Echo planar imaging (EPI) is the method of choice for the majority of functional magnetic resonance imaging (fMRI), yet EPI is prone to geometric distortions and thus misaligns with conventional anatomical reference data. The poor geometric correspondence between functional and anatomical data can lead to severe misplacements and corruption of detected activation patterns. However, recent advances in imaging technology have provided EPI data with increasing quality and resolution. Here we present a framework for deriving cortical surface reconstructions directly from high-resolution EPI-based reference images that provide anatomical models exactly geometric distortion-matched to the functional data. Anatomical EPI data with 1mm isotropic voxel size were acquired using a fast multiple inversion recovery time EPI sequence (MI-EPI) at 7T, from which quantitative T1 maps were calculated. Using these T1 maps, volumetric data mimicking the tissue contrast of standard anatomical data were synthesized using the Bloch equations, and these T1-weighted data were automatically processed using FreeSurfer. The spatial alignment between T2(⁎)-weighted EPI data and the synthetic T1-weighted anatomical MI-EPI-based images was improved compared to the conventional anatomical reference. In particular, the alignment near the regions vulnerable to distortion due to magnetic susceptibility differences was improved, and sampling of the adjacent tissue classes outside of the cortex was reduced when using cortical surface reconstructions derived directly from the MI-EPI reference. The MI-EPI method therefore produces high-quality anatomical data that can be automatically segmented with standard software, providing cortical surface reconstructions that are geometrically matched to the BOLD fMRI data.

Diffusion magnetic resonance imaging (MRI) methods for axon diameter mapping benefit from higher maximum gradient strengths than are currently available on commercial human scanners. Using a dedicated high-gradient 3T human MRI scanner with a maximum gradient strength of 300 mT/m, we systematically studied the effect of gradient strength on in vivo axon diameter and density estimates in the human corpus callosum. Pulsed gradient spin echo experiments were performed in a single scan session lasting approximately 2h on each of three human subjects. The data were then divided into subsets with maximum gradient strengths of 77, 145, 212, and 293 mT/m and diffusion times encompassing short (16 and 25 ms) and long (60 and 94 ms) diffusion time regimes. A three-compartment model of intra-axonal diffusion, extra-axonal diffusion, and free diffusion in cerebrospinal fluid was fitted to the data using a Markov chain Monte Carlo approach. For the acquisition parameters, model, and fitting routine used in our study, it was found that higher maximum gradient strengths decreased the mean axon diameter estimates by two to three fold and decreased the uncertainty in axon diameter estimates by more than half across the corpus callosum. The exclusive use of longer diffusion times resulted in axon diameter estimates that were up to two times larger than those obtained with shorter diffusion times. Axon diameter and density maps appeared less noisy and showed improved contrast between different regions of the corpus callosum with higher maximum gradient strength. Known differences in axon diameter and density between the genu, body, and splenium of the corpus callosum were preserved and became more reproducible at higher maximum gradient strengths. Our results suggest that an optimal q-space sampling scheme for estimating in vivo axon diameters should incorporate the highest possible gradient strength. The improvement in axon diameter and density estimates that we demonstrate from increasing maximum gradient strength will inform protocol development and encourage the adoption of higher maximum gradient strengths for use in commercial human scanners.

Irreversible white matter (WM) damage, including severe demyelination and axonal loss, is a main determinant of long-term disability in multiple sclerosis (MS). Non-invasive detection of changes in microstructural WM integrity in the disease is challenging since commonly used imaging metrics lack the necessary sensitivity, especially in the early phase of the disease. This study aims at assessing microstructural WM abnormalities in early-stage MS by using ultra-high gradient strength multi-shell diffusion MRI and the restricted signal fraction (FR) from the Composite Hindered and Restricted Model of Diffusion (CHARMED), a metric sensitive to the volume fraction of axons. In 22 early MS subjects (disease duration ≤5 years) and 15 age-matched healthy controls, restricted fraction estimates were obtained through the CHARMED model along with conventional Diffusion Tensor Imaging (DTI) metrics. All imaging parameters were compared cross-sectionally between the MS subjects and controls both in WM lesions and normal-appearing white matter (NAWM). We found a significant reduction in FR focally in WM lesions and widespread in the NAWM in MS patients relative to controls (corrected p < .05). Signal fraction changes in NAWM were not driven by perilesional tissue, nor were they influenced by proximity to the ventricles, challenging the hypothesis of an outside-in pathological process driven by CSF-mediated immune cytotoxic factors. No significant differences were found in conventional DTI parameters. In a cross-validated classification task, FR showed the largest effect size and outperformed all other diffusion imaging metrics in discerning lesions from contralateral NAWM. Taken together, our data provide evidence for the presence of widespread microstructural changes in the NAWM in early MS stages that are, at least in part, unrelated to focal demyelinating lesions. Interestingly, these pathological changes were not yet detectable by conventional diffusion imaging at this early disease stage, highlighting the sensitivity and value of multi-shell diffusion imaging for better characterizing axonal microstructure in MS.

Diffusion tensor imaging (DTI) enables non-invasive, cyto-architectural mapping of in vivo tissue microarchitecture through voxel-wise mathematical modeling of multiple magnetic resonance imaging (MRI) acquisitions, each differently sensitized to water diffusion. DTI computations are fundamentally estimation processes and are sensitive to noise and artifacts. Despite widespread adoption in the neuroimaging community, maintaining consistent DTI data quality remains challenging given the propensity for patient motion, artifacts associated with fast imaging techniques, and the possibility of hardware changes/failures. Furthermore, the quantity of data acquired per voxel, the non-linear estimation process, and numerous potential use cases complicate traditional visual data inspection approaches. Currently, quality inspection of DTI data has relied on visual inspection and individual processing in DTI analysis software programs (e.g. DTIPrep, DTI-studio). However, recent advances in applied statistical methods have yielded several different metrics to assess noise level, artifact propensity, quality of tensor fit, variance of estimated measures, and bias in estimated measures. To date, these metrics have been largely studied in isolation. Herein, we select complementary metrics for integration into an automatic DTI analysis and quality assurance pipeline. The pipeline completes in 24 hours, stores statistical outputs, and produces a graphical summary quality analysis (QA) report. We assess the utility of this streamlined approach for empirical quality assessment on 608 DTI datasets from pediatric neuroimaging studies. The efficiency and accuracy of quality analysis using the proposed pipeline is compared with quality analysis based on visual inspection. The unified pipeline is found to save a statistically significant amount of time (over 70%) while improving the consistency of QA between a DTI expert and a pool of research associates. Projection of QA metrics to a low dimensional manifold reveal qualitative, but clear, QA-study associations and suggest that automated outlier/anomaly detection would be feasible.

The neurophysiological basis of variability in the latency of evoked neural responses has been of interest for decades. We describe a method to identify white matter pathways that may contribute to inter-individual variability in the timing of neural activity. We investigated the relation of the latency of peak visual responses in occipital cortex as measured by magnetoencephalography (MEG) to fractional anisotropy (FA) in the entire brain as measured with diffusion tensor imaging (DTI) in eight healthy young adults. This method makes no assumptions about the anatomy of white matter connections. Visual responses were evoked during a saccadic paradigm and were time-locked to arrival at a saccadic goal. The latency of the peak visual response was inversely related to FA in bilateral parietal and right lateral frontal white matter adjacent to cortical regions that modulate early visual responses. These relations suggest that biophysical properties of white matter affect the timing of early visual responses. This preliminary report demonstrates a non-invasive, unbiased method to relate the timing information from evoked-response experiments to the biophysical properties of white matter measured with DTI.

Axon diameter mapping using high-gradient diffusion MRI has generated great interest as a noninvasive tool for studying trends in axonal size in the human brain. One of the main barriers to mapping axon diameter across the whole brain is accounting for complex white matter fiber configurations (e.g., crossings and fanning), which are prevalent throughout the brain. Here, we present a framework for generalizing axon diameter index estimation to the whole brain independent of the underlying fiber orientation distribution using the spherical mean technique (SMT). This approach is shown to significantly benefit from the use of real-valued diffusion data with Gaussian noise, which reduces the systematic bias in the estimated parameters resulting from the elevation of the noise floor when using magnitude data with Rician noise. We demonstrate the feasibility of obtaining whole-brain orientationally invariant estimates of axon diameter index and relative volume fractions in six healthy human volunteers using real-valued diffusion data acquired on a dedicated high-gradient 3-Tesla human MRI scanner with 300 mT/m maximum gradient strength. The trends in axon diameter index are consistent with known variations in axon diameter from histology and demonstrate the potential of this generalized framework for revealing coherent patterns in axonal structure throughout the living human brain. The use of real-valued diffusion data provides a viable solution for eliminating the Rician noise floor and should be considered for all spherical mean approaches to microstructural parameter estimation.

Strong gradient systems can improve the signal-to-noise ratio of diffusion MRI measurements and enable a wider range of acquisition parameters that are beneficial for microstructural imaging. We present a comprehensive diffusion MRI dataset of 26 healthy participants acquired on the MGH-USC 3 T Connectome scanner equipped with 300 mT/m maximum gradient strength and a custom-built 64-channel head coil. For each participant, the one-hour long acquisition systematically sampled the accessible diffusion measurement space, including two diffusion times (19 and 49 ms), eight gradient strengths linearly spaced between 30 mT/m and 290 mT/m for each diffusion time, and 32 or 64 uniformly distributed directions. The diffusion MRI data were preprocessed to correct for gradient nonlinearity, eddy currents, and susceptibility induced distortions. In addition, scan/rescan data from a subset of seven individuals were also acquired and provided. The MGH Connectome Diffusion Microstructure Dataset (CDMD) may serve as a test bed for the development of new data analysis methods, such as fiber orientation estimation, tractography and microstructural modelling.

The temporal monocular crescent (TMC) is the most peripheral portion of the visual field whose perception relies solely on input from the ipsilateral eye. According to a handful of post-mortem histological studies in humans and non-human primates, the TMC is represented visuotopically within the most anterior portion of the primary visual cortical area (V1). However, functional evidence of the TMC visuotopic representation in human visual cortex is rare, mostly due to the small size of the TMC representation (~6% of V1) and due to the technical challenges of stimulating the most peripheral portion of the visual field inside the MRI scanner. In this study, by taking advantage of custom-built MRI-compatible visual stimulation goggles with curved displays, we successfully stimulated the TMC region of the visual field in eight human subjects, half of them right-eye dominant, inside a 3 ​T MRI scanner. This enabled us to localize the representation of TMC, along with the blind spot representation (another visuotopic landmark in V1), in all volunteers, which match the expected spatial pattern based on prior anatomical studies. In all hemispheres, the TMC visuotopic representation was localized along the peripheral border of V1, within the most anterior portion of the calcarine sulcus, without any apparent extension into the second visual area (V2). We further demonstrate the reliability of this localization within/across experimental sessions, and consistency in the spatial location of TMC across individuals after accounting for inter-subject structural differences.

Estimation of causal interactions between brain areas is necessary for elucidating large-scale functional brain networks underlying behavior and cognition. Granger causality analysis of time series data can quantitatively estimate directional information flow between brain regions. Here, we show that such estimates are significantly improved when the temporal sampling rate of functional magnetic resonance imaging (fMRI) is increased 20-fold. Specifically, healthy volunteers performed a simple visuomotor task during blood oxygenation level dependent (BOLD) contrast based whole-head inverse imaging (InI). Granger causality analysis based on raw InI BOLD data sampled at 100-ms resolution detected the expected causal relations, whereas when the data were downsampled to the temporal resolution of 2 s typically used in echo-planar fMRI, the causality could not be detected. An additional control analysis, in which we SINC interpolated additional data points to the downsampled time series at 0.1-s intervals, confirmed that the improvements achieved with the real InI data were not explainable by the increased time-series length alone. We therefore conclude that the high-temporal resolution of InI improves the Granger causality connectivity analysis of the human brain.

This paper compares a range of compartment models for diffusion MRI data on in vivo human acquisitions from a standard 60mT/m system (Philips 3T Achieva) and a unique 300mT/m system (Siemens Connectom). The key aim is to determine whether both systems support broadly the same models or whether the Connectom higher gradient system supports significantly more complex models. A single volunteer underwent 8h of acquisition on each system to provide uniquely wide and dense sampling of the available space of pulsed-gradient spin-echo (PGSE) measurements. We select a set of promising models from the wide set of possible three-compartment models for in vivo white matter (WM) that previous work and preliminary experiments suggest as strong candidates, but extend them to fit for compartmental T2 and diffusivity. We focus on the corpus callosum where the WM fibre architecture is simplest and compare their ability to explain the measured data, using Akaike's information criterion (AIC), and to predict unseen data, using cross-validation. We also compare the stability of parameter estimates in the presence of i) noise, using bootstrapping, and ii) spatial variation, using visual assessment and comparison with anatomical knowledge. Broadly similar models emerge from the AIC and cross-validation experiments in both data sets. Specifically, a three-compartment model consisting of either a Bingham distribution of sticks or a Cylinder for the intracellular compartment, an anisotropic diffusion tensor (DT) model for the extracellular compartment, as well as an isotropic CSF compartment, performs consistently well. However, various other models also perform well and no single model emerges as clear winner. The WM data (with virtually no CSF contamination) do not support compartmental T2 but partially support compartmental diffusivity. Evaluation of parameter stability favours simpler models than those identified by AIC or cross-validation. They suggest that the level of complexity in models underpinning currently popular microstructure imaging techniques such as NODDI, CHARMED, or ActiveAx, where the number of free parameters is about 4 or 5 rather than 10 or 11, may reflect the level of complexity achievable for a useful technique on current systems, although the 300mT/m data may support more complex models.

Physiological noise arising from a variety of sources can significantly degrade the detection of task-related activity in BOLD-contrast fMRI experiments. If whole head spatial coverage is desired, effective suppression of oscillatory physiological noise from cardiac and respiratory fluctuations is quite difficult without external monitoring, since traditional EPI acquisition methods cannot sample the signal rapidly enough to satisfy the Nyquist sampling theorem, leading to temporal aliasing of noise. Using a combination of high speed magnetic resonance inverse imaging (InI) and digital filtering, we demonstrate that it is possible to suppress cardiac and respiratory noise without auxiliary monitoring, while achieving whole head spatial coverage and reasonable spatial resolution. Our systematic study of the effects of different moving average (MA) digital filters demonstrates that a MA filter with a 2 s window can effectively reduce the variance in the hemodynamic baseline signal, thereby achieving 57%-58% improvements in peak z-statistic values compared to unfiltered InI or spatially smoothed EPI data (FWHM = 8.6 mm). In conclusion, the high temporal sampling rates achievable with InI permit significant reductions in physiological noise using standard temporal filtering techniques that result in significant improvements in hemodynamic response estimation.

The MGH-USC CONNECTOM MRI scanner housed at the Massachusetts General Hospital (MGH) is a major hardware innovation of the Human Connectome Project (HCP). The 3T CONNECTOM scanner is capable of producing a magnetic field gradient of up to 300 mT/m strength for in vivo human brain imaging, which greatly shortens the time spent on diffusion encoding, and decreases the signal loss due to T2 decay. To demonstrate the capability of the novel gradient system, data of healthy adult participants were acquired for this MGH-USC Adult Diffusion Dataset (N=35), minimally preprocessed, and shared through the Laboratory of Neuro Imaging Image Data Archive (LONI IDA) and the WU-Minn Connectome Database (ConnectomeDB). Another purpose of sharing the data is to facilitate methodological studies of diffusion MRI (dMRI) analyses utilizing high diffusion contrast, which perhaps is not easily feasible with standard MR gradient system. In addition, acquisition of the MGH-Harvard-USC Lifespan Dataset is currently underway to include 120 healthy participants ranging from 8 to 90 years old, which will also be shared through LONI IDA and ConnectomeDB. Here we describe the efforts of the MGH-USC HCP consortium in acquiring and sharing the ultra-high b-value diffusion MRI data and provide a report on data preprocessing and access. We conclude with a demonstration of the example data, along with results of standard diffusion analyses, including q-ball Orientation Distribution Function (ODF) reconstruction and tractography.

The parameter selection for diffusion MRI experiments is dominated by the "k-q tradeoff" whereby the Signal to Noise Ratio (SNR) of the images is traded for either high spatial resolution (determined by the maximum k-value collected) or high diffusion sensitivity (effected by b-value or the q vector) but usually not both. Furthermore, different brain regions (such as gray matter and white matter) likely require different tradeoffs between these parameters due to the size of the structures to be visualized or the length-scale of the microstructure being probed. In this case, it might be advantageous to combine information from two scans - a scan with high q but low k (high angular resolution in diffusion but low spatial resolution in the image domain) to provide maximal information about white matter fiber crossing, and one low q but high k (low angular resolution but high spatial resolution) for probing the cortex. In this study, we propose a method, termed HIgh b-value and high Resolution Integrated Diffusion (HIBRID) imaging, for acquiring and combining the information from these two complementary types of scan with the goal of studying diffusion in the cortex without compromising white matter fiber information. The white-gray boundary and pial surface obtained from anatomical scans are incorporated as prior information to guide the fusion. We study the complementary advantages of the fused datasets, and assess the quality of the HIBRID data compared to either alone.

Automatic cerebral cortical surface reconstruction is a useful tool for cortical anatomy quantification, analysis and visualization. Recently, the Human Connectome Project and several studies have shown the advantages of using T1-weighted magnetic resonance (MR) images with sub-millimeter isotropic spatial resolution instead of the standard 1-mm isotropic resolution for improved accuracy of cortical surface positioning and thickness estimation. Nonetheless, sub-millimeter resolution images are noisy by nature and require averaging multiple repetitions to increase the signal-to-noise ratio for precisely delineating the cortical boundary. The prolonged acquisition time and potential motion artifacts pose significant barriers to the wide adoption of cortical surface reconstruction at sub-millimeter resolution for a broad range of neuroscientific and clinical applications. We address this challenge by evaluating the cortical surface reconstruction resulting from denoised single-repetition sub-millimeter T1-weighted images. We systematically characterized the effects of image denoising on empirical data acquired at 0.6 mm isotropic resolution using three classical denoising methods, including denoising convolutional neural network (DnCNN), block-matching and 4-dimensional filtering (BM4D) and adaptive optimized non-local means (AONLM). The denoised single-repetition images were found to be highly similar to 6-repetition averaged images, with a low whole-brain averaged mean absolute difference of ~0.016, high whole-brain averaged peak signal-to-noise ratio of ~33.5 dB and structural similarity index of ~0.92, and minimal gray matter-white matter contrast loss (2% to 9%). The whole-brain mean absolute discrepancies in gray matter-white matter surface placement, gray matter-cerebrospinal fluid surface placement and cortical thickness estimation were lower than 165 μm, 155 μm and 145 μm-sufficiently accurate for most applications. These discrepancies were approximately one third to half of those from 1-mm isotropic resolution data. The denoising performance was equivalent to averaging ~2.5 repetitions of the data in terms of image similarity, and 1.6-2.2 repetitions in terms of the cortical surface placement accuracy. The scan-rescan variability of the cortical surface positioning and thickness estimation was lower than 170 μm. Our unique dataset and systematic characterization support the use of denoising methods for improved cortical surface reconstruction at sub-millimeter resolution.

To estimate microstructure-related parameters from diffusion MRI data, biophysical models make strong, simplifying assumptions about the underlying tissue. The extent to which many of these assumptions are valid remains an open research question. This study was inspired by the disparity between the estimated intra-axonal axial diffusivity from literature and that typically assumed by the Neurite Orientation Dispersion and Density Imaging (NODDI) model (d∥=1.7μm2/ms). We first demonstrate how changing the assumed axial diffusivity results in considerably different NODDI parameter estimates. Second, we illustrate the ability to estimate axial diffusivity as a free parameter of the model using high b-value data and an adapted NODDI framework. Using both simulated and in vivo data we investigate the impact of fitting to either real-valued or magnitude data, with Gaussian and Rician noise characteristics respectively, and what happens if we get the noise assumptions wrong in this high b-value and thus low SNR regime. Our results from real-valued human data estimate intra-axonal axial diffusivities of ∼2-2.5μm2/ms, in line with current literature. Crucially, our results demonstrate the importance of accounting for both a rectified noise floor and/or a signal offset to avoid biased parameter estimates when dealing with low SNR data.

Recently proposed dynamic magnetic resonance (MR) inverse imaging (InI) is a novel parallel imaging reconstruction technique capable of improving the temporal resolution of blood-oxygen level-dependent (BOLD) contrast functional MRI (fMRI) to the order of milliseconds at the cost of moderate spatial resolution. Volumetric InI reconstructs spatial information from projection data by solving ill-posed inverse problems using simultaneous acquisitions from a RF coil array. Previously a spatial filtering technique based on linearly constrained minimum variance (LCMV) beamformer was suggested to localize the hemodynamic changes of dynamic InI data with improved spatial resolution and sensitivity. Here we report an advancement of the spatial filtering method, which combines the eigenspace projection of the measured data and the L1-norm minimization of the spatial filters' output noise amplitude, to further improve the detection power of BOLD contrast fMRI data. Using numerical simulation and in vivo data, we demonstrate that this eigenspace linearly constrained minimum amplitude (eLCMA) beamformer can reconstruct spatiotemporal hemodynamic signals with high statistical significance values and high spatial resolution in event-related two-choice reaction time visuomotor experiments.

The acquisition time of BOLD contrast functional MRI (fMRI) data with whole-brain coverage typically requires a sampling rate of one volume in 1-3s. Although the volumetric sampling time of a few seconds is adequate for measuring the sluggish hemodynamic response (HDR) to neuronal activation, faster sampling of fMRI might allow for monitoring of rapid physiological fluctuations and detection of subtle neuronal activation timing information embedded in BOLD signals. Previous studies utilizing a highly accelerated volumetric MR inverse imaging (InI) technique have provided a sampling rate of one volume per 100 ms with 5mm spatial resolution. Here, we propose a novel modification of this technique, the echo-shifted InI, which allows TE to be longer than TR, to measure BOLD fMRI at an even faster sampling rate of one volume per 25 ms with whole-brain coverage. Compared with conventional EPI, echo-shifted InI provided an 80-fold speedup with similar spatial resolution and less than 2-fold temporal SNR loss. The capability of echo-shifted InI to detect HDR timing differences was tested empirically. At the group level (n=6), echo-spaced InI was able to detect statistically significant HDR timing differences of as low as 50 ms in visual stimulus presentation. At the level of individual subjects, significant differences in HDR timing were detected for 400 ms stimulus-onset differences. Our results also show that the temporal resolution of 25 ms is necessary for maintaining the temporal detecting capability at this level. With the capabilities of being able to distinguish the timing differences in the millisecond scale, echo-shifted InI could be a useful fMRI tool for obtaining temporal information at a time scale closer to that of neuronal dynamics.

The engineering of a 3 T human MRI scanner equipped with 300 mT/m gradients - the strongest gradients ever built for an in vivo human MRI scanner - was a major component of the NIH Blueprint Human Connectome Project (HCP). This effort was motivated by the HCP's goal of mapping, as completely as possible, the macroscopic structural connections of the in vivo healthy, adult human brain using diffusion tractography. Yet, the 300 mT/m gradient system is well suited to many additional types of diffusion measurements. Here, we present three initial applications of the 300 mT/m gradients that fall outside the immediate scope of the HCP. These include: 1) diffusion tractography to study the anatomy of consciousness and the mechanisms of brain recovery following traumatic coma; 2) q-space measurements of axon diameter distributions in the in vivo human brain and 3) postmortem diffusion tractography as an adjunct to standard histopathological analysis. We show that the improved sensitivity and diffusion-resolution provided by the gradients are rapidly enabling human applications of techniques that were previously possible only for in vitro and animal models on small-bore scanners, thereby creating novel opportunities to map the microstructure of the human brain in health and disease.

Diffusion tensor magnetic resonance imaging (DTI) is unsurpassed in its ability to map tissue microstructure and structural connectivity in the living human brain. Nonetheless, the angular sampling requirement for DTI leads to long scan times and poses a critical barrier to performing high-quality DTI in routine clinical practice and large-scale research studies. In this work we present a new processing framework for DTI entitled DeepDTI that minimizes the data requirement of DTI to six diffusion-weighted images (DWIs) required by conventional voxel-wise fitting methods for deriving the six unique unknowns in a diffusion tensor using data-driven supervised deep learning. DeepDTI maps the input non-diffusion-weighted (b ​= ​0) image and six DWI volumes sampled along optimized diffusion-encoding directions, along with T1-weighted and T2-weighted image volumes, to the residuals between the input and high-quality output b = 0 image and DWI volumes using a 10-layer three-dimensional convolutional neural network (CNN). The inputs and outputs of DeepDTI are uniquely formulated, which not only enables residual learning to boost CNN performance but also enables tensor fitting of resultant high-quality DWIs to generate orientational DTI metrics for tractography. The very deep CNN used by DeepDTI leverages the redundancy in local and non-local spatial information and across diffusion-encoding directions and image contrasts in the data. The performance of DeepDTI was systematically quantified in terms of the quality of the output images, DTI metrics, DTI-based tractography and tract-specific analysis results. We demonstrate rotationally-invariant and robust estimation of DTI metrics from DeepDTI that are comparable to those obtained with two b ​= ​0 images and 21 DWIs for the primary eigenvector derived from DTI and two b ​= ​0 images and 26-30 DWIs for various scalar metrics derived from DTI, achieving 3.3-4.6 × ​acceleration, and twice as good as those of a state-of-the-art denoising algorithm at the group level. The twenty major white-matter tracts can be accurately identified from the tractography of DeepDTI results. The mean distance between the core of the major white-matter tracts identified from DeepDTI results and those from the ground-truth results using 18 ​b ​= ​0 images and 90 DWIs measures around 1-1.5 ​mm. DeepDTI leverages domain knowledge of diffusion MRI physics and power of deep learning to render DTI, DTI-based tractography, major white-matter tracts identification and tract-specific analysis more feasible for a wider range of neuroscientific and clinical studies.

We present a whole-brain in vivo diffusion MRI (dMRI) dataset acquired at 760 μm isotropic resolution and sampled at 1260 q-space points across 9 two-hour sessions on a single healthy participant. The creation of this benchmark dataset is possible through the synergistic use of advanced acquisition hardware and software including the high-gradient-strength Connectom scanner, a custom-built 64-channel phased-array coil, a personalized motion-robust head stabilizer, a recently developed SNR-efficient dMRI acquisition method, and parallel imaging reconstruction with advanced ghost reduction algorithm. With its unprecedented resolution, SNR and image quality, we envision that this dataset will have a broad range of investigational, educational, and clinical applications that will advance the understanding of human brain structures and connectivity. This comprehensive dataset can also be used as a test bed for new modeling, sub-sampling strategies, denoising and processing algorithms, potentially providing a common testing platform for further development of in vivo high resolution dMRI techniques. Whole brain anatomical T1-weighted and T2-weighted images at submillimeter scale along with field maps are also made available.