parameters for the prototype 3D EPI and multi-echo 3D gradient echo (GRE) scans for comparing the effect of CAIPI shifts.Parameters not explicitly stated are identical to those used in 1-mm and 0.78-mm healthy participant scans stated in the manuscript.
financial support by the Austrian Federal Ministry for Digital and Economic Affairs and the National Foundation for Research, Technology and Development is also gratefully acknowledged.SB and AS acknowledge funding through an ARC Linkage grant (LP200301393).Open access publishing facilitated by The University of Queensland, as part of the Wiley -The University of Queensland agreement via the Council of Australian University Librarians.The acceleration pattern 3 × R kz is incompatible with this example even if k z blips are used.Figure S2.Residual segmentation phase errors (red arrows) in the 1-mm 3D EPI scans of Participant 3. Figure S3.Residual segmentation phase errors (red arrows) in the unaccelerated 0.65-mm 3D EPI scans of Participant 5. Despite their clear appearance of the artifacts in the T 2 *-weighted images, they do not significantly affect the appearance of susceptibility maps.Figure S4.T 2 *-weighted images for 0.78-mm acquisitions with different acceleration factors and CAIPI shifts.Red arrows indicate areas where detail is lost when no CAIPI shift is used.Figure S5.Susceptibility maps for 0.78-mm acquisitions with different acceleration factors and CAIPI shifts.
ky × R kz with either a ky or kz shift.CAIPI patterns that require a k z blip or a different ETL are not allowed and are grayed out.In these cases, when the CAIPI pattern is represented as a shift in k y , the acceleration factor in ky, R ky , is not an integer multiple of the shots per partition in the unaccelerated acquisition (red text).

Table S1 .
Imaging parameters for the prototype 3D EPI and multi-echo 3D gradient echo (GRE) scans for comparing the effect of CAIPI shifts.Parameters not explicitly stated are identical to those used in 1-mm and 0.78-mm healthy participant scans stated in the manuscript.How to cite this article: Tourell M, Jin J, Bachrata B, et al.Three-dimensional EPI with shot-selective CAIPIRIHANA for rapid high-resolution quantitative susceptibility mapping at 3 T. Magn Reson Med.2024;92:997-1010.doi: 10.1002/mrm.30101