Search

Deep neural networks show great promise for classifying brain diseases and making prognostic assessments based on neuroimaging data, but large, labeled training datasets are often required to achieve high predictive accuracy. Here we evaluated a range oftransfer learningor pre-training strategies to create useful MRI representations for downstream tasks that lack large amounts of training data, such as Alzheimer’s disease (AD) classification. To test our proposed pre-training strategies, we analyzed 4,098 3D T1-weighted brain MRI scans from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort and independently validated with an out-of-distribution test set of 600 scans from the Open Access Series of Imaging Studies (OASIS3) cohort for detecting AD. First, we trained 3D and 2D convolutional neural network (CNN) architectures. We tested combinations of multiple pre-training strategies based on (1) supervised, (2) contrastive learning, and (3) self-supervised learning - using pre-training data within versus outside the MRI domain. In our experiments, the 3D CNN pre-trained with contrastive learning provided the best overall results - when fine-tuned on T1-weighted scans for AD classification - outperformed the baseline by 2.8% when trained with all of the training data from ADNI. We also show test performance as a function of the training dataset size and the chosen pre-training method. Transfer learning offered significant benefits in low data regimes, with a performance boost of 7.7%. When the pre-trained model was used for AD classification, we were able to visualize an improved clustering of test subjects’ diagnostic groups, as illustrated via a uniform manifold approximation (UMAP) projection of the high-dimensional model embedding space. Further, saliency maps indicate the additional activation regions in the brain scan using pre-training, that then maximally contributed towards the final prediction score.

Parkinson9s disease (PD) and Alzheimer9s disease (AD) are progressive neurodegenerative disorders that affect millions of people worldwide. In this work, we propose a deep learning approach to classify these diseases based on 3D T1-weighted brain MRI. We analyzed several datasets including the Parkinson9s Progression Markers Initiative (PPMI), an independent dataset from the University of Pennsylvania School of Medicine (UPenn), the Alzheimer9s Disease Neuroimaging Initiative (ADNI), and the Open Access Series of Imaging Studies (OASIS) dataset. PPMI and ADNI were partitioned to train (70%), validate (20%), and test (10%) a 3D convolutional neural network (CNN) for PD and AD classification. The UPenn and OASIS datasets were used as independent test sets to evaluate the model performance during inference. We also implemented a random forest classifier as a baseline model by extracting key radiomics features from the same T1-weighted MRI scans. The proposed 3D CNN model was trained from scratch for the classification tasks. For AD classification, the 3D CNN model achieved an ROC-AUC of 0.878 on the ADNI test set and an average ROC-AUC of 0.789 on the OASIS dataset. For PD classification, the proposed 3D CNN model achieved an ROC-AUC of 0.667 on the PPMI test set and an average ROC-AUC of 0.743 on the UPenn dataset. We also found that model performance was largely maintained when using only 25% of the training dataset. The 3D CNN outperformed the random forest classifier for both the PD and AD tasks. The 3D CNN also generalized better on unseen MRI data from different imaging centers. Our results show that the proposed 3D CNN model was less prone to overfitting for AD than for PD classification. This approach shows promise for screening of PD and AD patients using only T1-weighted brain MRI, which is relatively widely available. This model with additional validation could also be used to help differentiate between challenging cases of AD and PD when they present with similarly subtle motor and non-motor symptoms.

Diffusion-weighted magnetic resonance imaging (dMRI) is sensitive to microstructural changes in the brain that occur with normal aging and Alzheimer’s disease (AD). There is much interest in which dMRI measures are most strongly correlated with clinical measures of AD severity, such as the clinical dementia rating (CDR), and biological processes that may be disrupted in AD, such as brain amyloid load measured using PET. Of these processes, some can be targeted using novel drugs. Since 2016, the Alzheimer’s Disease Neuroimaging Initiative (ADNI) has collected dMRI data from three scanner manufacturers across 58 sites using 7 different protocols that vary in angular resolution, scan duration, and in the number and distribution of diffusion-weighted gradients. Here, we assessed dMRI data from 730 of those individuals (447 cognitively normal controls, 214 with mild cognitive impairment, 69 with dementia; age: 74.1±7.9 years; 381 female/349 male). To harmonize data from different protocols, we applied ComBat, ComBat-GAM, and CovBat to dMRI metrics from 28 white matter regions of interest. We ranked all dMRI metrics in order of the strength of clinically relevant associations, and assessed how this depended on the harmonization methods employed. dMRI metrics were associated with age and clinical impairment, but also with amyloid positivity. All harmonization methods gave comparable results while enabling data integration across multiple scanners and protocols.

The brain’s white matter microstructure, as assessed using diffusion-weighted MRI (DWI), changes significantly with age and also exhibits significant sex differences. Here we examined the ability of a traditional diffusivity metric (fractional anisotropy derived from diffusion tensor imaging, DTI-FA) and advanced diffusivity metrics (fractional anisotropy derived from the tensor distribution function, TDF-FA; neurite orientation dispersion and density imaging measures of intracellular volume fraction, NODDI-ICVF; orientation dispersion index, NODDI-ODI; and isotropic volume fraction, NODDI-ISOVF) to detect sex differences in white matter aging. We also created normative aging reference curves based on sex. Diffusion tensor imaging (DTI) applies a single-tensor diffusion model to single-shell DWI data, while the tensor distribution function (TDF) fits a continuous distribution of tensors to single-shell DWI data. Neurite orientation dispersion and density imaging (NODDI) fits a multi-compartment model to multi-shell DWI data to distinguish intra- and extracellular contributions to diffusion. We analyzed these traditional and advanced diffusion measures in a large population sample available through the UK Biobank (15,394 participants; age-range: 45-80 years) by using linear regression and fractional polynomials. Advanced diffusivity metrics (NODDI-ODI, NODDI-ISOVF, TDF-FA) detected significant sex differences in aging, whereas a traditional metric (DTI-FA) did not. These findings suggest that future studies examining sex differences in white matter aging may benefit from including advanced diffusion measures.

Diffusion-weighted MRI (dMRI) offers a range of measures that are sensitive to brain aging and neurodegeneration. Here we analyzed data from 318 participants (mean age: 75.4±7.9 years; 143 men/175 women) from the third phase of the Alzheimer’s Disease Neuroimaging Initiative (ADNI3), who were each scanned with one of six different diffusion MRI protocols using scanners from three different manufacturers. We computed 4 standard diffusion tensor imaging (DTI) anisotropy and diffusivity indices, and one advanced anisotropy index based on the tensor distribution function (TDF), in 24 white matter regions of interest. Modeling protocol effects, we ranked the diffusion indices for their strength of correlation with 3 standard clinical measures of cognitive impairment: the ADAS-Cog, MMSE, and sum-of-boxes Clinical Dementia Rating. Across all dMRI indices and cognitive measures, the cingulum-hippocampal region and the uncinate showed some of the strongest associations with cognitive impairment; largest effect sizes were detected with axial diffusivity (AxDDTI). While fractional anisotropy (FA) derived from the DTI model was the weakest in detecting associations with cognitive measures, FA derived from the TDF detected widespread, robust associations. Protocol differences affected dMRI indices; however by modeling protocol effects, we were able to pool dMRI data from multiple acquisition protocols and detect consistent associations with cognitive impairment and age. dMRI indices computed from the upgraded scanning protocols in ADNI3 were sensitive to cognitive impairment in brain aging, offering a benchmark to compare to future multi-shell or multi-compartment diffusion indices.

Alzheimer’s disease is heterogeneous and despite some consistent neuropathological hallmarks, different clinical forms have been identified, including non-amnestic presentations. Even in amnestic forms, the presentation of the disease can differ across individuals, in terms of age of onset, dynamics of progression and specific impairment profiles. Different distributions of neurofibrillary tangles exist in AD, and these are linked with structural differences detectable on ante-mortem MRI , but these are hard to identify in the earlier stages of disease. In the present work, we validate and test a previously proposed method for identifying subtypes of cortical atrophy in AD, based on MRI data from an independent case/control study of individuals defined by pathophysiological biomarkers. We implemented a clustering method based on the Louvain modularity method, and tested it across a range of pre-processing parameters. Our cohort of participants was comprised of 111 participants (mean age: 67.7 year; range: 51-91), including 37 cognitively normal controls, 43 prodromal AD, and 31 demented AD patients. We identified 4 patient clusters with distinct atrophy patterns either predominantly in the temporal lobes (groups 0 and 1), in the parietal and temporal lobes (group 2), or in the frontal and temporal lobes (group 3). Further evaluation of neuro-psychological characteristics of each patient cluster will be carried out in the future. In conclusion, the modularity-based clustering method may help to identify specific subtypes of atrophy in neurological diseases such as AD.

Traumatic brain injury (TBI) is a major public health issue around the world. Pediatric TBI patients are at risk of long-term disabilities, as a brain injury sustained during development can affect on-going maturational processes. The white matter (WM) in particular is vulnerable, as myelination continues into the third decade of life and beyond, and poor myelination of tracts can result in decreased integration within brain networks. In addition, variability and heterogeneity are hallmarks of TBI, e.g., injury-related variables and symptoms. These issues combined with small sample sizes limit the power and generalizability of individual studies. In the present study, we employed a meta-analytic approach, combining data across 4 pediatric TBI samples resulting in 104 TBI (75M/29F) and 114 control participants (70M/44F) between 7-18 years, using harmonized processing and analysis as part of the ENIGMA consortium (Enhancing NeuroImaging Genetics through Meta-Analysis). We report lower fractional anisotropy (FA) values in TBI patients across several post-injury windows, particularly in central WM tracts. Within the TBI patient group, we also report marginally significant results of lower FA in younger TBI patients, patients scanned closer to time of injury, and female patients. Although this meta-analytic approach yielded the largest sample size reported yet in pediatric moderate/severe TBI (msTBI) neuroimaging, our trends indicate that larger sample sizes are needed in further studies. As additional cohorts join the ENIGMA Pediatric moderate/severe TBI (msTBI) effort, more robust effects will be revealed.

Diffusion MRI (dMRI) is widely used to study the brain’s white matter (WM) microstructure in a range of psychiatric and neurological diseases. As the diffusion tensor model has limitations in brain regions with crossing fibers, novel diffusion MRI reconstruction models may offer more accurate measures of tissue properties, and a better understanding of the brain abnormalities in specific diseases. Here we studied a large sample of 249 participants with 22q11.2 deletion syndrome (22q11DS), a neurogenetic condition associated with high rates of developmental neuropsychiatric disorders, and 224 age-matched healthy controls (HC) (age range: 8-35 years). Participants were scanned with dMRI at eight centers worldwide. Using a meta-analytic approach, we assessed the profile of group differences in four diffusion anisotropy measures to better understand the patterns of WM microstructural abnormalities and evaluate their consistency across alternative measures. When assessed in atlas-defined regions of interest, we found statistically significant differences for all anisotropy measures, all showing a widespread but not always coinciding pattern of effects. The tensor distribution function fractional anisotropy (TDF-FA) showed largest effect sizes all in the same direction (greater anisotropy in 22q11DS than HC). Fractional anisotropy based on the tensor model (FA) showed the second largest effect sizes after TDF-FA; some regions showed higher mean values in 22q11DS, but others lower. Generalized fractional anisotropy (GFA) showed the opposite pattern to TDF-FA with most regions showing lower anisotropy in 22q11DS versus HC. Anisotropic power maps (AP) showed the lowest effect sizes also with a mixed pattern of effects across regions. These results were also consistent across skeleton projection methods, with few differences when projecting anisotropy values from voxels sampled on the FA map or projecting values from voxels sampled from each anisotropy map. This study highlights that different mathematical definitions of anisotropy may lead to different profiles of group differences, even in large, well-powered population studies. Further studies of biophysical models derived from multi-shell dMRI and histological validations may help to understand the sources of these differences. 22q11DS is a promising model to study differences among novel anisotropy/dMRI measures, as group differences are relatively large and there exist animal models suitable for histological validation.

Polygenic risk scores (PRSs) may be used to investigate the effects of genetic risk for disease on complex human traits. Here we set out to determine how the overall genetic risk for Alzheimer’s disease (AD) shapes brain structure in non-AD populations. PRS scores were computed using results from the International Genomics of Alzheimer's Project (IGAP) study. PRSs were computed at 14 different significance thresholds in the IGAP results. The effect of PRS as a predictor of brain morphometry was mapped voxelwise on brain structure as determined by tensor-based morphometry (TBM) in three cohorts: ADNI1, ADNI2, and HUNT. Our multi-cohort TBM framework first tests associations in each cohort individually, then meta-analyzes findings in a common space. Higher PRS for AD was associated with greater ventricular and lower hippocampal volumes. Associations remained after removing the major AD risk gene, APOE, from the PRS. This cumulative influence of common genetic variants on brain-wide structural variation in nondemented individuals may pinpoint genetic and neurological pathways that contribute to the preclinical assessment of disease risk.

Outcome after a traumatic brain injury (TBI) is quite variable, and this variability is not solely accounted for by severity or demographics. Identifying sub-groups of patients who recover faster or more fully will help researchers and clinicians understand sources of this variability, and hopefully lead to new therapies for patients with a more prolonged recovery profile. We have previously identified two subgroups within the pediatric TBI patient population with different recovery profiles based on an ERP-derived (event-related potential) measure of interhemispheric transfer time (IHTT). Here we examine structural network topology across both patient groups and healthy controls, focusing on the ‘rich-club’ - the core of the network, marked by high degree nodes. These analyses were done at two points post-injury - 2-5 months (post-acute), and 13-19 months (chronic). In the post-acute time-point, we found that the TBI-slow group, those showing longitudinal degeneration, showed hyperconnectivity within the rich-club nodes relative to the healthy controls, at the expense of local connectivity. There were minimal differences between the healthy controls and the TBI-normal group (those patients who show signs of recovery). At the chronic phase, these disruptions were no longer significant, but closer analysis showed that this was likely due to the loss of power from a smaller sample size at the chronic time-point, rather than a sign of recovery. We have previously shown disruptions to white matter (WM) integrity that persist and progress over time in the TBI-slow group, and here we again find differences in the TBI-slow group that fail to resolve over the first year post-injury.

Mild traumatic brain injury (mTBI) is characterized clinically by a closed head injury involving differential or rotational movement of the brain inside the skull. Over 3 million mTBIs occur annually in the United States alone. Many of the individuals who sustain an mTBI go on to recover fully, but around 20% experience persistent symptoms. These symptoms often last for many weeks to several months. The thalamus, a structure known to serve as a global networking or relay system for the rest of the brain, may play a critical role in neurorehabiliation and its integrity and connectivity after injury may also affect cognitive outcomes. To examine the thalamus, conventional tractography methods to map corticothalamic pathways with diffusion-weighted MRI (DWI) lead to sparse reconstructions that may contain false positive fibers that are anatomically inaccurate. Using a specialized method to zero in on corticothalamic pathways with greater robustness, we noninvasively examined corticothalamic fiber projections using DWI, in 68 service members. We found significantly lower fractional anisotropy (FA), a measure of white matter microstructural integrity, in pathways projecting to the left pre- and postcentral gyri – consistent with sensorimotor deficits often found post-mTBI. Mapping of neural circuitry in mTBI may help to further our understanding of mechanisms underlying recovery post-TBI.

State-of-the-art data analysis methods in genetics and related fields have advanced beyond massively univariate analyses. However, these methods suffer from the limited amount of data available at a single research site. Recent large-scale multi-centric imaging-genetic studies, such as ENIGMA, have to rely on meta-analysis of mass univariate models to achieve critical sample sizes for uncovering statistically significant associations. Indeed, model parameters, but not data, can be securely and anonymously shared between partners. We propose here partial least squares (PLS) as a multivariate imaging-genetics model in meta-studies. In particular, we propose an online estimation approach to partial least squares for the sequential estimation of the model parameters in data batches, based on an approximation of the singular value decomposition (SVD) of partitioned covariance matrices. We applied the proposed approach to the challenging problem of modeling the association between 1,167,117 genetic markers (SNPs, single nucleotide polymorphisms) and the brain cortical and sub-cortical atrophy (354,804 anatomical surface features) in a cohort of 639 individuals from the Alzheimer's Disease Neuroimaging Initiative. We compared two different modeling strategies (sequential- and meta-PLS) to the classic non-distributed PLS. Both strategies exhibited only minimal approximation errors of model parameters. The proposed approaches pave the way to the application of multivariate models in large scale imaging-genetics meta-studies, and may lead to novel understandings of the complex brain phenotype-genotype interactions.

We present an algorithm to align cortical surface models based on structural connectivity. We follow the continuous connectivity approach,1, 2 assigning a dense connectivity to every surface point-pair. We adapt and modify an approach for aligning low-rank functional networks based on eigenvalue decomposition of individual connectomes.3 The spherical demons framework then provides a natural setting for inter-subject connectivity alignment, enforcing a smooth, anatomically plausible correspondence, and allowing us to incorporate anatomical as well as connectivity information. We apply our algorithm to 98 diffusion MRI images in an Alzheimer's Disease study, and 731 healthy subjects from the Human Connectome Project. Our method consistently reduces connectome variability due to misalignment. Further, the approach reveals subtle disease effects on structural connectivity which are not seen when registering only cortical anatomy.

Fractional anisotropy derived from the single-tensor model (FADTI) in diffusion MRI (dMRI) is the most widely used metric to characterize white matter (WM) micro-architecture in disease, despite known limitations in regions with extensive fiber crossing. Models such as the tensor distribution function (TDF), which represents the diffusion profile as a probabilistic mixture of tensors, have been proposed to reconstruct multiple underlying fibers. Although complex HARDI acquisition protocols are rare in clinical studies, the TDF and TDF-derived scalar FA metric (FATDF) have been shown to be advantageous even for data with modest angular resolution. However, further evaluation and validation of the metric are necessary. Here we compared the test-retest reliability of FATDF and FADTI in clinical quality data by computing the intra-class correlation (ICC) between dMRI scans collected 3 months apart. When FATDF and FADTI were calculated at various angular resolutions, FATDF ICC in both the corpus callosum and in a full axial slice were consistently more stable across scans, as compared to FADTI.

Traumatic brain injury (TBI) causes extensive damage to the white matter (WM) of the brain, which can be evaluated with diffusion-weighted magnetic resonance imaging (dMRI). Diffusion MRI can be used to map the WM tracts and their integrity, but offers limited understanding of the biochemical basis of any differences. Magnetic resonance spectroscopy (MRS) measures neural metabolites that reflect neuronal health, inflammation, demyelination, and other consequences of TBI. We combined whole-brain MRS with dMRI to investigate WM dysfunction following pediatric TBI, using “tract-based spectroscopy”. Deficits in N-acetylaspartate (NAA) correspond to regions of deficits in WM integrity, but choline showed minimal overlap with WM deficits. NAA is a marker of neuronal health, while choline is an inflammatory marker. A partial F-test showed that MRS measures improved our ability to predict long-term cognitive function. This is the first paper to combine MRS with dMRI-derived tracts on a whole-brain scale, offering insights into the biochemical correlates of WM tract dysfunction, following injury and potentially in other WM disorders.

White matter tractography is non-invasive method to study white matter microstructure within the brain and its connectivity across the different regions. Various neuro-degenerative diseases affect the white matter connectivity in the brain. In order to study the neurodegeneration and localize the affected fiber bundles, it is important to cluster the white matter fibers in an anatomically consistent manner. Clustering white matter fiber bundles in the brain is a challenging problem. The present approaches include region of interest (ROI) based clustering as well as template based clustering. A novel clustering technique using support vector machine framework is introduced. In this method, a conformal volumetric bijective mapping between the brain and the topologically equivalent sphere is established. The white matter fibers are then parameterized in this domain. Such a parameterization also introduces a spatial normalization without requiring any prior registration. We show that such a mapping is useful to learn statistical models of white matter fiber bundles and use it for clustering in a new subject.

In this paper, a Bayesian super resolution (SR) method obtains high resolution (HR) brain Diffusion-Weighted Magnetic Resonance Imaging (DMRI) images from degraded low resolution (LR) images. Under a Bayesian formulation, the unknown HR image, the acquisition process and the unknown parameters are modeled as stochastic processes. The likelihood model is modeled using a Gaussian distribution to estimate the error between the a linear representation and the observations. The prior is introduced as a Multivariate Gaussian Distribution, for which the inverse of the covariance matrix is approximated by Laplacian-like functions that model the local relationships, capturing thereby non-homogeneous relationships between neighbor intensities. Experimental results show the method outperforms the base line by 2.56 dB when using PSNR as a metric of quality in a set of 35 cases.

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder with atypical cortical maturation leading to a deficiency in social cognition and language. Numerous studies have attempted to classify ASD using brain measurements such as cortical thickness, surface area, or volume with promising results. However, the underpowered sample sizes of these studies limit external validity and generalizability at the population level. Large scale collaborations such as Enhancing NeuroImaging Genetics through Meta Analysis (ENIGMA) or the Autism Brain Imaging Data Exchange (ABIDE) aim to bring together like-minded scientists to further improve investigations into brain disorders. To the best of our knowledge, this study represents the largest classification analysis for detection of ASD vs. healthy age and sex matched controls using cortical thickness brain parcellations and intracranial volume normalized surface area and subcortical volumes. We were able to increase classification accuracy overall from 56% to 60% and for females only by 6%. These novel findings using Evolving Partitions to Improve Connectomics (EPIC) underscore the importance of large-scale data-driven approaches and collaborations in the discovery of brain disorders.

Outcomes after traumatic brain injury (TBI) are variable and only partially predicted by acute injury factors. With rich datasets, we can examine how numerous factors – cognitive scores, acute injury variables, demographic variables, and brain imaging variables – are interrelated and aid in outcome prediction. To help study this rich data, we applied CorEx, a novel method for unsupervised machine learning. CorEx decodes the hierarchical structure, identifying latent causes of dependence in the data. It groups predictor variables based on their joint information and inter-dependence. We examined 21 TBI patients 2-5 months post-injury along with healthy controls; both groups were assessed again 12 months later. Although we were limited in the number of participants, this tool for exploratory analysis found potential relationships between change in cognitive scores over the 12-month period and baseline brain volumes. Certain regional brain volumes measured post-injury could serve as predictors of patient recovery. As future planned analyses will examine greater sample sizes, we hope to perform follow-up statistical analysis of variables identified by CorEx in independent data.

The quest to identify genetic factors that shape the human brain has been greatly accelerated by imaging. The human brain functions as a complex network of integrated systems and connected processes, and a vast number of features can be observed and extracted from structural brain images -- including regional volume, shape, and other morphological features of given brain structures. This feature set can be considered as part of the structural network of the brain, which is under strong genetic control. However, it is unclear which of the imaging derived features serve as the most promising traits for discovering specific genes that affect brain structure. Here, we aim to create the first ever network of genetically correlated cortical sulcal features, and through a twin model, determine the degree of genetic correlation across the entire network. Building on functional brain network analysis, we consider the high-dimensional genetic correlation structure as a undirected graph with a complex network of multi-weighted hubs to uncover the underlying genetic core of sulcal morphometry.

Neuroimaging consortia such as ENIGMA can significantly improve power to discover factors that affect the human brain by pooling statistical inferences across cohorts to draw generalized conclusions from populations around the world. Voxelwise analyses such as tensor-based morphometry also allow an unbiased search for effects throughout the brain. Even so, such consortium-based analyses are limited by a lack of high-powered methods to harmonize voxelwise information across study populations and scanners. While the simplest approach may be to map all images to a single standard space, the benefits of cohort-specific templates have long been established. Here we studied methods to pool voxel-wise data across sites using templates customized for each cohort but providing a meaningful common space across all studies for voxelwise comparisons. As non-linear 3D MRI registrations represent mappings between images at millimeter resolution, we need to consider the reliability of these mappings. To evaluate these mappings, we calculated test-retest statistics on the volumetric maps of expansion and contraction. Further, we created study-specific brain templates for ten T1-weighted MRI datasets, and a common space from four study-specific templates. We evaluated the efficacy of using a two-step registration framework versus a single standard space. We found that the two-step framework more reliably mapped subjects to a common space.

Nonlinear registration algorithms are routinely used in brain imaging, to align data for inter-subject and group comparisons, and for voxelwise statistical analyses. To understand how the choice of registration method affects maps of functional brain connectivity in a sample of 611 twins, we evaluated three popular nonlinear registration methods: Advanced Normalization Tools (ANTs), Automatic Registration Toolbox (ART), and FMRIB's Nonlinear Image Registration Tool (FNIRT). Using both structural and functional MRI, we used each of the three methods to align the MNI152 brain template, and 80 regions of interest (ROIs), to each subject's T1-weighted (T1w) anatomical image. We then transformed each subject's ROIs onto the associated resting state functional MRI (rs-fMRI) scans and computed a connectivity network or functional connectome for each subject. Given the different degrees of genetic similarity between pairs of monozygotic (MZ) and same-sex dizygotic (DZ) twins, we used structural equation modeling to estimate the additive genetic influences on the elements of the function networks, or their heritability. The functional connectome and derived statistics were relatively robust to nonlinear registration effects.

Cognitive decline accompanies many debilitating illnesses, including Alzheimer’s disease (AD). In old age, brain tissue loss also occurs along with cognitive decline. Although blood tests are easier to perform than brain MRI, few studies compare brain scans to standard blood tests to see which kinds of information best predict future decline. In 504 older adults from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), we first used linear regression to assess the relative value of different types of data to predict cognitive decline, including 196 blood panel biomarkers, 249 MRI biomarkers obtained from the FreeSurfer software, demographics, and the AD-risk gene APOE. A subset of MRI biomarkers was the strongest predictor. There was no specific blood marker that increased predictive accuracy on its own, we found that a novel unsupervised learning method, CorEx, captured weak correlations among blood markers, and the resulting clusters offered unique predictive power.

Sparse learning enables dimension reduction and efficient modeling of high dimensional signals and images, but it may need to be tailored to best suit specific applications and datasets. Here we used sparse learning to efficiently represent functional magnetic resonance imaging (fMRI) data from the human brain. We propose a novel embedded sparse representation (ESR), to identify the most consistent dictionary atoms across different brain datasets via an iterative group-wise dictionary optimization procedure. In this framework, we introduced additional criteria to make the learned dictionary atoms more consistent across different subjects. We successfully identified four common dictionary atoms that follow the external task stimuli with very high accuracy. After projecting the corresponding coefficient vectors back into the 3-D brain volume space, the spatial patterns are also consistent with traditional fMRI analysis results. Our framework reveals common features of brain activation in a population, as a new, efficient fMRI analysis method.

Deficits in white matter (WM) integrity occur following traumatic brain injury (TBI), and often persist long after the visible scars have healed. Heterogeneity in injury types and locations can complicate analyses, making it harder to discover common biomarkers for tracking recovery. Here we apply a newly developed adaptive connectivity method, EPIC (evolving partitions to improve connectomics) to identify differences in structural connectivity that persist longitudinally. This data comes from a longitudinal study, in which we scanned participants (aged 8-19 years) with anatomical and diffusion MRI in both the post-acute and chronic phases (1-6 months and 13-19 months post-injury). To identify patterns of abnormal connectivity, we trained a model on data from 32 TBI patients in the post-acute phase and 45 well-matched healthy controls, reducing an initial 68x68 connectivity matrix to a 14x14 matrix. We then applied this reduced parcellation to the chronic data in participants who had returned for their chronic assessment (21 TBI and 26 healthy controls) and tested for group differences. We found significant differences in two connections, comprising callosal fibers and long anterior-posterior fibers, with the TBI group showing increased fiber density relative to controls. Longitudinal analysis revealed that these were connections that were decreasing over time in the healthy controls, as is a common developmental phenomenon, but they were increasing in the TBI group. While we cannot definitively tell why this may occur with our current data, this study provides targets for longitudinal tracking, and poses questions for future investigation.

Diffusion weighted imaging (DWI) can reveal the orientation of the underlying fiber populations in the brain. High angular resolution diffusion imaging (HARDI) is increasingly used to better resolve the orientation and mixing of fibers. Here, we assessed the added value of multi-shell q-space sampling on the reconstruction of major fibers using mathematical frameworks from q-ball imaging (QBI) and generalized q-sampling imaging (GQI), as compared to diffusion tensor imaging (DTI). We scanned a healthy mouse brain using 7-Tesla 5-shell HARDI (b=1000, 3000, 4000, 8000, 12000 s/mm2), also known as hybrid diffusion imaging (HYDI). We found that QBI may provide greater reconstruction accuracy for major fibers, which improves with the addition of higher b-value shells, unlike GQI or DTI (as expected). Although QBI is a special case of GQI, the major fiber orientation in QBI was more closely related to the orientation in DTI, rather than GQI. HYDI can aid the clinical outcomes of research and especially – more advanced human and animal connectomics projects to map the brain’s neural pathways and networks.

In the present work we study a family of generative network model and its applications for modeling the human connectome. We introduce a minor but novel variant of the Mixed Membership Stochastic Blockmodel and apply it and two other related model to two human connectome datasets (ADNI and a Bipolar Disorder dataset) with both control and diseased subjects. We further provide a simple generative classifier that, alongside more discriminating methods, provides evidence that blockmodels accurately summarize tractography count networks with respect to a disease classification task.

The aim of this paper is to assess the heritability of cerebral cortex, based on measurements of grey matter (GM) thickness derived from structural MR images (sMRI). With data acquired from a large twin cohort (328 subjects), an automated method was used to estimate the cortical thickness, and EM-ICP surface registration algorithm was used to establish the correspondence of cortex across the population. An ACE model was then employed to compute the heritability of cortical thickness. Heritable cortical thickness measures various cortical regions, especially in frontal and parietal lobes, such as bilateral postcentral gyri, superior occipital gyri, superior parietal gyri, precuneus, the orbital part of the right frontal gyrus, right medial superior frontal gyrus, right middle occipital gyrus, right paracentral lobule, left precentral gyrus, and left dorsolateral superior frontal gyrus.

Banks of high-quality, multimodal neurological images offer new possibilities for analyses based on brain registration. To take full advantage of these, current algorithms should be significantly enhanced. We present here a new brain registration method driven simultaneously by the structural intensity and the total diffusion information of MRI scans. Using the two modalities together allows for a better alignment of general and specific aspects of the anatomy. Furthermore, keeping the full diffusion tensor in the cost function, rather than only some of its scalar measures, will allow for a thorough statistical analysis once the Jacobian of the transformation is obtained.

Tensor-Based Morphometry (TBM) is an increasingly popular method for group analysis of brain MRI data. The main steps in the analysis consist of a nonlinear registration to align each individual scan to a common space, and a subsequent statistical analysis to determine morphometric differences, or difference in fiber structure between groups. Recently, we implemented the Statistically-Assisted Fluid Registration Algorithm or SAFIRA, 1 which is designed for tracking morphometric differences among populations. To this end, SAFIRA allows the inclusion of statistical priors extracted from the populations being studied as regularizers in the registration. This flexibility and degree of sophistication limit the tool to expert use, even more so considering that SAFIRA was initially implemented in command line mode. Here, we introduce a new, intuitive, easy to use, Matlab-based graphical user interface for SAFIRA’s multivariate TBM. The interface also generates different choices for the TBM statistics, including both the traditional univariate statistics on the Jacobian matrix, and comparison of the full deformation tensors. 2 This software will be freely disseminated to the neuroimaging research community.

We develop a new algorithm to compute voxel-wise shape differences in tensor-based morphometry (TBM). As in standard TBM, we non-linearly register brain T1-weighed MRI data from a patient and control group to a template, and compute the Jacobian of the deformation fields. In standard TBM, the determinants of the Jacobian matrix at each voxel are statistically compared between the two groups. More recently, a multivariate extension of the statistical analysis involving the deformation tensors derived from the Jacobian matrices has been shown to improve statistical detection power. 7 However, multivariate methods comprising large numbers of variables are computationally intensive and may be subject to noise. In addition, the anatomical interpretation of results is sometimes difficult. Here instead, we analyze the eigenvalues and the eigenvectors of the Jacobian matrices. Our method is validated on brain MRI data from Alzheimer’s patients and healthy elderly controls from the Alzheimer’s Disease Neuro Imaging Database.

Imaging genetics is an emerging methodology that combines genetic information with imaging-derived metrics to understand how genetic factors impact observable structural, functional, and quantitative phenotypes. Many of the most well-known genetic studies are based on Genome-Wide Association Studies (GWAS), which use large populations of related or unrelated individuals to associate traits and disorders with individual genetic factors. Merging imaging and genetics may potentially lead to improved power of association in GWAS because imaging traits may be more sensitive phenotypes, being closer to underlying genetic mechanisms, and their quantitative nature inherently increases power. We are developing SOLAR-ECLIPSE (SE) imaging genetics software which is capable of performing genetic analyses with both large-scale quantitative trait data and family structures of variable complexity. This program can estimate the contribution of genetic commonality among related subjects to a given phenotype, and essentially answer the question of whether or not the phenotype is heritable. This central factor of interest, heritability, offers bounds on the direct genetic influence over observed phenotypes. In order for a trait to be a good phenotype for GWAS, it must be heritable: at least some proportion of its variance must be due to genetic influences. A variety of family structures are commonly used for estimating heritability, yet the variability and biases for each as a function of the sample size are unknown. Herein, we investigate the ability of SOLAR to accurately estimate heritability models based on imaging data simulated using Monte Carlo methods implemented in R. We characterize the bias and the variability of heritability estimates from SOLAR as a function of sample size and pedigree structure (including twins, nuclear families, and nuclear families with grandparents).

The detection, segmentation and quantification of multiple sclerosis (MS) lesions on magnetic resonance images (MRI) has been a very active field for the last two decades because of the urge to correlate these measures with the effectiveness of pharmacological treatment. A myriad of methods has been developed and most of these are non specific for the type of lesions and segment the lesions in their acute and chronic phases together. On the other hand, radiologists are able to distinguish between several stages of the disease on different types of MRI images. The main motivation of the work presented here is to computationally emulate the visual perception of the radiologist by using modeling principles of the neuronal centers along the visual system. By using this approach we are able to detect the lesions in the majority of the images in our population sample. This type of approach also allows us to study and improve the analysis of brain networks by introducing a priori information.

Introduction: Dynamic Contrast-Enhanced-Magnetic Resonance Imaging (DCE-MRI) appears to provide an unambiguous means of tracking the outcome of photodynamic therapy (PDT) of brain tumors with the photosensitizer Pc 4. The increase in Gd enhancement observed after Pc 4-PDT may be due to a temporary opening of the blood-brain-barrier which, as noted by others, may offer a therapeutic window. Methods: We injected 2.5 x 105 U87 cells into the brains of 9 athymic nude rats. After 8-9 days peri-tumor DCE-MRI images were acquired on a 7.0 T microMRI scanner before and after the administration of 150 μL Gd. DCE-MRI scans were repeated three times following Pc 4-PDT. Results: The average, normalized peak enhancement in the tumor region, approximately 30-90 seconds after Gd administration, was 1.31 times greater than baseline (0.03 Standard Error [SE]) prior to PDT and was 1.44 (0.02 SE) times baseline in the first Post-PDT scans (Day 11), a statistically significant (p ≈ 0.014, N=8) increase over the Pre- PDT scans, and was 1.38 (0.02 SE) times baseline in the second scans (Day 12), also a statistically significant (p ≈ 0.008, N=7) increase. Observations were mixed in the third Post-PDT scans (Day 13), averaging 1.29 (0.03 SE) times baseline (p ≈ 0.66, N=7). Overall a downward trend in enhancement was observed from the first to the third Post-PDT scans. Discussion: DCE-MRI may provide an unambiguous indication of brain tumor PDT outcome. The initial increase in DCE-MRI signal may correlate with a temporary, PDT-induced opening of the blood-brain-barrier, creating a potential therapeutic window.

The tremendous pace of development in brain imaging technologies has revolutionized our ability to investigate brain structure and function. Techniques are now available to capture features of anatomy and function at both molecular and whole-brain scales, mapping neuronal dynamics and gene expression as well as growth and degenerative processes that span multiyear time scales. The number of brain imaging investigations is also increasing exponentially. A major goal of these studies is to analyze how the dynamically changing brain varies across age, gender, disease, across multiple imaging modalities, and in large human populations. To tackle these questions, many laboratories are using sophisticated algorithms for brain image analysis. Engineering approaches drawn from computer vision, image analysis, computer graphics, and artificial intelligence research fields are required to manipulate, analyze, and communicate brain data. Novel image analysis algorithms continue to uncover new patterns of altered structure and function in individuals and clinical populations, and mathematical strategies are being developed to relate these patterns to clinical, demographic, and genetic parameters. In this chapter, we review current challenges in brain image analysis, focusing on the main algorithms, their technical foundations, and their scientific and clinical applications. The approaches include methods for automated registration and segmentation, anatomical parameterization and modeling, tissue classification and shape analysis, and pathology detection in individuals or groups. Algorithms are also described for generating digital brain atlases. Atlases are fundamental to brain image analysis, as they offer a powerful framework to synthesize the results of disparate imaging studies. Built from one or more representations of the brain, atlases are annotated representations of anatomy in a 3D coordinate system. They serve as standardized templates on which other brain maps can be overlaid, for subsequent comparison and integration. To align new imaging data with an atlas, a variety of registration algorithms may be employed (see also Chapter 8 for other applications). Once registered, brain maps can be pooled across subjects and combined mathematically and statistically. As such, atlases provide a standardized 3D coordinate system to express observations from different individuals and a framework for interlaboratory communication.

Robust and automatic non-rigid registration depends on many parameters that have not yet been systematically explored. Here we determined how tissue classification influences non-linear fluid registration of brain MRI. Twin data is ideal for studying this question, as volumetric correlations between corresponding brain regions that are under genetic control should be higher in monozygotic twins (MZ) who share 100% of their genes when compared to dizygotic twins (DZ) who share half their genes on average. When these substructure volumes are quantified using tensor-based morphometry, improved registration can be defined based on which method gives higher MZ twin correlations when compared to DZs, as registration errors tend to deplete these correlations. In a study of 92 subjects, higher effect sizes were found in cumulative distribution functions derived from statistical maps when performing tissue classification before fluid registration, versus fluidly registering the raw images. This gives empirical evidence in favor of pre-segmenting images for tensor-based morphometry.

We developed an automated analysis pipeline to analyze 3D changes in ventricular morphology; it provides a highly sensitive quantitative marker of Alzheimer's disease (AD) progression for MRI studies. In the ADNI image database, we created expert delineations of the ventricles, as parametric surface meshes, in 6 brain MRI scans. These 6 images and their embedded surfaces were fluidly registered to MRI scans of 80 AD patients, 80 individuals with mild cognitive impairment (MCI), and 80 healthy controls. Surface averaging within subjects greatly reduced segmentation error. Surface-based statistical maps revealed powerful correlations between surface morphology at baseline and (1) diagnosis, (2) cognitive performance (MMSE scores), (3) depression, and (4) predicted future decline, over a 1 year interval, in 3 standard clinical scores (MMSE, global and sum-of-boxes CDR). We used a false discovery rate method (FDR) method based on cumulative probability plots to find that 40 subjects were sufficient to discriminate AD from normal groups. 60 and 119 subjects, respectively, were required to correlate ventricular enlargement with MMSE and clinical depression. Surface-based FDR, along with multi-atlas fluid registration to reduce segmentation error, will allow researchers to (1) estimate sample sizes with adequate power to detect groups differences, and (2) compare the power of mapping methods head-to-head, optimizing cost-effectiveness for future clinical trials.

We propose a unified variational approach for registration of gene expression data to neuroanatomical mouse atlas in two dimensions. The proposed energy (minimized in the unknown displacement u) is composed of three terms: a standard data fidelity term based on L 2 similarity measure, a regularizing term based on nonlinear elasticity (allowing larger smooth deformations), and a geometric penalty constraint for landmark matching. We overcome the difficulty of minimizing the nonlinear elasticity functional by introducing an auxiliary variable v that approximates ∇u, the Jacobian of the unknown displacement u. We therefore minimize now the functional with respect to the unknowns u (a vector-valued function of two dimensions) and v (a two-by-two matrix-valued function). An additional quadratic term is added, to insure good agreement between v and ∇u. In this way, the nonlinearity in the derivatives of the unknown u no longer exists in the obtained Euler-Lagrange equations, producing simpler implementations. Several satisfactory experimental results show that gene expression data are mapped to a mouse atlas with good landmark matching and smooth deformation. We also present comparisons with the biharmonic regularization. An advantage of the proposed nonlinear elasticity model is that usually no numerical correction such as regridding is necessary to keep the deformation smooth, while unifying the data fidelity term, regularization term, and landmark constraints in a single minimization approach.

In the past decade, information theory has been studied extensively in computational imaging. In particular, image matching by maximizing mutual information has been shown to yield good results in multimodal image registration. However, there have been few rigorous studies to date that investigate the statistical aspect of the resulting deformation fields. Different regularization techniques have been proposed, sometimes generating deformations very different from one another. In this paper, we present a novel model for multimodal image registration. The proposed method minimizes a purely information-theoretic functional consisting of mutual information matching and unbiased regularization. The unbiased regularization term measures the magnitude of deformations using either asymmetric Kullback-Leibler divergence or its symmetric version. The new multimodal unbiased matching method, which allows for large topology preserving deformations, was tested using pairs of two and three dimensional serial MRI images. We compared the results obtained using the proposed model to those computed with a well-known mutual information based viscous fluid registration. A thorough statistical analysis demonstrated the advantages of the proposed model over the multimodal fluid registration method when recovering deformation fields and corresponding Jacobian maps.

Inter-subject analysis of anatomical and functional brain imaging data requires the images to be registered to a common coordinate system in which anatomical features are aligned. Intensity-based volume registration methods can align subcortical structures well, but the variability in sulcal folding patterns typically results in misalignment of the cortical surface. Conversely, surface-based registration using sulcal features can produce excellent cortical alignment but the mapping between brains is restricted to the cortical surface. Here we describe a method for volumetric registration that also produces a one-to-one point correspondence between cortical surfaces. This is achieved by first parameterizing and aligning the cortical surfaces. We then use a constrained harmonic mapping to define a volumetric correspondence between brains. Finally, the correspondence is refined using an intensity-based warp. We evaluate the performance of our proposed method in terms of the inter-subject alignment of expert-labeled sub-cortical structures after registration.

A fully automated application was developed and used for the registration of T1-weighted magnetic resonance images (MRIs) for Alzheimer patients. Two methods for image registration were implemented and compared: affine and nonlinear registration. Nonlinear registration uses continuum-mechanics-based elastic deformation. The affine registration algorithm is linear and is generated by an amplitude-modulated phase-only filter. The nonlinear registration method uses an elastic transformation generated by Navier-Stokes continuum-mechanics models. The validation method to quantitatively compare the performance of the affine and nonlinear registration algorithms uses root-mean-square error and three-dimensional volume rendering.

We developed a general method for global conformal parameterizations based on the structure of the cohomology group of holomorphic one-forms with or without boundaries. 1, 2 For genus zero surfaces, our algorithm can find a unique mapping between any two genus zero manifolds by minimizing the harmonic energy of the map. In this paper, we apply the algorithm to the cortical surface matching problem. We use a mesh structure to represent the brain surface. Further constraints are added to ensure that the conformal map is unique. Empirical tests on MRI data show that the mappings preserve angular relationships, are stable in MRIs acquired at different times, and are robust to differences in data triangulation, and resolution. Compared with other brain surface conformal mapping algorithms, our algorithm is more stable and has good extensibility.

Presently available anatomic atlases provide useful coordinate systems such as the ubiquitous Talairach system but are sorely lacking in both spatial resolution and completeness. An appropriately sampled anatomic specimen can provide the additional detail necessary to accurately localize activation sites as well as provide other structural perspectives such as chemoarchitecture. We collected serial section postmortem anatomic data from several whole human head and brain specimens using a cryosectioning technique. Tissue imaged so that voxel resolution was 200 microns or better at full color. These high resolution datasets along with collections of MR data were placed within a common coordinate system and used to produce a probabilistic representation. This approach represents anatomy within a coordinate system as a probability. Coordinate locations are assigned a confidence limit to describe the likelihood that a given location belongs to an anatomic structure based upon the population of specimens. A variety of warping strategies are discussed to provide statistics on morphometric variability and probability. High dimensional anatomically based warps utilizing sulcal anatomy are described. These data are an important and necessary part of the comprehensive structural and functional analyses that focus on the mapping of the human brain.