Evaluation of simultaneous multi-slice readout-segmented diffusion-weighted MRI acquisition in human optic nerve measurements

Diffusion-weighted magnetic resonance imaging (dMRI) is the only available method to measure the tissue properties of white matter tracts in living human brains and has opened avenues for neuroscientific and clinical studies on human white matter. However, dMRI using conventional simultaneous multi-slice (SMS) single-shot echo planar imaging (ssEPI) still presents challenges in the analyses of some specific white matter tracts, such as the optic nerve, which are heavily affected by susceptibility-induced artifacts. In this study


Introduction
Advancements in diffusion-weighted magnetic resonance imaging (dMRI) and tractography algorithms have enabled identification of major white matter tracts in living human brains [1][2][3][4][5][6][7]. This method has opened up avenues to evaluate the tissue properties of white matter tracts in individuals and compare them with distributions in a healthy population [8]. This approach has been confirmed to be useful for understanding the effects of retinal disorders on white matter [9][10][11][12], the relationship between language-related white matter tracts and reading performance [13][14][15][16], the relationship between white matter tracts in the limbic system and psychiatric disorders [17], age dependency of white matter tissue properties [18][19][20], and lateralization of white matter tracts [21][22][23].
Although dMRI studies can provide insights into the field of cognitive and clinical neuroimaging, the existing methods show several limitations. Researchers usually acquire dMRI data using single-shot echo planar imaging (ssEPI). However, dMRI data acquired using ssEPI often shows image distortions and signal dropout due to susceptibilityinduced artifacts in the brain areas near the paranasal sinus or petrous apex/mastoid air complexes. This limitation poses challenges in identifying white matter pathways in areas affected by susceptibilityinduced artifacts, such as the optic nerve, which is located near the paranasal sinus. Assessment of the optic nerve is essential to evaluate the impact of disease by which it is directly damaged, such as optic neuritis [24] and glaucoma [12,25,26]. In addition, dMRI measurements of the optic nerve is important in understanding how amblyopia [27,28] affects white matter pathways; amblyopia commonly affects only one eye and its optic nerve has to be compared to that of the other eye. While obtaining dMRI measurements of the optic nerve using the conventional ssEPI method is not impossible [26,29,30], challenges in measurements limit the opportunity to study the optic nerve in relation to the health and diseases of the human visual system [31].
A few approaches have been proposed to improve dMRI measurements of the optic nerve. One approach is EPI acquisition with a reduced field of view to achieve higher image quality [32][33][34][35][36]. However, this approach limits the opportunity to evaluate the optic nerve along with other visual pathways, such as the optic tract and optic radiation, using the same dataset. An alternative approach is to use multi-shot EPI readout for dMRI acquisition to reduce echo spacing and achieve a higher resolution while reducing artifacts. The use of a multi-shot EPI approach for dMRI measurement of the human optic nerve has already been proposed [37], and various multi-shot methods for dMRI data acquisitions have been proposed to reduce the limitations of dMRI measurements using ssEPI [38][39][40][41][42][43][44][45]. Readout-segmented echo-planar imaging (rsEPI) is a multi-shot method that divides the acquisition space (k-space) into multiple segments along the readout direction (multi-shot acquisition) to enable the acquisition of dMRI data with a shorter echo time, reduced echo spacing, and reduced echo train length [46]. Several clinical studies have demonstrated that rsEPI is a useful approach for evaluation of the optic nerve and optic chiasm as it provides superior image quality with less distortion than ssEPI [47][48][49][50][51][52]. In addition, using rsEPI, a recent study identified the olfactory tract, a white matter tract affected by susceptibility-induced artifacts, from the dMRI data acquired from living humans [93]. However, a major limitation of multi-shot EPI methods, including rsEPI, is the prolonged acquisition time because of the need to collect data from multiple segments [53]. The prolonged acquisition time of rsEPI limits its application when whole-brain acquisition and tractography are required, as such studies require data acquisition from a large number of slices and diffusion gradient directions. Frost et al. (2015) proposed combining simultaneous multi-slice (SMS) acquisition [54][55][56] and rsEPI to overcome the prolonged acquisition time in rsEPI [57]. This method, named SMS rsEPI, could substantially reduce the dMRI acquisition time in comparison with rsEPI acquisition without SMS [57]. Thus, the development of SMS rsEPI methods has made dMRI measurements with rsEPI more practical for neuroscience and clinical studies. However, the brain areas in which dMRI acquisition with SMS rsEPI has an advantage over the commonly used SMS ssEPI method remain unclear.
This study aimed to evaluate the advantage of SMS rsEPI over SMS ssEPI in a tract-specific analysis of dMRI datasets on living human brains. To this end, we acquired dMRI data from 11 healthy volunteers by using both SMS rsEPI and SMS ssEPI. SMS ssEPI data were acquired multiple times to match the scan time with SMS rsEPI. We performed experiments with two different voxel sizes to evaluate the generality of findings. In our data analysis, we primarily focused on the optic nerve, which is commonly affected by susceptibility-induced artifacts, to clarify whether SMS rsEPI is an advantageous method for the measurement of this tract.

Participants
Eleven healthy participants (mean age: 23.8 years; age range: 19-34 years; four females) participated in this study. All participants had normal or corrected-to-normal vision and participated in two experiments (Experiment 1 and Experiment 2). The study protocol was approved by the local ethics and safety committees at Center for Information and Neural Networks (CiNet), Advanced ICT Research Institute, National Institute of Information and Communications Technology (NICT) and conducted in accordance with the ethical standards stated in the Declaration of Helsinki. Written informed consent was obtained from all participants.

MRI acquisition method
All MRI data were acquired using the SIEMENS 3 T scanner (MAG-NETOM Prisma, Siemens Healthcare, Erlangen, Germany) at CiNet, NICT, with the 32-channel head coil. The T1-weighted MRI data and all dMRI data were acquired during the same session.

T1-weighted MRI data acquisition
We obtained T1-weighted magnetization-prepared rapid gradientecho (MP-RAGE) images (voxel size: 1 mm isotropic; repetition time [TR]: 1990 ms; echo time [TE]: 3.37 ms; inversion time: 900 ms) from all participants. This image was used to identify regions of interest (ROIs) for tractography analysis and to estimate the border between the white and gray matter in subsequent analyses (see below). This image was also used as a reference image when evaluating distortions of dMRI images (see Fig. 2 and the other was acquired using the SMS rsEPI sequence, which is a work-in-progress sequence (research sequence) provided by SIEMENS (number of readout segments: 5; TR: 4390 ms; TE: 54 ms; ES: 0.34 ms; echo train length: 17 ms). Of note, the TR used for SMS ssEPI was the same in both experiments and above the minimum value. SMS rsEPI data was acquired with the reacquisition scheme, which was proposed in previous studies [46,57]. Both dMRI image sets were acquired with SMS excitation (SMS factor: 2) using blipped Controlled Aliasing In Parallel Imaging Results In Higher Acceleration (CAIPIRINHA) [58], in-plane acceleration (iPAT: 2) using generalized autocalibrating partially parallel acquisitions (GRAPPA) [59], and posterior-anterior phase-encoding direction with no slice gap. The slice orientation was transverse. Phase partial Fourier was not applied for the acquisition of either image set. Four non-diffusionweighted (b = 0) images (hereafter called "low b-value images") were acquired for each image set. During the acquisition of each image set, one low b-value image was acquired at the beginning and three at the end. In addition, we acquired two low b-value images with a reversed phase-encoding direction (anterior-posterior) for each image set. dMRI acquisition with SMS ssEPI was repeated five times to match the scan time with SMS rsEPI (SMS rsEPI: 13 min 11 s; SMS ssEPI: 12 min 55 s for five repetitions). For six participants, we acquired the SMS ssEPI image set earlier than the SMS rsEPI image set. For the remaining five participants, we acquired the SMS rsEPI image set earlier than the SMS ssEPI image set. The details of the acquisition parameters are listed in Table 1.

Experiment 2.
Experiment 2 aimed to evaluate the dMRI dataset acquired with relatively smaller voxel size. To this end, we acquired dMRI data by using spin-echo EPI with 1.8-mm isotropic voxels and 30 isotropically distributed diffusion weighting at b = 1000 s/mm 2 (FoV: 198 × 198 × 129.6 mm 3 ; number of slices: 72; diffusion scheme: monopolar). The data acquisition methods were largely the same as those used in Experiment 1, except for the following points. First, acquisition parameters differed slightly from those in Experiment 1 in both SMS ssEPI (TR: 4000 ms; TE: 70 ms; ES: 0.70 ms; echo train length: 38.5 ms) and SMS rsEPI (TR: 4520 ms; TE: 56 ms; ES: 0.34 ms; echo train length: 18.7 ms). Second, the acquisition time differed from that in Experiment 1 (SMS rsEPI: 13 min 34 s; SMS ssEPI: 13 min 5 s for five repetitions). The details of the acquisition parameters are listed in Table 1.
Both imaging and diffusion gradients were taken into account in the b-matrix calculations, which were used for subsequent analyses, in the SMS ssEPI and SMS rsEPI sequences.

Behavioral task.
During dMRI data acquisition, we instructed the participants to perform a fixation task to minimize eye movements. The participants viewed an MR-compatible screen monitor (BOLDscreen 32, Cambridge Research Systems, Kent, UK) located at the exit of the magnet bore via a mirror mounted over their eyes. A fixation point was presented at the center of the screen with a gray background. The participants were instructed to press a button when the color of the fixation dot (green or red) changed. The change in the fixation color occurred at a random timing once per 4 s on average. The participants were able to respond to changes in the fixation color within 1 s on most occasions (mean and standard deviation: 92.66% ± 10.19% for all participants), suggesting that eye movements are minimized during dMRI data acquisition.

dMRI data preprocessing
We performed susceptibility-induced distortion correction in the dMRI data by using the FSL TOPUP tool [62]. While we performed image comparisons of dMRI data before and after this correction (see Results), all subsequent analyses were performed based on the dMRI data with susceptibility-induced distortion correction. The dMRI data were corrected for eddy-current distortions and participant motion by using the FSL EDDY tool [63], and then aligned with T1-weighted MRI data by using a 14-parameter constrained nonlinear registration in the vistasoft distribution (mrDiffusion; https://github.com/vistalab/ vistasoft). B-matrices were reoriented after eddy-current correction and co-registration with T1-weighted images by applying the same transformation for the dMRI images [64]. We fit the diffusion tensor model to the dMRI data by using RESTORE (Robust Estimation of Tensors by Outlier Rejection; [65]). The resulting eigenvalues in each voxel were used to compute the FA, axial diffusivity (AD), and radial diffusivity (RD) [66] according to following formula: where λ 1 , λ 2 , and λ 3 are three eigenvalues obtained by fitting the diffusion tensor model. Among these eigenvalues, λ 1 is the principal eigenvalue of the diffusion tensor model. In addition, we also fitted constrained spherical deconvolution (CSD) for subsequent tractography analysis of major white matter tracts by using MRTrix3 (https://www.mrtrix.org/; [67]). For the dMRI data acquired with SMS ssEPI, fitting of tensor model and CSD was performed for the data combined across multiple repetitions.

Defining ROIs for tractography on visual pathways
We defined ROIs for tractography on the T1-weighted images acquired from each subject, which were defined using the methods already established in previous studies [8][9][10][11]26].

Optic nerve head.
We manually defined the position of the optic nerve head by visual inspection of a T1-weighted image in accordance with the method used in a previous study [26]. The ROIs for the optic nerve head were a 4-mm radius sphere centered slightly posterior to the optic nerve head at the back of the eye (Supplementary Fig. 1).

Optic chiasm.
The optic chiasm ROI was identified from the T1weighted image using an automated segmentation pipeline implemented in FreeSurfer (recon-all; [60]).

Lateral geniculate nucleus.
The lateral geniculate nucleus (LGN) was identified using methods described in previous publications [9][10][11]. In brief, we performed deterministic tractography by defining the optic chiasm as a seed to determine streamline termination near the LGN. We then defined the LGN ROIs as a 4-mm radius sphere that covered the endpoints of the streamlines from the optic chiasm. The anatomical validity of this method was evaluated in a previous study based on a comparison with histological data [11].

Primary visual cortex.
The ROI for the primary visual cortex (V1) was identified from the T1-weighted image using the Brodmann atlas implemented in the FreeSurfer pipeline [60,68].

Tractography
We performed tractography on the dMRI data to identify the white matter tracts in each participant. We used tractography algorithms, tractography parameters, and streamline exclusion criteria, that have already been established in previous studies to identify each white matter tract [8][9][10][11]26].

Optic nerve.
We identified the optic nerve using a method described in a previous study [26]. Specifically, we used CSD-based probabilistic tractography (iFOD2) implemented in MRTrix3 (fiber orientation distribution amplitude cutoff: 0.01; minimum streamline length: 20; maximum streamline length: 100) [69] and generated 5000 streamlines between a pair of ROIs (left/right optic nerve head and optic chiasm). We used this method because it is difficult to identify the optic nerve using tensor-based probabilistic tractography or whole-brain tractography owing to the relatively lower signal quality in this area. After generating streamlines between ROIs, we removed streamlines that met the following criteria: (1) streamline length ≥ 3 S.D. longer than the median streamline length in the tract and (2) streamline position ≥3 S.D. away from the median position of the tract [8]. This removal step was repeated three times. Finally, we rejected streamlines whose median distance to the tract core among all nodes was larger than 4.5 mm. Fig. 1 and Supplementary Fig. 2 show the optic nerve identified by tractography.

Optic tract.
We identified the optic tract using ConTrack [70] with the methods described in previous studies [9,11]. We sampled 5000 candidate streamlines connecting the optic chiasm and LGN (angle threshold: 90 • ; step size: 1 mm; maximum streamline length: 80 mm) in both hemispheres. We then selected the 100 streamlines with the highest scores in the ConTrack scoring process [71].

Optic radiation.
We identified the optic radiation using Con-Track [70], which is known to have sufficient sensitivity to identify the optic radiation, including Meyer's loop, from relatively low-resolution DTI data [71]. Using ConTrack, we sampled 100,000 candidate streamlines connecting the LGN and V1 (angle threshold: 90 • ; step size: 1 mm; maximum streamline length: 240 mm). We sampled a larger number of candidate streamlines for the optic radiation in comparison with those used for the optic nerve and optic tract, since the optic radiation has a larger volume and crosses other fibers in the Meyer's loop region [72]. Tracking was restricted using a white matter mask generated by tissue segmentation by FSL FAST. We refined the optic radiation streamlines on the basis of an outlier rejection process. First, we selected 30,000 streamlines with the highest scores in the ConTrack scoring process [71]. Next, we removed streamlines passing through voxels where the mean diffusivity exceeded 1 μm 2 /ms to minimize partial voluming with lateral ventricles. Third, using manually defined "NOT" ROIs, we excluded streamlines passing through the corpus callosum, anterior commissure, or brainstem or traversing the white matter inferior to the hippocampus as biologically implausible streamlines. Finally, we further removed streamlines based on the following criteria: (1) streamline length ≥ 3 S.D. longer than the median streamline length in the tract and (2) streamline position ≥3 S.D. away from the median position of the tract [8]. ConTrack has been used to successfully identify the optic radiation in previous publications [9][10][11][12]71,[73][74][75][76][77][78][79].

Other major white matter tracts.
We also identified other major white matter tracts to evaluate the extent to which the effect observed in the visual pathways can be generalized to other parts of the white matter. To this end, we generated whole-brain streamlines by using CSDbased probabilistic tractography implemented in MRTrix3 (iFOD2) [67] and taking the gray matter/white matter interface region, which was defined based on FSL FAST, as the seed voxels [80]. Two million streamlines were generated using the default parameters (step size: 0.5 x voxel size; angle, 45 deg.; minimum length of streamline: 2 x voxel size; maximum length of streamline: 100 x voxel size; fiber orientation distribution amplitude cutoff, 0.1). We identified the corticospinal tract (CST), forceps major, forceps minor, and inferior longitudinal fasciculus (ILF) from whole-brain streamlines by using the AFQ MATLAB toolbox (https://github.com/yeatmanlab/AFQ) [8]. After identifying each tract, outlier streamlines were further excluded using the following criteria: (1) streamline length > 4 S.D. longer than the median streamline length in the tract and (2) streamline position >4 S.D. away from the median position of the tract.

Evaluation of tissue properties along the tract
We evaluated the tissue properties of each visual white matter tract using methods used in previous studies [8,9,11,73,75,79]. Briefly, each streamline was resampled to 100 equidistant nodes. The tissue properties (FA, AD, and RD) were calculated for each node in each streamline. The properties at each node were summarized by taking the weighted average of the tissue properties (such as FA values) on each streamline within that node. The weight of each streamline was based on the Mahalanobis distance from the tract core. We excluded the first and last 10 nodes from the tissue property of the tract core to exclude voxels close to gray/white matter interfaces, where the tract was likely to intersect heavily with other tracts. We summarized the profile of each tract with a vector of 80 values representing FA, AD, and RD sampled at equidistant locations along the central portion of the tract. We calculated the average FA, AD, and RD values among 80 nodes for comparisons between the SMS rsEPI and SMS ssEPI datasets in each tract.

Statistical comparisons
For each tract, we performed a two-tailed two-sample t-test on FA to evaluate the statistical differences between the SMS rsEPI and SMS ssEPI datasets. We defined statistical significance (α) as P = 0.004, which is equivalent to Bonferroni correction for 12 comparisons (left/right optic nerve, left/right optic tract, left/right optic radiation, left/right CST, forceps major, forceps minor, and left/right ILF). We also calculated Cohen's d to estimate the effect size of the difference in FA between the SMS rsEPI and SMS ssEPI datasets. Statistical analyses were performed by using the MATLAB Statistics and Machine Learning Toolbox.
In addition, we performed a supplementary analysis to compare FA along the optic nerve between images acquired with different voxel sizes (Experiment 1 and 2) by performing a two-tailed two-sample t-test. We defined α as P = 0.05 in this supplementary analysis.  Fig. 2B show the images before applying susceptibility-induced distortion corrections. SMS ssEPI data showed susceptibility-induced image distortions near the optic nerve, which was located close to the paranasal sinus (the location of eyeballs in a T1weighted image is overlaid as yellow dotted lines in Fig. 2B). In contrast, SMS rsEPI data evidently showed smaller amounts of distortions at the optic nerve than the SMS ssEPI data in both Experiment 1 and 2 datasets (Fig. 2B). In addition to reduction of image distortions near the optic nerve, SMS rsEPI data also showed smaller amounts of image distortions and signal pile-up at the posterior part of the temporal lobe in these axial slices, which is near the petrous apex and mastoid air complexes (the location of the posterior end of the temporal pole in this slice is also shown as yellow dotted lines in Fig. 2B). Both SMS rsEPI and SMS ssEPI data did not show significant image distortions in the cerebellum (Fig. 2B).

Qualitative comparisons of image quality
The two right columns of Fig. 2 depict images after susceptibilityinduced distortion correction using the FSL TOPUP tool [62]. After correction, image distortions near the optic nerve in SMS ssEPI data were substantially reduced, and assessment of image quality differences solely by visual inspection was difficult (see Supplementary Fig. 3 for images acquired from other participants). In the subsequent analysis of diffusivity and its statistical comparisons, we focused on the data after the susceptibility-induced distortion.

Optic nerve
Using tractography, we successfully identified the optic nerve in both hemispheres in all participants ( Fig. 1 and Supplementary Fig. 2). We then compared FA along the optic nerve between the SMS rsEPI and SMS ssEPI datasets. Figure 3 depicts the profile of FA along the optic nerve in the SMS rsEPI (red) and SMS ssEPI datasets (blue) acquired in Experiment 1. In both hemispheres, the SMS rsEPI dataset exhibited a higher FA than the SMS ssEPI dataset. The differences were more prominent at the middle part of the optic nerve than at the anterior and posterior end. We then performed a statistical comparison between the datasets, by comparing the FA values averaged across all nodes along the optic nerve (Fig. 3,  right panel). In both hemispheres, SMS rsEPI data exhibited significantly higher FA than SMS ssEPI data along the optic nerve (left hemisphere; d = 0.73; paired t-test, t 10 = 6.42; 95% confidence interval [CI] = 0.03-0.06, P < 0.001; right hemisphere; d = 0.99; paired t-test, t 10 = 6.50; 95% CI = 0.03-0.06, P < 0.001). Figure 4 depicts the optic nerve results in Experiment 2, in which dMRI data were acquired with the smaller voxel size. Consistent with the results of Experiment 1, SMS rsEPI data showed a higher FA than SMS ssEPI data in the optic nerve (Fig. 4). Statistical comparisons suggest that SMS rsEPI exhibited significantly higher FA than SMS ssEPI data along the optic nerve (left hemisphere; d = 0.88; paired t-test, t 10 = 7.47; 95% CI = 0.04-0.07, P < 0.001; right hemisphere; d = 1.05; paired t-test, t 10 = 10.01; 95% CI = 0.04-0.06, P < 0.001). Taken together, these results indicate that SMS rsEPI data showed higher FA values along the optic nerve in the two datasets with different voxel sizes even after the susceptibility-induced distortion was corrected.
FA was computed from the AD and RD. We also examined these variables to better understand the source of the increased FA values along the optic nerve in the SMS rsEPI dataset. In both Experiments 1 and 2, the SMS rsEPI dataset showed larger AD and RD values than the SMS ssEPI datasets (Supplementary Fig. 4). The effect size of the difference was much larger in AD (Experiment 1, d = 2.22 and 2.33; Experiment 2, d = 2.77 and 1.89 for left and right hemisphere, respectively) than that in RD (Experiment 1, d = 1.44 and 1.51; Experiment 2, d = 1.82 and 1.23 for left and right hemisphere, respectively). Therefore, while SMS rsEPI data showed increased diffusivity in both the axial and radial directions, the increased sensitivity for AD explains the increased FA in the SMS rsEPI dataset.
In the main analyses, we analyzed tissue properties along the optic nerve in SMS rsEPI and SMS ssEPI by using tractography performed on the SMS rsEPI dataset to match the definition of the tract and voxels belonging to the optic nerve between the datasets. These procedures can potentially bias measurements in the SMS ssEPI dataset if there are any remaining image distortions even after preprocessing and the optic nerve voxels identified in SMS rsEPI data do not precisely match those in the SMS ssEPI dataset. To address this concern, we compared the FA values along the optic nerve when the tract was identified by tractography performed on each dMRI dataset (SMS rsEPI and SMS ssEPI). Even in this case, SMS rsEPI data showed higher FA values than SMS ssEPI data ( Supplementary Fig. 5; Experiment 1: d = 0.57 and 0.68; Experiment 2: d = 0.75 and 0.86 for left and right hemisphere, respectively).

Optic tract
We then analyzed the optic tract (Fig. 5A), which is a white matter tract from the optic chiasm to the LGN. While the optic tract also belongs to the early visual white matter pathways, it is in a more posterior and superior position and may be relatively less affected by susceptibilityinduced artifacts than the optic nerve. We evaluated the FA along the optic tract in the SMS rsEPI and SMS ssEPI datasets.

Optic radiation
We also analyzed the optic radiation (Fig. 6A), which is a white matter tract connecting the LGN and V1. The optic radiation is in an even more posterior and superior position than the optic tract. Therefore, this tract may be affected by susceptibility-induced artifacts to a lesser extent.

Comparisons of diffusivity measurements in other white matter tracts
We also analyzed six major white matter tracts ( Fig. 7A; left and right CST, forceps major, forceps minor, and left and right ILF) and compared FA between the SMS rsEPI and SMS ssEPI datasets. In Experiment 1, SMS rsEPI data showed higher FA values than SMS ssEPI data along the right ILF (d = 0.49; t 10 = 10.60; P < 0.001). However, we did not find significant differences between the datasets in the other five tracts (d = 0.01, 0.24, 0.06, − 0.01, and 0.08; t 10 = 0.31, 3.10, 2.10, − 0.06, and 1.36; P = 0.76, 0.01, 0.06, 0.95, and 0.20 for the left and right CST, forceps major, forceps minor, and left ILF, respectively). In Experiment 2, SMS rsEPI data showed significantly higher FA values than SMS ssEPI data in the right CST and ILF (d = 0.37 and 0.33; t 10 = 4.69 and 9.82, respectively; P < 0.001 in both tracts). In contrast, we did not find any significant differences between the datasets in the other four tracts (d = − 0.04, − 0.09, 0.11, and 0.02; t 10 = − 0.33, − 1.26, 0.59, and 0.17; P = 0.75, 0.24, 0.57, and 0.87 for the left CST, forceps major, forceps minor, and left ILF, respectively).

Discussion
In this study, we evaluated the advantages of SMS rsEPI over SMS ssEPI in dMRI measurements of white matter tracts. We particularly focused on the optic nerve because optic nerve assessments are important to evaluate the consequences of visual disorders. However, they are often affected by susceptibility-induced artifacts in conventional SMS ssEPI acquisition [31]. We visually confirmed that the SMS rsEPI data showed smaller amounts of susceptibility-induced distortions near the optic nerve than SMS ssEPI data, indicating the advantage of SMS rsEPI in measuring this tract. Even after distortion correction, the SMS rsEPI dataset showed higher FA values along the optic nerve than the SMS ssEPI dataset. The increase in FA values along the optic nerve was consistently observed in two datasets acquired with different voxel sizes and also observed in the optic tract. In contrast, the increase of FA values was insignificant in some other major white matter tracts, such as the optic radiation, suggesting that the effect observed in early visual pathways is not generalizable to the entire white matter. In summary, our results demonstrate that SMS rsEPI is a promising approach for measuring the tissue properties of white matter tracts affected by susceptibility-induced artifacts, such as the optic nerve.

SMS rsEPI shows smaller distortion and higher FA in dMRI measurements of the optic nerve
This study aimed to investigate the advantages of the SMS rsEPI method in comparison with SMS ssEPI for dMRI measurements on the human optic nerve. Our results suggested that SMS rsEPI data showed smaller amounts of image distortions (Fig. 2) and higher FA values (Fig. 3) along the optic nerve, supporting the idea that SMS rsEPI will be an advantageous method for dMRI measurements on this pathway. One interpretive limitation of this result is that a higher FA does not always indicate an improvement in the dMRI measurements. Moreover, FA depends on many different factors (partial voluming, crossing fibers, and microstructural properties such as myelination), and the relationship between signal-to-noise ratio and FA can be complex, as demonstrated by a numerical simulation study in which the measurements with lower signal-to-noise ratio induced overestimation or underestimation of FA [81]. Due to a lack of ground truth in the in vivo dMRI measurements performed on human brains, the interpretation of increased FA along the optic nerve remains speculative. However, since the optic nerve does not cross other tracts and most optic nerve fibers are myelinated [82], the optic nerve should show a higher FA value if the measurement does not suffer from lower signal quality and artifacts. Therefore, a lower FA value along the optic nerve observed in the SMS ssEPI dataset may reflect the limited sensitivity of dMRI measurements in identifying diffusion anisotropy that must exist in the optic nerve. However, the same interpretation for FA results will not be applicable to other tracts such as the ILF, which include regions that cross other fiber tracts [83].
Several possible reasons can be used to explain the source of FA differences in the optic nerve between the SMS rsEPI and SMS ssEPI datasets. The first possible reason is the influence of the longer echo train in SMS ssEPI than that in SMS rsEPI. It seems likely that T 2 * decay during the longer echo train affected the point spread function and consequently caused blurring and partial volume effects in the SMS ssEPI dataset. The second possible reason is that an increased level of distortion in the SMS ssEPI images results in blurring after distortion correction during preprocessing, and this blurring effect can be greater than when the same procedure is applied to SMS rsEPI. This blurring effect may result in reduced FA in the SMS ssEPI dataset. Lastly, it is also possible that differences in general image quality and noise distribution between the datasets may result in FA differences. It is known that noise of MRI data with a lower signal-to-noise ratio follows non-Gaussian distribution [84]. Such noise distribution may affect the estimation of diffusivity measurements [85]. If this effect is more substantial in SMS ssEPI images rather than SMS rsEPI images, such difference can also cause FA differences, as observed in this study. Of note, images acquired via SMS rsEPI and SMS ssEPI were reconstructed through different processes, as SMS rsEPI data from multiple shots are combined in kspace before conversion to modulus images. This difference may impact the noise distributions.
The impact of some of the aforementioned factors might significantly depend on the acquisition parameters selected, such as the b-value and voxel size, since these can affect the achievable echo spacing in the EPI echo train. In this study, we performed two experiments with different voxel sizes (2 mm and 1.8 mm isotropic) and found that dependencies on the voxel size may be small or absent in the optic nerve (Figs. 3 and 4). However, we may observe dependencies on the voxel size if we test different combinations of voxel size and other parameters, such as the bvalue. Since it was difficult to test all combinations of the acquisition parameters in a single study, dependencies on the acquisition parameters should be further assessed in future investigations.

Dependency on white matter tracts
Our results also suggest that the potential advantage of SMS rsEPI depends on the white matter tracts. For example, the optic radiation did not show statistically significant differences between the SMS rsEPI and SMS ssEPI datasets in either experiment (Fig. 6). This is most likely because the optic radiation is not located near the paranasal sinus, and dMRI measurements of this tract are less affected by susceptibilityinduced distortion. In fact, previous studies have demonstrated that it is quite possible to identify the optic radiation and its tissue changes caused by retinal disorders by using the ssEPI approach [9,11]. However, we did not exclude the possibility that the advantage of SMS rsEPI is smaller in the optic radiation, since FA measurements do not strongly reflect the improvement of signal quality because the structural properties of the tract (tract size and crossing fibers) differ from those of the optic nerve [72].
We also found that some association pathways, such as the right ILF, showed higher FA values in the SMS rsEPI dataset (Fig. 7). This result can be explained by the fact that the ILF is located close to the temporal Comparisons of FA along other major tracts between SMS rsEPI and SMS ssEPI datasets. A. The left corticospinal tract (CST, orange), forceps major (purple), forceps minor (magenta), and left inferior longitudinal fasciculus (ILF, yellow) identified by tractography on SMS rsEPI data acquired from a representative participant (participant P1) overlaid on a sagittal slice of a T1weighted image, which is located medial to the CST and ILF. B. Effect size (d, vertical axis) on the FA difference between SMS rsEPI and SMS ssEPI datasets in white matter tracts (blue, Exp. 1; red, Exp. 2). A positive value indicates that FA was higher in the SMS rsEPI dataset than in the SMS ssEPI dataset. Asterisk indicates statistical significance (P < 0.004). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) pole, which is an area often affected by susceptibility-induced artifacts due to its proximity to the petrous apex and mastoid air complexes.

Limitations and future directions
One notable limitation of this study is that we only tested one hardware (SIEMENS MAGNETOM 3 T Prisma) for comparisons between SMS rsEPI and SMS ssEPI. While this magnet has a high-performance gradient system (maximum gradient amplitude: 80 mT/m, slew rate: 200 mT/m/s), the disadvantages of SMS ssEPI can be reduced if MR systems with ultra-high-performance gradient coils become widely available, because such systems can substantially reduce echo spacing and TE on dMRI measurements using SMS ssEPI [86,87]. In contrast, the advantage of SMS rsEPI may become much larger when using an MR system with a higher static magnetic field (7 T or beyond), in which the limitations imposed by T 2 * decay became more challenging [53]. For these reasons, the advantages of SMS rsEPI over SMS ssEPI, as shown in this study, are not fully generalizable to other hardwares with different gradient performance levels and static magnetic field strengths.
SMS acquisition is very effective when a large number of slices are necessary to cover the whole brain. It is particularly useful when researchers aim to evaluate the impact of disease on the optic nerve, together with other white matter tracts, such as the optic tract, optic radiation, and other association pathways. However, if smaller coverage is adequate, rsEPI without SMS acquisition may be more advantageous for measurement of the optic nerve as this may avoid the signal-to-noise ratio reduction associated with SMS. This point needs to be clarified by future studies in which rsEPI with and that without SMS are compared.
While this study focused entirely on healthy young adult participants, future investigations should be performed on clinical populations with various ages to validate the clinical benefits of SMS rsEPI. Since the results of this study demonstrated that SMS rsEPI is advantageous for optic nerve measurements, one possible extension of this study is to use SMS rsEPI for acquiring dMRI data from patients with disorders directly affecting the optic nerve, such as glaucoma [12,25,26,29,75] and optic neuritis [24,32,34,36,50,51]. Accurate measurements of the optic nerve will also facilitate evaluations of amblyopia [27,28], in which the degree of decreased vision can differ between the two eyes, necessitating measurement of the tissue properties of the optic nerve in individual hemispheres. The generalizability of our results to participants from different age groups needs to be evaluated, considering the age dependency of dMRI measurements on white matter tracts [18][19][20]23]. As the prevalence rate of certain visual disorders, such as glaucoma, is higher in older age groups [88], such evaluation is essential to clarify the benefit of SMS rsEPI for clinical practice. Future studies should aim to determine the extent to which SMS rsEPI has improved sensitivity in identifying tissue changes along the optic nerve occurring as a consequence of these visual disorders.
In our study, SMS rsEPI exhibited a smaller amount of image distortion than SMS ssEPI (Fig. 2), which indicates that it holds promise for future application in diagnostic imaging. For further advancement of the clinical application of dMRI, tissue properties of the optic nerve of one individual have to be quantitatively evaluated in relation to the distribution of properties in a healthy population. A tract-specific approach [8,89,90], as performed in this study, is an important step to enable quantitative, statistical evaluation of a specific tract of interest in individual brains. However, further studies on its reliability and generalizability are required to validate an SMS rsEPI-based tractography approach for clinical investigations [91,92]. An extension of this work should provide further insights into how SMS rsEPI can be best used for clinical evaluation of the human optic nerve.

Conclusions
This study demonstrated that dMRI acquisition with SMS rsEPI offered advantages in measurement of the optic nerve in comparison with SMS ssEPI in terms of reduced susceptibility-induced artifacts and increased diffusion anisotropy along the tract. This result will open up avenues to perform health-and disease-related investigations on optic nerve tissue properties with a reasonable acquisition time.

Ethics statement
The study protocol was approved by the local ethics and safety committees at Center for Information and Neural Networks (CiNet), Advanced ICT Research Institute, National Institute of Information and Communications Technology (NICT) and conducted in accordance with the ethical standards stated in the Declaration of Helsinki. Written informed consent was obtained from all study participants.

Funding
This study was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI (JP17H04684 and JP21H03789 to H.T.). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of Competing Interest
WL and HK are employed by Siemens Healthcare. Siemens Healthcare did not have any additional role in the study design, data collection and analysis. Other three authors have declared that no competing interests exist.