Background & Summary

Multimodal microstructural MRI has shown increased sensitivity and specificity to microstructural changes in various brain disease and injury models in the preclinical setting. Here, we present an in vivo longitudinal imaging dataset in the healthy mouse brain, which includes structural T2-weighted, magnetization transfer (MT), and advanced diffusion MRI (dMRI) data. There were no hardware or software changes during data acquisition, and all protocols for a single timepoint in each mouse were acquired in the same session. Each of 12 C57Bl/6 mice were scanned at 6 different timepoints, between 3–8 months of age (Fig. 1). Importantly, this dataset provides imaging data in the same mice over time, which provides greater statistical power compared to cross-sectional studies, to detect changes in brain maturation, as myelination continues to increase between three and six months1. The data were acquired with the goals of forming a control dataset and investigating microstructural changes in the healthy mouse brain.

Fig. 1
figure 1

Overview of study design. At Day 0, all mice were 12–14 weeks old. Each C57Bl/6 mouse (n = 12, six males and six females) was scanned at 6 different timepoints, comprising a total of 72 MRI sessions. After Week 20, four of the mice (two males and two females) underwent ex vivo imaging.

Full size image

To optimize potential applications of this dataset, we provide dMRI pulse sequences and protocols, source data (DICOM format), code to process source data, unprocessed and preprocessed data (NIFTI format), and quantitative MRI metric maps. We have made this dataset publicly available as other groups may not have access to all resources required to undertake a longitudinal MRI study. This includes hardware, software (specifically custom pulse sequences to implement advanced dMRI protocols), and time/personnel required. Analysis of test-retest reproducibility of the MRI metrics, using a subset of the data, have been published elsewhere2,3. We envision utility of this dataset in the microstructural MRI field to develop and test methods that model temporal brain dynamics, registration and preprocessing pipelines, and biophysical models of brain microstructure. Sex and age-dependent differences can be investigated, as the dataset includes an equal number of male and female age-matched mice. In vivo to ex vivo changes, arising from perfusion and fixation processes, can be explored, as a subset of ex vivo data has been included.

MT imaging has been used extensively to investigate changes in myelin content and integrity4,5. The MT imaging protocol applied here enables computation of the widely used MT ratio (MTR), and the more recently developed MT saturation index (MTsat)6. As MTR is confounded by T1 effects, flip angle inhomogeneities, and choice of sequence parameters, MTsat was developed to reduce T1 dependence and improve specificity to myelin, while maintaining a feasible scan time. MTsat shows higher white matter contrast in the brain than MTR3,6, and has been shown to correlate more with disability metrics than MTR in patients with multiple sclerosis7.

Developing advanced dMRI techniques, beyond the conventional diffusion tensor imaging (DTI) model, is currently of broad interest, as DTI lacks the specificity to identify unique microstructural environments8. The advanced dMRI methods applied here include oscillating gradient spin echo (OGSE) dMRI9,10, implemented by varying the oscillating gradient frequency, and microscopic anisotropy (μA) dMRI11,12,13,14, implemented via tensor valued diffusion encoding. In addition to advanced dMRI metrics, traditional DTI metrics are also provided. OGSE dMRI provides additional insight, compared to conventional dMRI, by increasing sensitivity to smaller spatial scales. This is a robust dataset to explore the frequency dependence of OGSE dMRI metrics, which may provide insight into the relevant mesoscopic structures affecting water diffusion15. Evidence of a linear dependence of mean diffusivity on the square root of OGSE frequency has been demonstrated in healthy and globally ischaemic rodent brain tissue16 and in healthy human white matter17. In contrast to the widely used fractional anisotropy metric (FA)8, which confounds true microstructural changes with fiber orientation dispersion, the microscopic anisotropy (μA) metric quantifies water diffusion anisotropy independent of orientation dispersion11,18,19. Importantly, μA dMRI can provide estimates of cell shape11,18,20,21,22,23,24,25. Additionally, diffusional kurtosis estimated from the μA protocol includes linear kurtosis (arising from the linear tensor encoding (LTE) acquisitions) and isotropic kurtosis (arising from the spherical tensor encoding (STE) acquisitions), which can be related to cell size heterogeneity20.

As myelin is MR-invisible in diffusion-weighted scans, recent studies have applied both dMRI and MT methods for a more comprehensive view of microstructural changes26,27,28. Thus, there may be interest in investigating longitudinal changes by jointly assessing MT and dMRI data, and additionally testing biophysical models using the combined OGSE, µA, and MT data.

Methods

Subjects

All animal procedures were approved by the University of Western Ontario Animal Care Committee and were consistent with guidelines established by the Canadian Council on Animal Care. Twelve adult C57Bl/6 mice (six male and six female) were scanned at six timepoints. They were between 12–14 weeks old at the first timepoint (Fig. 1). Before scanning, anesthesia was induced by placing the animals in an induction chamber with 4% isoflurane and an oxygen flow rate of 1.5 L/min. Following induction, isoflurane was maintained during the imaging session at 1.8% with an oxygen flow rate of 1.5 L/min through a custom-built nose cone. At the end of the study, the mice were euthanized. The mice were anesthetized with ketamine/xylazine (2:1) and then underwent trans-cardiac perfusion with ice-cold saline, followed by 4% paraformaldehyde in phosphate-buffer saline (PBS).

In vivo MRI Acquisition

In vivo MRI experiments were performed on a 9.4 Tesla (T) Bruker small animal scanner, running ParaVision 6.0.1, equipped with a gradient coil set of 1 T/m strength (slew rate = 4100 T/m/s). A single channel transceive surface coil (20 mm × 25 mm), built in-house, was fixed in place directly above the mouse head to maximize signal-to-noise ratio (SNR). The mouse holder (which included ear bars and a bite bar), nose cone, and surface coil were fixed onto a support, which was placed into the scanner (Fig. 2a). This ensured consistent positioning of the mouse head in the scanner at each session. 30 slices, with a slice thickness of 400 µm (anatomical scans) or 500 µm (diffusion-weighted scans), were required for full brain acquisition for all protocols. Anatomical images were acquired at each session for each subject using a T2-weighted TurboRARE sequence. A brief overview of the protocols is given in Table 1.

Fig. 2
figure 2

Schematic of experimental setup for in vivo and ex vivo imaging sessions. (a) In vivo setup showing the 3D printed mouse holder and surface coil securely attached to a support. The cross-section of the mouse holder depicts how the mouse is secured in place with a nose cone, bite bar, and ear bars. (b) Ex vivo setup showing the 3D printed mouse brain holder, which can hold two extracted brains, and the 3D printed plastic container, which holds the mouse holder and is filled with Christo-lube. Both the mouse brain holder and container were custom designed to fit in the MP30 volume coil. The MP30 volume coil is securely attached to the support, with the isocentre marked in red.

Full size image
Table 1 Brief sequence details for in vivo imaging. For a full list of parameters, the exported protocols are included in the repository.
Full size table

OGSE and µA dMRI

Each dMRI protocol was acquired with single-shot spin echo echo-planar-imaging (EPI) readout with partial Fourier imaging in the phase encode direction with 80% of k-space being sampled. For each dMRI protocol, a single reverse phase encoded b = 0 s/mm2 volume was acquired at the end of the diffusion sequence for subsequent use in TOPUP29 and EDDY30, from FMRIB Software Library (FSL, Oxford, UK)31, to correct for susceptibility and eddy current induced distortions.

The OGSE dMRI protocol included a PGSE sequence (with gradient duration = 11 ms and diffusion time = 13.8 ms) and four OGSE sequences with oscillating gradient frequencies of 50 Hz, 100 Hz, 145 Hz, and 190 Hz at b = 800 s/mm2 (10 directions for each frequency). The lowest OGSE frequency (50 Hz) uses the newly introduced frequency tuned bipolar (FTB) waveforms to reduce TE of the acquisition32. The μA sequence was implemented with linear (LTE) and spherical tensor (STE) encodings, as shown in Table 1, at b = 2000 s/mm2 (30 directions for each of LTE and STE) and b = 1000 s/mm2 (12 directions). Details about gradient waveforms and gradient modulation power spectra for the OGSE and µA protocols implemented here are presented in Rahman et al.2.

MT Imaging

The MT protocol required 50 minutes total scan time and comprised three FLASH-3D (fast low angle shot) scans and one RF transmit field (B1) map scan acquired using the actual flip-angle imaging (AFI) method33 to correct for local variations in flip angle. An MT-weighted scan, and reference T1-weighted and PD-weighted scans (MTw, T1w, and PDw respectively) were acquired by appropriate choice of the repetition time (TR) and the flip angle (α). MT-weighting was achieved by applying an off-resonance Gaussian-shaped RF pulse (12 ms duration, 385° nominal flip angle, 3.5 kHz frequency offset from water resonance, 5 µT RF peak amplitude) prior to the excitation.

Ex vivo MRI Acquisition

Ex vivo MRI experiments were performed on a subset of four mice (two male and two female) after the last in vivo scan. The mouse IDs of ex vivo data are: NR1_F (female), NR2_F (female), NR7_M (male), and NR8_M (male). NR1_F and NR2_F were scanned with the skull attached to the brain to minimize chances of tissue deformation, while NR7_M and NR8_M were scanned with the skull removed.

Ex vivo imaging was also performed on the 9.4 Tesla (T) Bruker small animal scanner, running ParaVision 6.0.1, equipped with a gradient coil set of 1 T/m strength (slew rate = 4100 T/m/s). A 3D printed mouse brain holder, holding two mouse brains at a time, was placed into a 3D printed plastic container and submerged with lubricant (Christo-lube MCG 1009; Engineered Custom Lubricants) to avoid magnetic susceptibility-related distortion artifacts (Fig. 2b). The mouse brain holder and container were custom designed to fit in the MP30 volume coil (Agilent, Palo Alto, CA, USA). The container was then slid into the volume coil (fixed on a support) and taped onto the support. The design of the mouse brain holder and container ensured that the mouse brain was positioned at the isocentre of the volume coil and the design of the support ensured consistent positioning of the mouse brain in the scanner at each session. 30 slices, with a slice thickness of 400 µm (anatomical scans) or 500 µm (diffusion-weighted scans), were required for full brain acquisition for all protocols. Anatomical images were acquired for each brain using a T2-weighted TurboRARE sequence. Due to field-of-view (FOV) constraints, one brain was imaged at a single session (although the mouse holder was designed to hold two brains). A brief overview of the protocols is given in Table 2. The total ex vivo scan time for each brain was 15 hours.

Table 2 Brief sequence details for ex vivo imaging. For a full list of parameters, the exported protocols are included in the repository.
Full size table

OGSE and µA dMRI

Each dMRI protocol was acquired with multi-shot spin echo echo-planar-imaging (EPI) readout with 2 shots and partial Fourier imaging in the phase encode direction with 75% of k-space being sampled. Reverse phase-encoded volumes were not acquired for ex vivo data.

The OGSE dMRI protocol included a PGSE sequence (with gradient duration = 11 ms and diffusion time = 13.8 ms) and four OGSE sequences with oscillating gradient frequencies of 50 Hz, 80 Hz, 115 Hz, and 150 Hz at b = 1600 s/mm2 (10 directions for each frequency), with the lowest OGSE frequency using the FTB waveform. The μA sequence was implemented with linear (LTE) and spherical tensor (STE) encodings, as shown in Table 2, at b = 1320 s/mm2 (6 directions for each of LTE and STE), b = 2640 s/mm2 (9 directions), and b = 4000 s/mm2 (15 directions).

MT Imaging

MT-weighting was achieved by applying an off-resonance Gaussian-shaped RF pulse, with the same parameters for in vivo imaging, prior to the excitation.

Data analysis pipeline

The data analysis pipeline was built using Snakemake34 (described in greater detail in the Usage Notes section). A Snakemake workflow defines data analysis in terms of rules that are specified in the “Snakefile.” Fig. 3 outlines the data analysis steps from DICOM to scalar map generation.

Fig. 3
figure 3

Flowchart outlining data analysis steps from DICOM to scalar map generation. The “rules” listed are those used in the Snakemake pipeline, using Snakemake 3.13.3. The software dependencies and versions of each analysis step are included.

Full size image

OGSE and µA dMRI recon and preprocessing

For the dMRI protocols, averages were acquired separately on the scanner and the complex-valued averages were combined using in-house MATLAB code which included reconstruction of partial Fourier data using POCS (Projection onto Convex Sets)35, correction for frequency and signal drift associated with gradient coil heating36, and Marchenko-Pastur denoising of complex-valued data37. If averages were not collected separately, this step can simply be skipped. Importantly, the pipeline can be used for both complex-valued and magnitude data. After the averages were combined, images were preprocessed using Gibbs ringing correction from the MRtrix3 package38, followed by TOPUP29 and EDDY30 from FMRIB Software Library (FSL, Oxford, UK)31. Using the data collected with reverse phase-encode blips, the susceptibility-induced off-resonance field was estimated using TOPUP. Then, EDDY was run to correct for eddy current induced distortions (volume-by-volume), perform motion correction, and apply the results from TOPUP.

Brain masks

For each protocol, brain masks were produced using the skull stripping tool from BrainSuite (v. 19b)39 and manually edited, as needed. For dMRI data, an initial brain mask (after dMRI averages were combined) was created by registering the T2 brain masks to a b = 0 s/mm2 volume, using ANTs software40. This initial brain mask was required for EDDY in the dMRI preprocessing step. After dMRI preprocessing, a final brain mask was produced and manually edited using BrainSuite. Brain masks for images from the T2, MT, and dMRI protocols have been included in the repository.

Scalar map generation

Scalar maps are shown in Fig. 4 (in vivo) and Fig. 5 (ex vivo). The scalar maps provided in the repository are summarized in Supplementary Table 1. Although briefly described here, more details describing the scalar maps are presented in previous reproducibility studies of the data2,3.

Fig. 4
figure 4

In vivo scalar maps. Other DTI metric maps (such as axial and radial diffusivity) are not shown here but have been included in the repository. MTR: magnetization transfer ratio; MTsat: magnetization transfer saturation; FA: fractional anisotropy; MD: mean diffusivity; ΔMD: mean diffusivity difference between MD (190 Hz) and MD (0 Hz); Λ: diffusion dispersion rate; µA: microscopic anisotropy; µFA: microscopic fractional anisotropy; KLTE: linear kurtosis calculated from LTE volumes; KSTE: isotropic kurtosis calculated from STE volumes.

Full size image
Fig. 5
figure 5

Ex vivo scalar maps. Other DTI metric maps (such as axial and radial diffusivity) are not shown here but have been included in the repository. MTR: magnetization transfer ratio; MTsat: magnetization transfer saturation; FA: fractional anisotropy; MD: mean diffusivity; ΔMD: mean diffusivity difference between MD (190 Hz) and MD (0 Hz); Λ: diffusion dispersion rate; µA: microscopic anisotropy; µFA: microscopic fractional anisotropy; KLTE: linear kurtosis calculated from LTE volumes; KSTE: isotropic kurtosis calculated from STE volumes.

Full size image

Scalar maps of MTR and MTsat were generated from the MT protocol. MTw, PDw, and T1w images were used to calculate MTsat maps, following the original method proposed by Helms et al.6 and outlined by Hagiwara et al.41 and Rahman et al.3. Furthermore, B1 maps are available to correct for small residual higher-order dependencies of the MT saturation on the local RF transmit field to further improve spatial uniformity, as suggested by Weiskopf et al.42.

From the OGSE data, maps of MD at each frequency were generated using MRtrix338. The mean diffusivity difference, ΔMD, was calculated as the difference between MD acquired at the highest frequency (190 Hz (in vivo) or 150 Hz (ex vivo)) and MD acquired at the lowest frequency (0 Hz). To characterize the power law relationship between MD and OGSE frequency (f)15, the slope of linear regression of MD with f,0.5 the diffusion dispersion rate (Λ), was calculated. From the µA data, maps of microscopic anisotropy (µA), microscopic fractional anisotropy (µFA), and diffusion kurtosis arising from LTE and STE acquisitions (KLTE and KSTE respectively) were generated by fitting the powder-averaged STE and LTE signals versus b-value to the diffusion kurtosis model14. The powder-averaged signal, in diffusion MRI, refers to the average signal intensity over all directions in a specific b-shell11. As a reference for the OGSE and µA metrics, DTI metrics have been included in the repository.

Data Records

The datasets, exported MRI protocols, Snakemake pipeline, and in-house MATLAB code are available in the Federated Research Data Repository (FRDR) at https://doi.org/10.20383/103.059443.

Datasets are arranged in ‘Data’ as mouseID_sex/timepoint/MRI_contrast. The timepoint includes in vivo timepoints (Day0, Day3, Week1, Week4, Week8, and Week20) and the ex vivo timepoint (ex_vivo). For dMRI data, preprocessed data and scalar maps are arranged in ‘DiffusionDataPreproc’ with the same structure of mouseID_sex/timepoint/MRI_contrast.

Structural T2-weighted dataset

Folder

Filename

Filename Extensions

Data

T2-weighted scan:

mouseID_sex/timepoint/T2_TurboRARE_AX150150500_A16/ T2_TurboRARE_AX150150500_A16 (in vivo)

mouseID_sex/timepoint/ T2_TurboRARE_AX100100500_A48/ T2_TurboRARE_AX100100500_A48 (ex vivo)

.dcm

.nii.gz

.json

_method.json

Data

Brain mask:

mouseID_sex/timepoint/T2_TurboRARE_AX150150500_A16/ T2_TurboRARE_AX150150500_A16_mask (in vivo)

mouseID_sex/timepoint/ T2_TurboRARE_AX100100500_A48/ T2_TurboRARE_AX100100500_A48_mask (ex vivo)

.nii.gz

Data

T2-weighted scan with brain mask applied:

mouseID_sex/timepoint/T2_TurboRARE_AX150150500_A16/ T2_TurboRARE_AX150150500_A16_Wmask (in vivo)

mouseID_sex/timepoint/ T2_TurboRARE_AX100100500_A48/ T2_TurboRARE_AX100100500_A48_Wmask (ex vivo)

.nii.gz

MT Imaging dataset

Folder

Filename

Filename Extensions

Data

MTw scan:

mouseID_sex/timepoint/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz/ MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz (in vivo)

mouseID_sex/timepoint/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz (ex vivo)

.dcm

.nii.gz

.json

_method.json

_visu_pars.json

Data

PDw scan:

mouseID_sex/timepoint/MToff_PD_GRE_3D_150 × 400_12A/MToff_PD_GRE_3D_150 × 400_12A (in vivo)

mouseID_sex/timepoint/ MToff_PD_GRE_3D_100 × 400_36A/ MToff_PD_GRE_3D_100 × 400_36A (ex vivo)

.dcm

.nii.gz

.json

_method.json

_visu_pars.json

Data

T1w scan:

mouseID_sex/timepoint/ MToff_T1_GRE_3D_150 × 400_12A/MToff_T1_GRE_3D_150 × 400_12A (in vivo)

mouseID_sex/timepoint/ MToff_T1_GRE_3D_100 × 400_36A/ MToff_T1_GRE_3D_100 × 400_36A (ex vivo)

.dcm

.nii.gz

.json

_method.json

_visu_pars.json

Data

B1 map data (2 volumes acquired from 2 TRs):

mouseID_sex/timepoint/rpAFI_mouse_2/rpAFI_mouse_2 (in vivo and ex vivo)

.dcm

.nii.gz

.json

_method.json

_visu_pars.json

Data

B1 map (from the scanner):

mouseID_sex/timepoint/rpAFI_mouse_1/rpAFI_mouse_1 (in vivo and ex vivo)

.dcm

.nii.gz

.json

_method.json

_visu_pars.json

Data

B1 map (resampled to match MTw scan’s resolution):

mouseID_sex/timepoint/rpAFI_mouse_1/rpAFI_mouse_1_vol2_RS (in vivo and ex vivo)

.nii.gz

Data

Text file detailing which B1 map slices have artifacts (0 for slices with the banding artifact and 1 for slices without artifacts):

mouseID_sex/timepoint/rpAFI_mouse_1/rpAFI_mouse_1 (in vivo and ex vivo)

.csv

Data

Brain mask:

mouseID_sex/timepoint/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz_mask (in vivo)

mouseID_sex/timepoint/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz_mask (ex vivo)

.nii.gz

Data

MTR – scalar map:

mouseID_sex/timepoint/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz_mtr (in vivo)

mouseID_sex/timepoint/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz_mtr (ex vivo)

.nii.gz

Data

MTsat – scalar map:

mouseID_sex/timepoint/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz_mtsat (in vivo)

mouseID_sex/timepoint/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz/ MTon_GRE_3D_100 × 400_36A_5uT_385FA_3500Hz_mtsat (ex vivo)

.nii.gz

OGSE dMRI dataset

Folder

Filename

Filename Extensions

Data

OGSE dMRI scan (complex-valued data):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10 (in vivo)

mouseID_sex/timepoint/OGSE_res130150500/OGSE_res130150500 (ex vivo)

.dcm

_real.nii.gz

_imaginary.nii.gz

_real.json

_imaginary.json

_method.json

_visu_pars.json

.bmat

.bvec

.bval

Data

OGSE dMRI scan (after averages are combined):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb (in vivo)

mouseID_sex/timepoint/OGSE_res130150500/OGSE_res130150500_aveComb (ex vivo)

.nii.gz

.bmat

.bvec

.bval

Data

b0 scan acquired with reverse PE (complex-valued data):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10_b0_reversePE/OGSE_5Shapes_1A_5Rep_TR10_b0_reversePE_aveComb (in vivo)

(not acquired for ex vivo)

.dcm

_real.nii.gz

_imaginary.nii.gz

_real.json

_imaginary.json

_method.json

_visu_pars.json

.bmat

.bvec

.bval

Data

b0 scan acquired with reverse PE (after averages are combined):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10_b0_reversePE/OGSE_5Shapes_1A_5Rep_TR10_b0_reversePE_aveComb (in vivo)

.nii.gz

.bmat

.bvec

.bval

Data

Mean b0 volume extracted from dataset after averages are combined (with normal PE):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_mean_b0 (in vivo)

.nii.gz

DiffusionDataPreproc

Preprocessed Dataset:

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc (in vivo)

mouseID_sex/timepoint/OGSE_res130150500/OGSE_res130150500_aveComb_preproc (ex vivo)

.nii.gz

.bvec

.bval

DiffusionDataPreproc

Preprocessed Dataset split into separate frequencies (in vivo):

PGSE or 0 Hz:

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000

50 Hz OGSE:

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f050

100 Hz OGSE:

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f100

145 Hz OGSE:

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f145

190 Hz OGSE:

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f190

Preprocessed Dataset split into separate frequencies (ex vivo):

PGSE or 0 Hz:

mouseID_sex/timepoint/OGSE_res130150500/OGSE_ res130150500_aveComb_preproc_f000

50 Hz OGSE:

mouseID_sex/timepoint/OGSE_ res130150500/OGSE_ res130150500_aveComb_preproc_f050

80 Hz OGSE:

mouseID_sex/timepoint/OGSE_ res130150500/OGSE_ res130150500_aveComb_preproc_f080

115 Hz OGSE:

mouseID_sex/timepoint/OGSE_ res130150500/OGSE_ res130150500_aveComb_preproc_f115

150 Hz OGSE:

mouseID_sex/timepoint/OGSE_ res130150500/OGSE_ res130150500_aveComb_preproc_f150

.nii.gz

.bvec

.bval

DiffusionDataPreproc

Scalar Maps generated for each frequency (in vivo and ex vivo)

For example, for PGSE or 0 Hz (in vivo):

Axial Diffusivity (AD):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000_AD

Radial Diffusivity (RD):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000_RD

Mean Diffusivity (MD):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000_MD

Fractional Anisotropy (FA):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000_FA

Color Fractional Anisotropy (Color FA):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000_FAvec

Voxelwise Diffusion Dispersion Rate (Λ):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_f000_MD_Gfactor

.nii.gz

DiffusionDataPreproc

Other Scalar Maps

Mean Diffusivity Difference (between 190 Hz OGSE and PGSE (0 Hz)):

mouseID_sex/timepoint/OGSE_5Shapes_1A_5Rep_TR10/OGSE_5Shapes_1A_5Rep_TR10_aveComb_preproc_delMD (in vivo)

mouseID_sex/timepoint/OGSE_res130150500/OGSE_res130150500_aveComb_preproc_delMD (ex vivo)

.nii.gz

  1. As the same brain mask is used for both dMRI datasets, the brain mask has been included in the µA dMRI dataset.

µA dMRI dataset

Folder

Filename

Filename Extensions

Data

µA dMRI scan (complex-valued data):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10 (in vivo)

mouseID_sex/timepoint/uFA_res130150500/uFA_res130150500 (ex vivo)

.dcm

_real.nii.gz

_imaginary.nii.gz

_real.json

_imaginary.json

_method.json

_visu_pars.json

.bmat

.bvec

.bval

Data

µA dMRI scan (after averages are combined):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/ uFA_2Shapes_1A_3Rep_TR10_aveComb (in vivo)

mouseID_sex/timepoint/uFA_res130150500/ uFA_res130150500 (ex vivo)

.nii.gz

.bmat

.bvec

.bval

Data

b0 scan acquired with reverse PE (complex-valued data):mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10_b0_reversePE/ uFA_2Shapes_1A_3Rep_TR10_b0_reversePE (in vivo)

(not acquired for ex vivo)

.dcm

_real.nii.gz

_imaginary.nii.gz

_real.json

_imaginary.json

_method.json

_visu_pars.json

.bmat

.bvec

.bval

Data

b0 scan acquired with reverse PE (after averages are combined):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10_b0_reversePE/ uFA_2Shapes_1A_3Rep_TR10_b0_reversePE_aveComb (in vivo)

.nii.gz

.bmat

.bvec

.bval

Data

Mean b0 volume extracted from dataset after averages are combined (normal PE):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_mean_b0 (in vivo)

.nii.gz

DiffusionDataPreproc

Preprocessed Dataset:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc (in vivo)

mouseID_sex/timepoint/uFA_res130150500/uFA_res130150500_aveComb_preproc (ex vivo)

.nii.gz

.bvec

.bval

.isiso

DiffusionDataPreproc

Scalar maps (in vivo):

(for ex vivo scalar maps, replace ‘uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10’ with ‘uFA_res130150500/uFA_res130150500’)

Axial Diffusivity (AD) – acquired with b1000 LTE volumes:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_AD

Radial Diffusivity (RD) – acquired with b1000 LTE volumes:mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_RD

Mean Diffusivity (MD) – acquired with b1000 LTE volumes:mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_MD

Fractional Anisotropy (FA) – acquired with b1000 LTE volumes:mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_FA

Fractional Anisotropy (FA) – acquired with b2000 LTE volumes:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_b2000_FAColor Fractional Anisotropy (FA) – acquired with b1000 LTE volumes:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_b1000_FAvec

Color Fractional Anisotropy (FA) – acquired with b2000 LTE volumes:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_b2000_FAvec

Microscopic Anisotropy (µA):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_uAMicroscopic Fractional Anisotropy (µFA):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_uFA

Linear Kurtosis – calculated from LTE volumes (KLTE):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_Klin

Isotropic Kurtosis – calculated from STE volumes (KSTE):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_Kiso

.nii.gz

DiffusionDataPreproc

Initial brain mask (for use in EDDY in the preprocessing step):

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_mask (in vivo)

mouseID_sex/timepoint/uFA_res130150500/uFA_res130150500_aveComb_preproc_mask (ex vivo)

.nii.gz

DiffusionDataPreproc

Final brain mask:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_mask_after (in vivo)

mouseID_sex/timepoint/uFA_res130150500/uFA_res130150500_aveComb_preproc_mask_after (ex vivo)

.nii.gz

DiffusionDataPreproc

Mean b0 volume extracted from dataset after preprocessing:

mouseID_sex/timepoint/uFA_2Shapes_1A_3Rep_TR10/uFA_2Shapes_1A_3Rep_TR10_aveComb_preproc_mean_b0 (in vivo)

.nii.gz

Templates and Atlas for Registration

Folder

Filename

Filename Extensions

Registration

Turone Atlas (as downloaded from https://www.nitrc.org/projects/tmbta_2019/):

atlas/TMBTA_Brain_Template

Turone Atlas Labels (as downloaded from https://www.nitrc.org/projects/tmbta_2019/):

atlas/TMBTA_Brain_Labels

.nii

Registration

Downsampled Turone Atlas (to be used for registration):

atlas/TMBTA_Brain_Template_reorient_smoothed0_2_RS_Gaussian

Downsampled Turone Atlas Labels:atlas/TMBTA_Brain_Labels_reorient_RS_Gaussian

.nii.gz

Registration

Study-specific templates:

ANTStemplate_T2/T2_template (T2-weighted template)ANTStemplate_MT/MT_template (MTw template)

ANTStemplate_FA/FA_template (FA template)

.nii.gz

Registration

Registration transforms (Affine transform to register individual images to template space):

In each template folder:

template_contrast_mouseID_sex_timepoint***GenericAffine

where ‘contrast’ is ‘T2,’ ‘b2000_FA,’ or ‘MTon’ and *** are 3 numbers outputted by the ANTs template building command

For example: template_T2_NR1_F_1week600GenericAffine

.mat

Registration

Registration transforms (Symmetric diffeomorphic transform to register individual images to template space):

In each template folder:

template_contrast_mouseID_sex_timepoint***Warp

template_contrast_mouseID_sex_timepoint***InverseWarp

.nii.gz

Registration

Registration transforms (between templates):

FA template to T2 template:

FAtemplate_to_T2template/FA_T2_SynMI0_005_transform0GenericAffine.mat (affine transform)

FAtemplate_to_T2template/FA_T2_SynMI0_005_transform1Warp.nii.gz (symmetric diffeomorphic transform)

FAtemplate_to_T2template/FA_T2_SynMI0_005_transform1InverseWarp.nii.gz (inverse symmetric diffeomorphic transform)

MTw template to T2 template:

MTtemplate_to_T2template/ MT_T2_SynCI0.005_transform0GenericAffine.mat (affine transform)

MTtemplate_to_T2template/ MT_T2_SynCI0.005__transform1Warp.nii.gz (symmetric diffeomorphic transform)

MTtemplate_to_T2template/ MT_T2_SynCI0.005__transform1InverseWarp.nii.gz (inverse symmetric diffeomorphic transform)

 

Registration

Registration transforms (from T2 template to the downsampled atlas):

T2template_to_atlas/T2_atlas_SynMI0_00005_transform0GenericAffine.mat (affine transform)

T2template_to_atlas/T2_atlas_SynMI0_00005_transform1Warp.nii.gz (symmetric diffeomorphic transform)

 

Imaging protocols

To optimize utility of the protocols, imaging protocols were exported from a Bruker ParaVision 6.0.1 system (OGSE and µA dMRI protocols), which was used for data collection, and are included. The files can be imported into the Bruker ParaVision system to run all protocols. ParaVision 6.0.1 compiled binaries for the custom diffusion MRI pulse sequences are available at doi.org/10.17605/OSF.IO/5EUSW, while the other scans used vendor-provided sequences. Imaging protocols and compiled binaries for a Bruker ParaVision 7.0.0 system are also included, for convenience. The diffusion MRI pulse sequence source code is available upon reasonable request.

Technical Validation

As 3D printed custom designed parts and the surface/volume coil were fixed onto a support, which was placed into the scanner, this ensured consistent positioning of the mouse head in the scanner at each session and prevented motion artifacts. Raw and preprocessed dMRI data were visually inspected to ensure good preprocessing results, as shown previously2 and in Fig. 6a (in vivo) and 6b (ex vivo).

Fig. 6
figure 6

In vivo (a) and ex vivo (b) raw and preprocessed dMRI data. Raw data (after combining averages) is shown in the top row and preprocessed data is shown in the bottom row. Representative b = 0 s/mm2 images are shown for both the OGSE and µA protocols. From the OGSE protocol, representative diffusion weighted images from a single diffusion gradient direction are shown from PGSE and OGSE with the highest frequency used in this study (190 Hz (in vivo) and 150 Hz (ex vivo)), at b = 800 s/mm2 (in vivo) and b = 1600 s/mm2 (ex vivo). From the µA protocol, diffusion weighted images from a single diffusion gradient direction are shown from the LTE and STE sequences, at b = 2000 s/mm2 (in vivo) and b = 4000 s/mm2 (ex vivo). Adapted from Rahman et al.2.

Full size image

The only artifact observed in the in vivo data was a banding artifact in the rostral region of the brain in most of the B1 maps, which were acquired as part of the MT protocol. Thus, the MTsat maps included in the repository were generated without applying the B1 correction. Users have the option to turn the B1 correction on or off. If the B1 correction is on, the correction will be applied only to the slices which showed no banding artifact in the B1 map. Although the B1 maps have an artifact and the correction cannot be applied to all brain slices, inhomogeneities in the transmitted RF field are inherently compensated to some degree when calculating MTsat6. The B1 maps were acquired to correct for small residual higher-order dependencies of the MT saturation on the local RF transmit field to further improve spatial uniformity, as suggested by Weiskpof et al.42. Thus, the B1 correction is a finetuning for MTsat maps rather than a substantial part of the calculation, and the MTsat maps can still be analyzed without the correction.

For ex vivo data, as mouse IDs NR7_M and NR8_M were scanned with the skull removed, slight deformation of the tissue is observed at the superior edges of the brain. Mouse IDs NR1_F and NR2_F show banding artifacts in the caudal region of the brain in the B1 maps.

Test-retest reproducibility

Test-retest analysis is an additional tool for technical validation. Test-retest comparisons have been performed using data from two timepoints: Day 3 and Week 12,3. Bland-Altman plots and coefficients of variation (CVs) revealed that most of the μA dMRI metrics are reproducible in both ROI-based and voxelwise analysis, while the OGSE dMRI metrics are only reproducible in ROI-based analysis. MTR and MTsat show high reproducibility (CVs < 10%) in both voxelwise and ROI-based analyses. The previous test-retest analysis also shows that given feasible preclinical sample sizes (10–15), the MRI metrics may provide sensitivity to subtle microstructural changes (6–8%).

Signal-to-noise ratio measurements

For dMRI data, SNR maps were calculated by dividing the powder-averaged magnitude signal (of the combined averages) by the noise. Noise was calculated from each of the real and imaginary components of the complex-valued data as the standard deviation of the background signal from a single average of a single direction divided by \(\sqrt{\left({\rm{number}}\,{\rm{of}}\,{\rm{averages}}\right)\cdot \left({\rm{number}}\,{\rm{of}}\,{\rm{directions}}\right)}\), and averaged over the real and imaginary components. For MT data, SNR maps were calculated by dividing the magnitude signal by the standard deviation of background signal.

To maximize SNR, a surface coil, built in-house, was used for in vivo imaging. As expected with a surface coil, a gradient of SNR change can be seen in the superior-inferior direction of the brain, compared to the commercially available MP30 volume coil, which was used for ex vivo imaging (Fig. 7). This gradient of SNR change does not seem to affect voxel-wise CV maps to the same extent, as shown in Rahman et al.2, which could be due to the denoising quality.

Fig. 7
figure 7

SNR maps of in vivo and ex vivo images. SNR maps for a single b = 0 s/mm2 image are shown for all dMRI protocols, and SNR maps for the powder average of the highest b-values are shown for all protocols (b = 800 s/mm2 for OGSE-190 Hz (in vivo), b = 2000 s/mm2 for μA-STE (in vivo), b = 1600 s/mm2 for OGSE-150 Hz (ex vivo), and b = 4000 s/mm2 for μA-STE (ex vivo)). SNR maps for MTw and PDw scans are shown for MT MRI. Adapted from Rahman et al.2.

Full size image

Usage Notes

Data analysis pipeline - snakemake

The data preprocessing and analysis pipeline was built using Snakemake34, a reproducible and adaptable Python-based workflow management system. Snakemake itself is easily deployable via the Conda package manager (https://conda.io). Instructions and further information can be found at https://snakemake.github.io.

The workflow, called the “Snakefile,” contains all data analysis steps such as DICOM to NIFTI conversion, preprocessing data, and scalar map generation. This involves FSL31, MRtrix338, and ANTs40 commands, as well as MATLAB functions and bash scripts. Users can easily modify and add rules to the pipeline.

Example snakemake usage

Below are example Snakemake commands, which can be run from the command line, to process dicoms to preprocessed data and scalar maps. These instructions have also been included in the README of the code directory. The Snakemake rules used are listed directly below each command. The filepaths and filenames, which the user must change, are in italics, and any number of files can be converted at once. Importantly, most of the code assumes that the dicom or NIFTI filename matches the name of the folder that it is in. All brain masks (for each MRI contrast) have been provided in the repository and the user should copy the masks to their respective folders, as the code assumes that these masks exist. Alternatively, the user can edit the code to run without masks.

Anatomical data

To convert the anatomical dicoms (which include T2 and all MT related dicoms) to NIFTI format, the following Snakemake command can be used:

$ snakemake --cores 1 filepath/{mouse#1_sex/timepoint,mouse#2_sex/timepoint,mouse#3_sex/timepoint}/dicom_foldername/dicom_filename.json

[Rules: dcmTOnii_anat]

For example, a real use case of the above command, with the actual filepaths and filenames to acquire T2-weighted NIFTIs may be:

$ snakemake --cores 1 Data/{NR1_F/Day0,NR1_F/Day3,NR2_F/Day0}/T2_TurboRARE_AX150150500_A16/T2_TurboRARE_AX150150500_A16.json

Before acquiring MT metric maps, users must make a brain mask using the MT-weighted images (with software such as BrainSuite) and save the mask as “MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz_mask.nii.gz” in the folder “MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz.” As brain masks are also provided in the repository, users can also copy the mask, instead of creating a new one. To generate MT metric maps (MTR and MTsat), the following Snakemake command can be used:

$ snakemake --cores 1 filepath/{mouse#1_sex /timepoint,mouse#2_sex /timepoint,mouse#3_sex /timepoint}/ MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz/MTon_GRE_3D_150 × 400_12A_5uT_385FA_3500Hz_mtsat.nii.gz

[Rules: mtsat]

Diffusion MRI data

To convert a number of dicoms to combined averages (in NIFTI format, with partial Fourier reconstruction, correction for frequency and signal drift, and denoising) and generate the initial dMRI brain mask (needed for preprocessing), the following Snakemake command can be used:

$ snakemake --cores 1 filepath/{mouse#1_sex /timepoint,mouse#2_sex /timepoint,mouse#3_sex /timepoint}/dMRI_filename/dMRI_filename_aveComb_preproc_mask.nii.gz

[Rules: dcmTOnii_dMRI, combAve, get_preproc_mask]

The above command assumes that T2-weighted brain masks exist as “T2_TurboRARE_AX150150500_A16_mask.nii.gz” in the folder “T2_TurboRARE_AX150150500_A16,” as this mask is registered to dMRI space to create the initial dMRI brain mask. As the data acquired with reverse phase-encoding do not require an initial mask, since they are combined with the larger datasets (“uFA_2Shapes_1A_3Rep_TR10” and “OGSE_5Shapes_1A_5Rep_TR10”), the command to convert dicoms with reverse phase-encoding to combined averages is:

$ snakemake --cores 1 filepath/{mouse#1_sex /timepoint,mouse#2_sex /timepoint,mouse#3_sex /timepoint}/dMRI_filename/dMRI_filename_aveComb.nii.gz

[Rules: dcmTOnii_dMRI, combAve]

After combined averages and initial dMRI brain masks are generated, preprocessing can be run by this command:

$ snakemake --cores 1 DiffusionDataPreproc/{mouse#1_sex /timepoint,mouse#2_sex /timepoint,mouse#3_sex /timepoint}/dMRI_filename/dMRI_filename_aveComb_preproc.nii.gz

[Rules: dMRIpreproc]

Note that the code assumes that the original NIFTI files are located in the “Data” folder and that FSL is being run from a singularity container. The user can change the code in “dMRIpreproc.sh” located in the folder “code_scidata_paper/dMRIpreproc” to align with their FSL environment. The above command will work with or without reverse phase-encoded data.

After the dMRI preprocessing step, final dMRI brain masks can be created, or the user can use the masks provided in the repository (“dMRI_filename_aveComb_preproc_mask_after.nii.gz”). The code assumes that the masks are named as they are in the repository. Alternatively, the user can acquire dMRI scalar maps without using brain masks by editing the code in the Snakefile. To acquire dMRI scalar maps, the following command can be run:

$ snakemake --cores 1 filepath/{mouse#1_sex /timepoint,mouse#2_sex /timepoint,mouse#3_sex /timepoint}/dMRI_filename/dMRI_filename_aveComb_preproc_mean_b0_Wmask.nii.gz

[Rules: get_dwimetric_maps]

The above command will generate scalar maps as well as a non-diffusion weighted (b0) NIFTI, averaged over all non-diffusion weighted volumes. This mean b0 NIFTI may be used to facilitate registration.

Image registration

The dMRI and MT data were not registered to a template or to the anatomical T2-weighted images to avoid errors from interpolation and registration inaccuracies, and as other researchers may prefer using their own registration pipelines. However, for flexible utility of the dataset, the anatomical T2-weighted images, study-specific templates, a downsampled atlas, and registration transforms have been included in the repository, so registration of the dMRI and MT data to anatomical space or an atlas is possible. Currently, the registration pipeline has been tested only with the in vivo dataset. Users may use the robust registration pipeline, based on ANTs commands, included in the Snakefile or tweak them accordingly. ANTs is an open source software package which comprises tools for image registration, template building and segmentation40. ANTs was chosen due to its flexibility and the robust performance of default ANTs registration parameters. Moreover, the nonlinear deformation algorithm used in ANTs was top ranked in a comparative study43.

The Turone atlas44, downsampled to the resolution of the in vivo T2-weighted images, was used for registration (Fig. 8a). Three study-specific templates, based on all images from all sessions, were created to facilitate the registration process. These templates included a T2 template, an FA template, and an MT-weighted template. Individual images can be registered to the downsampled atlas in three steps, as shown in Fig. 8b,c: (1) the FA maps and MT-weighted images are registered to the FA template and MT-weighted template, respectively; (2) the FA template and MT-weighted template are registered to the T2 template; (3) the T2 template is registered to the downsampled atlas. Each registration step involves an affine transformation, followed by a symmetric diffeomorphic transformation using ANTs’ Symmetric Normalization (SyN) algorithm. The registration transforms resulting from the previous three registration steps can be used to warp all dMRI metric maps and MT metric maps (MTR and MTsat) to the downsampled atlas space. Registration of all images to the atlas allows for voxel-wise analysis and atlas-based region-of-interest analysis. For region-of-interest analysis, atlas labels can be downloaded from https://www.nitrc.org/projects/tmbta_2019/45. Example ANTs commands, used for template-building and generating registration transforms, have been included in the text file ‘ANTs_Registration_Commands.txt.’ All other code to warp metric maps to the downsampled atlas space have been included in the Snakefile.

Fig. 8
figure 8

Schematic of registration steps. (a) The Turone atlas (60 µm isotropic resolution) was downsampled to the resolution of the T2-weighted images. (b) Registration steps to register individual FA maps to the downsampled atlas space. (c) Registration steps to register individual MT-weighted images to the downsampled atlas space. The registration transforms resulting from part (b,c) can be used to warp dMRI and MT metric maps to the downsampled atlas space.

Full size image