Small volume blood-brain barrier opening in macaques with a 1 MHz ultrasound phased array.

The use of focused ultrasound to open the blood-brain barrier (BBB) has the potential to deliver drugs to specific regions of the brain. The size of the BBB opening and ability to localize the opening determines the spatial extent and is a limiting factor in many applications of BBB opening where targeting a small brain region is desired. Here we evaluate the performance of a system designed for small opening volumes and highlight the unique challenges associated with pushing the spatial precision of this technique. To achieve small volume openings in cortical regions of the macaque brain, we tested a custom 1 MHz array transducer integrated into a magnetic resonance image-guided focused ultrasound system. Using real-time cavitation monitoring, we demonstrated twelve instances of single sonication, small volume BBB opening with average volumes of 59 ± 37 mm3 and 184 ± 2 mm3 in cortical and subcortical targets, respectively. We found high correlation between subject-specific acoustic simulations and observed openings when incorporating grey matter segmentation (R2 = 0.8577), and the threshold for BBB opening based on simulations was 0.53 MPa. Analysis of MRI-based safety assessment and cavitation signals indicate a safe pressure range for 1 MHz BBB opening and suggest that our system can be used to deliver drugs and gene therapy to small brain regions.

Disrupting nociceptive information processing flow through transcranial focused ultrasound neuromodulation of thalamic nuclei.

BACKGROUND: MRI-guided transcranial focused ultrasound (MRgFUS) as a next-generation neuromodulation tool can precisely target and stimulate deep brain regions with high spatial selectivity. Combined with MR-ARFI (acoustic radiation force imaging) and using fMRI BOLD signal as functional readouts, our previous studies have shown that low-intensity FUS can excite or suppress neural activity in the somatosensory cortex. OBJECTIVE: To investigate whether low-intensity FUS can suppress nociceptive heat stimulation-induced responses in thalamic nuclei during hand stimulation, and to determine how this suppression influences the information processing flow within nociception networks. FINDINGS: BOLD fMRI activations evoked by 47.5 °C heat stimulation of hand were detected in 24 cortical regions, which belong to sensory, affective, and cognitive nociceptive networks. Concurrent delivery of low-intensity FUS pulses (650 kHz, 550 kPa) to the predefined heat nociceptive stimulus-responsive thalamic centromedial_parafascicular (CM_para), mediodorsal (MD), ventral_lateral (VL_ and ventral_lateral_posteroventral (VLpv) nuclei suppressed their heat responses. Off-target cortical areas exhibited reduced, enhanced, or no significant fMRI signal changes, depending on the specific areas. Differentiable thalamocortical information flow during the processing of nociceptive heat input was observed, as indicated by the time to reach 10% or 30% of the heat-evoked BOLD signal peak. Suppression of thalamic heat responses significantly altered nociceptive processing flow and direction between the thalamus and cortical areas. Modulation of contralateral versus ipsilateral areas by unilateral thalamic activity differed. Signals detected in high-order cortical areas, such as dorsal frontal (DFC) and ventrolateral prefrontal (vlPFC) cortices, exhibited faster response latencies than sensory areas. CONCLUSIONS: The concurrent delivery of FUS suppressed nociceptive heat response in thalamic nuclei and disrupted the nociceptive network. This study offers new insights into the causal functional connections within the thalamocortical networks and demonstrates the modulatory effects of low-intensity FUS on nociceptive information processing.

Deep whole brain segmentation of 7T structural MRI.

7T magnetic resonance imaging (MRI) has the potential to drive our understanding of human brain function through new contrast and enhanced resolution. Whole brain segmentation is a key neuroimaging technique that allows for region-by-region analysis of the brain. Segmentation is also an important preliminary step that provides spatial and volumetric information for running other neuroimaging pipelines. Spatially localized atlas network tiles (SLANT) is a popular 3D convolutional neural network (CNN) tool that breaks the whole brain segmentation task into localized sub-tasks. Each sub-task involves a specific spatial location handled by an independent 3D convolutional network to provide high resolution whole brain segmentation results. SLANT has been widely used to generate whole brain segmentations from structural scans acquired on 3T MRI. However, the use of SLANT for whole brain segmentation from structural 7T MRI scans has not been successful due to the inhomogeneous image contrast usually seen across the brain in 7T MRI. For instance, we demonstrate the mean percent difference of SLANT label volumes between a 3T scan-rescan is approximately 1.73%, whereas its 3T-7T scan-rescan counterpart has higher differences around 15.13%. Our approach to address this problem is to register the whole brain segmentation performed on 3T MRI to 7T MRI and use this information to finetune SLANT for structural 7T MRI. With the finetuned SLANT pipeline, we observe a lower mean relative difference in the label volumes of ~8.43% acquired from structural 7T MRI data. Dice similarity coefficient between SLANT segmentation on the 3T MRI scan and the after finetuning SLANT segmentation on the 7T MRI increased from 0.79 to 0.83 with p<0.01. These results suggest finetuning of SLANT is a viable solution for improving whole brain segmentation on high resolution 7T structural imaging.

Small volume blood-brain barrier opening in macaques with a 1 MHz ultrasound phased array.

Focused ultrasound blood-brain barrier (BBB) opening is a promising tool for targeted delivery of therapeutic agents into the brain. The volume of opening determines the extent of therapeutic administration and sets a lower bound on the size of targets which can be selectively treated. We tested a custom 1 MHz array transducer optimized for cortical regions in the macaque brain with the goal of achieving small volume openings. We integrated this device into a magnetic resonance image guided focused ultrasound system and demonstrated twelve instances of small volume BBB opening with average opening volumes of 59 ± 37 mm 3 and 184 ± 2 mm 3 in cortical and subcortical targets, respectively. We developed real-time cavitation monitoring using a passive cavitation detector embedded in the array and characterized its performance on a bench-top flow phantom mimicking transcranial BBB opening procedures. We monitored cavitation during in-vivo procedures and compared cavitation metrics against opening volumes and safety outcomes measured with FLAIR and susceptibility weighted MR imaging. Our findings show small BBB opening at cortical targets in macaques and characterize the safe pressure range for 1 MHz BBB opening. Additionally, we used subject-specific simulations to investigate variance in measured opening volumes and found high correlation (R 2 = 0.8577) between simulation predictions and observed measurements. Simulations suggest the threshold for 1 MHz BBB opening was 0.53 MPa. This system enables BBB opening for drug delivery and gene therapy to be targeted to more specific brain regions.

Differential dose responses of transcranial focused ultrasound at brain regions indicate causal interactions.

We have previously shown that focused ultrasound (FUS) pulses in low pressure range exerted bidirectional and brain state-dependent neuromodulation in the nonhuman primate somatosensory cortices by fMRI. Here we aim to gain insights about the proposed neuron selective modulation of FUS and probe feedforward versus feedback interactions by simultaneously quantifying the stimulus (FUS pressures: 925, 425, 250 kPa) and response (% BOLD fMRI changes) function at the targeted area 3a/3b and off-target cortical areas at 7T. In resting-state, lowered intensities of FUS resulted in decreased fMRI signal changes at the target area 3a/3b and off-target area 1/2, S2, MCC, insula and auditory cortex, and no signal difference in thalamic VPL and MD nuclei. In activated states, concurrent high-intensity FUS significantly enhanced touch-evoked signals in area 1/2. Medium- and low-intensity FUS significantly suppressed touch-evoked BOLD signals in all areas except in the auditory cortex, VPL and MD thalamic nuclei. Distinct state dependent and dose-response curves led us to hypothesize that FUS's neuromodulatory effects may be mediated through preferential activation of different populations of neurons. Area 3a/3b may have distinct causal feedforward and feedback interactions with Area 1/2, S2, MCC, insula, and VPL. FUS offers a noninvasive neural stimulation tool for dissecting brain circuits and probing causal functional connections.

Ultra-high-resolution Mapping of Cortical Layers 3T-Guided 7T MRI.

7T MRI provides unprecedented resolution for examining human brain anatomy in vivo. For example, 7T MRI enables deep thickness measurement of laminar subdivisions in the right fusiform area. Existing laminar thickness measurement on 7T is labor intensive, and error prone since the visual inspection of the image is typically along one of the three orthogonal planes (axial, coronal, or sagittal view). To overcome this, we propose a new analytics tool that allows flexible quantification of cortical thickness on a 2D plane that is orthogonal to the cortical surface (beyond axial, coronal, and sagittal views) based on the 3D computational surface reconstruction. The proposed method further distinguishes high quality 2D planes and the low-quality ones by automatically inspecting the angles between the surface normals and slice direction. In our approach, we acquired a pair of 3T and 7T scans (same subject). We extracted the brain surfaces from the 3T scan using MaCRUISE and projected the surface to the 7T scan's space. After computing the angles between the surface normals and axial direction vector, we found that 18.58% of surface points were angled at more than 80° with the axial direction vector and had 2D axial planes with visually distinguishable cortical layers. 15.12% of the surface points with normal vectors angled at 30° or lesser with the axial direction, had poor 2D axial slices for visual inspection of the cortical layers. This effort promises to dramatically extend the area of cortex that can be quantified with ultra-high resolution in-plane imaging methods.

MASiVar: Multisite, multiscanner, and multisubject acquisitions for studying variability in diffusion weighted MRI.

PURPOSE: Diffusion-weighted imaging allows investigators to identify structural, microstructural, and connectivity-based differences between subjects, but variability due to session and scanner biases is a challenge. METHODS: To investigate DWI variability, we present MASiVar, a multisite data set consisting of 319 diffusion scans acquired at 3 T from b = 1000 to 3000 s/mm2 across 14 healthy adults, 83 healthy children (5 to 8 years), three sites, and four scanners as a publicly available, preprocessed, and de-identified data set. With the adult data, we demonstrate the capacity of MASiVar to simultaneously quantify the intrasession, intersession, interscanner, and intersubject variability of four common DWI processing approaches: (1) a tensor signal representation, (2) a multi-compartment neurite orientation dispersion and density model, (3) white-matter bundle segmentation, and (4) structural connectomics. Respectively, we evaluate region-wise fractional anisotropy, mean diffusivity, and principal eigenvector; region-wise CSF volume fraction, intracellular volume fraction, and orientation dispersion index; bundle-wise shape, volume, fractional anisotropy, and length; and whole connectome correlation and maximized modularity, global efficiency, and characteristic path length. RESULTS: We plot the variability in these measures at each level and find that it consistently increases with intrasession to intersession to interscanner to intersubject effects across all processing approaches and that sometimes interscanner variability can approach intersubject variability. CONCLUSIONS: This study demonstrates the potential of MASiVar to more globally investigate DWI variability across multiple levels and processing approaches simultaneously and suggests harmonization between scanners for multisite analyses should be considered before inference of group differences on subjects.

Bidirectional and state-dependent modulation of brain activity by transcranial focused ultrasound in non-human primates.

Transcranial focused ultrasound (FUS) stimulation under MRI guidance, coupled with functional MRI (fMRI) monitoring of effects, offers a precise, noninvasive technology to dissect functional brain circuits and to modulate altered brain functional networks in neurological and psychiatric disorders. Here we show that ultrasound at moderate intensities modulated neural activity bi-directionally. Concurrent sonication of somatosensory areas 3a/3b with 250 kHz FUS suppressed the fMRI signals produced there by peripheral tactile stimulation, while at the same time eliciting fMRI activation at inter-connected, off-target brain regions. Direct FUS stimulation of the cortex resulted in different degrees of BOLD signal changes across all five off-target regions, indicating that its modulatory effects on active and resting neurons differed. This is the first demonstration of the dual suppressive and excitative modulations of FUS on a specific functional circuit and of ability of concurrent FUS and MRI to evaluate causal interactions between functional circuits with neuron-class selectivity.

Concomitant modulation of BOLD responses in white matter pathways and cortex.

In response to a flickering visual stimulus, the BOLD response in primary visual cortex varies with the flickering frequency and is maximal when it is close to 8Hz. In previous studies we demonstrated that BOLD signals in specific white matter (WM) pathways covary with the alternations between stimulus conditions in a block design in similar manner to gray matter (GM) regions. Here we investigated whether WM tracts show varying responses to changes in flicker frequency and are modulated in the same manner as cortical areas. We used a Fourier analysis of BOLD signals to measure the signal amplitude and phase at the fundamental frequency of a block-design task in which flickering visual stimuli alternated with blank presentations, avoiding the assumption of any specific hemodynamic response function. The BOLD responses in WM pathways and the primary visual cortex were evaluated for flicker frequencies varying between 2 and 14Hz. The variations with frequency of BOLD signals in specific WM tracts followed closely those in primary visual cortex, suggesting that variations in cortical activation are directly coupled to corresponding BOLD signals in connected WM tracts. Statistically significant differences in the timings of BOLD responses were also measured between visual cortex and specific WM bundles. These results confirm that when cortical BOLD responses are modulated by selecting different task parameters, relevant WM tracts exhibit corresponding BOLD signals that are also affected.

Thickness of Deep Layers in the Fusiform Face Area Predicts Face Recognition.

People with superior face recognition have relatively thin cortex in face-selective brain areas, whereas those with superior vehicle recognition have relatively thick cortex in the same areas. We suggest that these opposite correlations reflect distinct mechanisms influencing cortical thickness (CT) as abilities are acquired at different points in development. We explore a new prediction regarding the specificity of these effects through the depth of the cortex: that face recognition selectively and negatively correlates with thickness of the deepest laminar subdivision in face-selective areas. With ultrahigh resolution MRI at 7T, we estimated the thickness of three laminar subdivisions, which we term "MR layers," in the right fusiform face area (FFA) in 14 adult male humans. Face recognition was negatively associated with the thickness of deep MR layers, whereas vehicle recognition was positively related to the thickness of all layers. Regression model comparisons provided overwhelming support for a model specifying that the magnitude of the association between face recognition and CT differs across MR layers (deep vs. superficial/middle) whereas the magnitude of the association between vehicle recognition and CT is invariant across layers. The total CT of right FFA accounted for 69% of the variance in face recognition, and thickness of the deep layer alone accounted for 84% of this variance. Our findings demonstrate the functional validity of MR laminar estimates in FFA. Studying the structural basis of individual differences for multiple abilities in the same cortical area can reveal effects of distinct mechanisms that are not apparent when studying average variation or development.

Tractography reproducibility challenge with empirical data (TraCED): The 2017 ISMRM diffusion study group challenge.

BACKGROUND: Fiber tracking with diffusion-weighted MRI has become an essential tool for estimating in vivo brain white matter architecture. Fiber tracking results are sensitive to the choice of processing method and tracking criteria. PURPOSE: To assess the variability for an algorithm in group studies reproducibility is of critical context. However, reproducibility does not assess the validity of the brain connections. Phantom studies provide concrete quantitative comparisons of methods relative to absolute ground truths, yet do no capture variabilities because of in vivo physiological factors. The ISMRM 2017 TraCED challenge was created to fulfill the gap. STUDY TYPE: A systematic review of algorithms and tract reproducibility studies. SUBJECTS: Single healthy volunteers. FIELD STRENGTH/SEQUENCE: 3.0T, two different scanners by the same manufacturer. The multishell acquisition included b-values of 1000, 2000, and 3000 s/mm2 with 20, 45, and 64 diffusion gradient directions per shell, respectively. ASSESSMENT: Nine international groups submitted 46 tractography algorithm entries each consisting 16 tracts per scan. The algorithms were assessed using intraclass correlation (ICC) and the Dice similarity measure. STATISTICAL TESTS: Containment analysis was performed to assess if the submitted algorithms had containment within tracts of larger volume submissions. This also serves the purpose to detect if spurious submissions had been made. RESULTS: The top five submissions had high ICC and Dice >0.88. Reproducibility was high within the top five submissions when assessed across sessions or across scanners: 0.87-0.97. Containment analysis shows that the top five submissions are contained within larger volume submissions. From the total of 16 tracts as an outcome relatively the number of tracts with high, moderate, and low reproducibility were 8, 4, and 4. DATA CONCLUSION: The different methods clearly result in fundamentally different tract structures at the more conservative specificity choices. Data and challenge infrastructure remain available for continued analysis and provide a platform for comparison. LEVEL OF EVIDENCE: 5 Technical Efficacy Stage: 1 J. Magn. Reson. Imaging 2020;51:234-249.

Generation of human thalamus atlases from 7 T data and application to intrathalamic nuclei segmentation in clinical 3 T T1-weighted images.

The thalamus serves as the central relay station for the brain. It processes and relays sensory and motor signals between different subcortical regions and the cerebral cortex and it can be divided into several neuronal clusters referred to as nuclei. Each of these can possibly be subdivided into sub-nuclei. Accurate and reliable identification of thalamic nuclei is important for surgical interventions and neuroanatomical studies. This is however a challenging task because the small size of the nuclei and the lack of contrast over the thalamus region in clinically acquired images does not permit the visualization of their boundaries. A number of methods have been developed for thalamus parcellation but the vast majority of these relies on diffusion imaging or functional imaging. The low resolution of these images only permit localizing the largest nuclei. In this work we propose a method to segment smaller nuclei. We first present a protocol to build histological-like atlases from a series of high-field (7 Tesla) MR images acquired with different pulse sequences that each permits to visualize the boundaries of a subset of the nuclei. We use this protocol to scan 9 subjects and we manually delineate 23 thalamic nuclei following the Morel atlas naming convention for each of these subjects. Manual contours for the nuclei are subsequently utilized to create statistical shape models. With these data, we compare four methods for the segmentation of thalamic nuclei in 3 T images we have also acquired for the 9 subjects included in the study: (1) single atlas, (2) multi atlas, (3) statistical shape, and (4) hierarchical statistical shape in which thalamic nuclei are hierarchically fitted to the images, starting from the largest ones. Results of a leave-one-out validation study conducted on the nine image sets we have acquired show that the multi atlas approach improves upon the single atlas approach for most nuclei. Segmentations obtained with the hierarchical statistical shape model yield the highest accuracy, with dice coefficients ranging from 0.53 to 0.90, mean surface errors from 0.27 mm to 0.64 mm, and maximum surface errors from 1.31 mm to 2.52 mm for all nuclei averaged across test cases. This suggests the feasibility of using such approach for localizing thalamic substructures in clinically acquired MR volumes. It may have a direct impact on surgeries such as Deep Brain Stimulation procedures that require the implantation of stimulating electrodes in specific thalamic nuclei.

Functional MRI and resting state connectivity in white matter - a mini-review.

Functional MRI (fMRI) signals are robustly detectable in white matter (WM) but they have been largely ignored in the fMRI literature. Their nature, interpretation, and relevance as potential indicators of brain function remain under explored and even controversial. Blood oxygenation level dependent (BOLD) contrast has for over 25 years been exploited for detecting localized neural activity in the cortex using fMRI. While BOLD signals have been reliably detected in grey matter (GM) in a very large number of studies, such signals have rarely been reported from WM. However, it is clear from our own and other studies that although BOLD effects are weaker in WM, using appropriate detection and analysis methods they are robustly detectable both in response to stimuli and in a resting state. BOLD fluctuations in a resting state exhibit similar temporal and spectral profiles in both GM and WM, and their relative low frequency (0.01-0.1 Hz) signal powers are comparable. They also vary with baseline neural activity e.g. as induced by different levels of anesthesia, and alter in response to a stimulus. In previous work we reported that BOLD signals in WM in a resting state exhibit anisotropic temporal correlations with neighboring voxels. On the basis of these findings, we derived functional correlation tensors that quantify the correlational anisotropy in WM BOLD signals. We found that, along many WM tracts, the directional preferences of these functional correlation tensors in a resting state are grossly consistent with those revealed by diffusion tensors, and that external stimuli tend to enhance visualization of specific and relevant fiber pathways. These findings support the proposition that variations in WM BOLD signals represent tract-specific responses to neural activity. We have more recently shown that sensory stimulations induce explicit BOLD responses along parts of the projection fiber pathways, and that task-related BOLD changes in WM occur synchronously with the temporal pattern of stimuli. WM tracts also show a transient signal response following short stimuli analogous to but different from the hemodynamic response function (HRF) characteristic of GM. Thus there is converging and compelling evidence that WM exhibits both resting state fluctuations and stimulus-evoked BOLD signals very similar (albeit weaker) to those in GM. A number of studies from other laboratories have also reported reliable observations of WM activations. Detection of BOLD signals in WM has been enhanced by using specialized tasks or modified data analysis methods. In this mini-review we report summaries of some of our recent studies that provide evidence that BOLD signals in WM are related to brain functional activity and deserve greater attention by the neuroimaging community.

Figure-Ground Modulation in the Human Lateral Geniculate Nucleus Is Distinguishable from Top-Down Attention.

Nearly all of the information that reaches the primary visual cortex (V1) of the brain passes from the retina through the lateral geniculate nucleus (LGN) of the thalamus. Although the LGN's role in relaying feedforward signals from the retina to the cortex is well understood [1, 2], the functional role of the extensive feedback it receives from the cortex has remained elusive [3-6]. Here, we investigated whether corticothalamic feedback may contribute to perceptual processing in the LGN in a manner that is distinct from top-down effects of attention [7-10]. We used high-resolution fMRI at 7 Tesla to simultaneously measure responses to orientation-defined figures in the human LGN and V1. We found robust enhancement of perceptual figures throughout the early visual system, which could be distinguished from the effects of covert spatial attention [11-13]. In a second experiment, we demonstrated that figure enhancement occurred in the LGN even when the figure and surrounding background were presented dichoptically (i.e., to different eyes). As binocular integration primarily occurs in V1 [14, 15], these results implicate a mechanism of automatic, contextually sensitive feedback from binocular visual cortex underlying figure-ground modulation in the LGN. Our findings elucidate the functional mechanisms of this core function of the visual system [16-18], which allows people to segment and detect meaningful figures in complex visual environments. The involvement of the LGN in this rich, contextually informed visual processing-despite showing minimal feedforward selectivity for visual features [19, 20]-underscores the role of recurrent processing at the earliest stages of visual processing.

Characterization of the hemodynamic response function in white matter tracts for event-related fMRI.

Accurate estimates of the BOLD hemodynamic response function (HRF) are crucial for the interpretation and analysis of event-related functional MRI data. To date, however, there have been no comprehensive measurements of the HRF in white matter (WM) despite increasing evidence that BOLD signals in WM change after a stimulus. We performed an event-related cognitive task (Stroop color-word interference) to measure the HRF in selected human WM pathways. The task was chosen in order to produce robust, distributed centers of activity throughout the cortex. To measure the HRF in WM, fiber tracts were reconstructed between each pair of activated cortical areas. We observed clear task-specific HRFs with reduced magnitudes, delayed onsets and prolonged initial dips in WM tracts compared with activated grey matter, thus calling for significant changes to current standard models for accurately characterizing the HRFs in WM and for modifications of standard methods of analysis of functional imaging data.

Inter-Scanner Harmonization of High Angular Resolution DW-MRI using Null Space Deep Learning.

Diffusion-weighted magnetic resonance imaging (DW-MRI) allows for non-invasive imaging of the local fiber architecture of the human brain at a millimetric scale. Multiple classical approaches have been proposed to detect both single (e.g., tensors) and multiple (e.g., constrained spherical deconvolution, CSD) fiber population orientations per voxel. However, existing techniques generally exhibit low reproducibility across MRI scanners. Herein, we propose a data-driven technique using a neural network design which exploits two categories of data. First, training data were acquired on three squirrel monkey brains using ex-vivo DW-MRI and histology of the brain. Second, repeated scans of human subjects were acquired on two different scanners to augment the learning of the network proposed. To use these data, we propose a new network architecture, the null space deep network (NSDN), to simultaneously learn on traditional observed/truth pairs (e.g., MRI-histology voxels) along with repeated observations without a known truth (e.g., scan-rescan MRI). The NSDN was tested on twenty percent of the histology voxels that were kept completely blind to the network. NSDN significantly improved absolute performance relative to histology by 3.87% over CSD and 1.42% over a recently proposed deep neural network approach. Moreover, it improved reproducibility on the paired data by 21.19% over CSD and 10.09% over a recently proposed deep approach. Finally, NSDN improved generalizability of the model to a third in vivo human scanner (which was not used in training) by 16.08% over CSD and 10.41% over a recently proposed deep learning approach. This work suggests that data-driven approaches for local fiber reconstruction are more reproducible, informative and precise and offers a novel, practical method for determining these models.

Phantom-based field maps for gradient nonlinearity correction in diffusion imaging.

Gradient coils in magnetic resonance imaging do not produce perfectly linear gradient fields. For diffusion imaging, the field nonlinearities cause the amplitude and direction of the applied diffusion gradients to vary over the field of view. This leads to site- and scan-specific systematic errors in estimated diffusion parameters such as diffusivity and anisotropy, reducing reliability especially in studies that take place over multiple sites. These errors can be substantially reduced if the actual scanner-specific gradient coil magnetic fields are known. The nonlinearity of the coil fields is measured by scanner manufacturers and used internally for geometric corrections, but obtaining and using the information for a specific scanner may be impractical for many sites that operate without special-purpose local engineering and research support. We have implemented an empirical field-mapping procedure using a large phantom combined with a solid harmonic approximation to the coil fields that is simple to perform and apply. Here we describe the accuracy and precision of the approach in reproducing manufacturer gold standard field maps and in reducing spatially varying errors in quantitative diffusion imaging for a specific scanner. Before correction, median B value error ranged from 33 - 41 relative to manufacturer specification at 100 mm from isocenter; correction reduced this to 0 - 4. On-axis spatial variation in the estimated mean diffusivity of an isotropic phantom was 2.2% - 4.1% within 60 mm of isocenter before correction, 0.5% - 1.6% after. Expected fractional anisotropy in the phantom was 0; highest estimated fractional anisotropy within 60 mm of isocenter was reduced from 0.024 to 0.012 in the phase encoding direction (48% reduction) and from 0.020 to 0.006 in the frequency encoding direction (72% reduction).

Learning Implicit Brain MRI Manifolds with Deep Learning.

An important task in image processing and neuroimaging is to extract quantitative information from the acquired images in order to make observations about the presence of disease or markers of development in populations. Having a low-dimensional manifold of an image allows for easier statistical comparisons between groups and the synthesis of group representatives. Previous studies have sought to identify the best mapping of brain MRI to a low-dimensional manifold, but have been limited by assumptions of explicit similarity measures. In this work, we use deep learning techniques to investigate implicit manifolds of normal brains and generate new, high-quality images. We explore implicit manifolds by addressing the problems of image synthesis and image denoising as important tools in manifold learning. First, we propose the unsupervised synthesis of T1-weighted brain MRI using a Generative Adversarial Network (GAN) by learning from 528 examples of 2D axial slices of brain MRI. Synthesized images were first shown to be unique by performing a cross-correlation with the training set. Real and synthesized images were then assessed in a blinded manner by two imaging experts providing an image quality score of 1-5. The quality score of the synthetic image showed substantial overlap with that of the real images. Moreover, we use an autoencoder with skip connections for image denoising, showing that the proposed method results in higher PSNR than FSL SUSAN after denoising. This work shows the power of artificial networks to synthesize realistic imaging data, which can be used to improve image processing techniques and provide a quantitative framework to structural changes in the brain.

SHARD: Spherical Harmonic-based Robust Outlier Detection for HARDI Methods.

High Angular Resolution Diffusion Imaging (HARDI) models are used to capture complex intra-voxel microarchitectures. The magnetic resonance imaging sequences that are sensitized to diffusion are often highly accelerated and prone to motion, physiologic, and imaging artifacts. In diffusion tensor imaging, robust statistical approaches have been shown to greatly reduce these adverse factors without human intervention. Similar approaches would be possible with HARDI methods, but robust versions of each distinct HARDI approach would be necessary. To avoid the computational and pragmatic burdens of creating individual robust HARDI analysis variants, we propose a robust outlier imputation model to mitigate outliers prior to traditional HARDI analysis. This model uses a weighted spherical harmonic fit of diffusion weighted magnetic resonance imaging scans to estimate the values which had been corrupted during acquisition to restore them. Briefly, spherical harmonics of 6th order were used to generate basis function which were weighted by diffusion signal for detection of outliers. For validation, a single healthy volunteer was scanned for a single session comprising of two scans one without head movement and the other with deliberate head movement at a b-value of 3000 s/mm2 with 64 diffusion weighted directions with a single b0 (5 averages) per scan. The deliberate motion from the volunteer created natural artifacts in the acquisition of one of the scans. The imputation model shows reduction in root mean squared error of the raw signal intensities and improvement for the HARDI method Q-ball in terms of the Angular Correlation Coefficient. The results reveal that there is quantitative and qualitative improvement. The proposed model can be used as general pre-processing model before implementing any HARDI model in general to restore the artifacts which are created because of the outlier diffusion signal in certain gradient volumes.

Towards Portable Large-Scale Image Processing with High-Performance Computing.

High-throughput, large-scale medical image computing demands tight integration of high-performance computing (HPC) infrastructure for data storage, job distribution, and image processing. The Vanderbilt University Institute for Imaging Science (VUIIS) Center for Computational Imaging (CCI) has constructed a large-scale image storage and processing infrastructure that is composed of (1) a large-scale image database using the eXtensible Neuroimaging Archive Toolkit (XNAT), (2) a content-aware job scheduling platform using the Distributed Automation for XNAT pipeline automation tool (DAX), and (3) a wide variety of encapsulated image processing pipelines called "spiders." The VUIIS CCI medical image data storage and processing infrastructure have housed and processed nearly half-million medical image volumes with Vanderbilt Advanced Computing Center for Research and Education (ACCRE), which is the HPC facility at the Vanderbilt University. The initial deployment was natively deployed (i.e., direct installations on a bare-metal server) within the ACCRE hardware and software environments, which lead to issues of portability and sustainability. First, it could be laborious to deploy the entire VUIIS CCI medical image data storage and processing infrastructure to another HPC center with varying hardware infrastructure, library availability, and software permission policies. Second, the spiders were not developed in an isolated manner, which has led to software dependency issues during system upgrades or remote software installation. To address such issues, herein, we describe recent innovations using containerization techniques with XNAT/DAX which are used to isolate the VUIIS CCI medical image data storage and processing infrastructure from the underlying hardware and software environments. The newly presented XNAT/DAX solution has the following new features: (1) multi-level portability from system level to the application level, (2) flexible and dynamic software development and expansion, and (3) scalable spider deployment compatible with HPC clusters and local workstations.