Discovery and visualization of structural biomarkers from MRI using transport-based morphometry

Review of structural neuroimaging and genetic findings in autism spectrum disorder - a clinical perspective

Autism spectrum disorders (ASDs) are neurodevelopmental conditions characterized by deficits in social relationships and communication and restrictive, repetitive behaviors and interests. ASDs form a heterogeneous group from a clinical and genetic perspective. Currently, ASDs diagnosis is based on the clinical observation of the individual's behavior. The subjective nature of behavioral diagnoses, in the context of ASDs heterogeneity, contributes to significant variation in the age at ASD diagnosis. Early detection has been proved to be critical in ASDs, as early start of appropriate therapeutic interventions greatly improve the outcome for some children. Structural magnetic resonance imaging (MRI) is widely used in the diagnostic work-up of neurodevelopmental conditions, including ASDs, mostly for brain malformations detection. Recently, the focus of brain imaging shifted towards quantitative MRI parameters, aiming to identify subtle changes that may establish early detection biomarkers. ASDs have a strong genetic component; deletions and duplications of several genomic loci have been strongly associated with ASDs risk. Consequently, a multitude of neuroimaging and genetic findings emerged in ASDs in the recent years. The association of gross or subtle changes in brain morphometry and volumes with different genetic defects has the potential to bring new insights regarding normal development and pathomechanisms of various disorders affecting the brain. Still, the clinical implications of these discoveries and the impact of genetic abnormalities on brain structure and function are unclear. Here we review the literature on brain imaging correlated with the most prevalent genomic imbalances in ASD, and discuss the potential clinical impact.

Discovering the gene-brain-behavior link in autism via generative machine learning

Autism is traditionally diagnosed behaviorally but has a strong genetic basis. A genetics-first approach could transform understanding and treatment of autism. However, isolating the gene-brain-behavior relationship from confounding sources of variability is a challenge. We demonstrate a novel technique, 3D transport-based morphometry (TBM), to extract the structural brain changes linked to genetic copy number variation (CNV) at the 16p11.2 region. We identified two distinct endophenotypes. In data from the Simons Variation in Individuals Project, detection of these endophenotypes enabled 89 to 95% test accuracy in predicting 16p11.2 CNV from brain images alone. Then, TBM enabled direct visualization of the endophenotypes driving accurate prediction, revealing dose-dependent brain changes among deletion and duplication carriers. These endophenotypes are sensitive to articulation disorders and explain a portion of the intelligence quotient variability. Genetic stratification combined with TBM could reveal new brain endophenotypes in many neurodevelopmental disorders, accelerating precision medicine, and understanding of human neurodiversity.

Invariance encoding in sliced-Wasserstein space for image classification with limited training data

Deep convolutional neural networks (CNNs) are broadly considered to be state-of-the-art generic end-to-end image classification systems. However, they are known to underperform when training data are limited and thus require data augmentation strategies that render the method computationally expensive and not always effective. Rather than using a data augmentation strategy to encode invariances as typically done in machine learning, here we propose to mathematically augment a nearest subspace classification model in sliced-Wasserstein space by exploiting certain mathematical properties of the Radon Cumulative Distribution Transform (R-CDT), a recently introduced image transform. We demonstrate that for a particular type of learning problem, our mathematical solution has advantages over data augmentation with deep CNNs in terms of classification accuracy and computational complexity, and is particularly effective under a limited training data setting. The method is simple, effective, computationally efficient, non-iterative, and requires no parameters to be tuned. Python code implementing our method is available at https://github.com/rohdelab/mathematical augmentation. Our method is integrated as a part of the software package PyTransKit, which is available at https://github.com/rohdelab/PyTransKit.

Optimal transport features for morphometric population analysis

Brain pathologies often manifest as partial or complete loss of tissue. The goal of many neuroimaging studies is to capture the location and amount of tissue changes with respect to a clinical variable of interest, such as disease progression. Morphometric analysis approaches capture local differences in the distribution of tissue or other quantities of interest in relation to a clinical variable. We propose to augment morphometric analysis with an additional feature extraction step based on unbalanced optimal transport. The optimal transport feature extraction step increases statistical power for pathologies that cause spatially dispersed tissue loss, minimizes sensitivity to shifts due to spatial misalignment or differences in brain topology, and separates changes due to volume differences from changes due to tissue location. We demonstrate the proposed optimal transport feature extraction step in the context of a volumetric morphometric analysis of the OASIS-1 study for Alzheimer's disease. The results demonstrate that the proposed approach can identify tissue changes and differences that are not otherwise measurable.

Effects of Mild Traumatic Brain Injury on Resting State Brain Network Connectivity in Older Adults

Older age is associated with worsened outcome after mild traumatic brain injury (mTBI) and a higher risk of developing persistent post-traumatic complaints. However, the effects of mTBI sequelae on brain connectivity at older age and their association with post-traumatic complaints remain understudied.We analyzed multi-echo resting-state functional magnetic resonance imaging data from 25 older adults with mTBI (mean age: 68 years, SD: 5 years) in the subacute phase (mean injury to scan interval: 38 days, SD: 9 days) and 20 age-matched controls. Severity of complaints (e.g. fatigue, dizziness) was assessed using self-reported questionnaires. Group independent component analysis was used to identify intrinsic connectivity networks (ICNs). The effects of group and severity of complaints on ICNs were assessed using spatial maps intensity (SMI) as a measure of within-network connectivity, and (static) functional network connectivity (FNC) as a measure of between-network connectivity.Patients indicated a higher total severity of complaints than controls. Regarding SMI measures, we observed hyperconnectivity in left-mid temporal gyrus (cognitive-language network) and hypoconnectivity in the right-fusiform gyrus (visual-cerebellar network) that were associated with group. Additionally, we found interaction effects for SMI between severity of complaints and group in the visual(-cerebellar) domain. Regarding FNC measures, no significant effects were found.In older adults, changes in cognitive-language and visual(-cerebellar) networks are related to mTBI. Additionally, group-dependent associations between connectivity within visual(-cerebellar) networks and severity of complaints might indicate post-injury (mal)adaptive mechanisms, which could partly explain post-traumatic complaints (such as dizziness and balance disorders) that are common in older adults during the subacute phase.

Investigating impact of cardiorespiratory fitness in reducing brain tissue loss caused by ageing

Mitigating the loss of brain tissue due to age is a major problem for an ageing population. Improving cardiorespiratory fitness has been suggested as a possible strategy, but the influenceon brain morphology has not been fully characterized. To investigate the dependent shifts in brain tissue distribution as a function of cardiorespiratory fitness, we used a 3D transport-based morphometry approach. In this study of 172 inactive older adults aged 58-81 (66.5 ± 5.7) years, cardiorespiratory fitness was determined by VO 2 peak (ml/kg/min) during graded exercise and brain morphology was assessed through structural magnetic resonance imaging. After correcting for covariates including age (in the fitness model), gender and level of education, we compared dependent tissue shifts with age to those due to V O 2   peak . We found a significant association between cardiorespiratory fitness and brain tissue distribution (white matter, r = 0.30, P = 0.003; grey matter, r = 0.40, P < 0.001) facilitated by direct visualization of the brain tissue shifts due to cardiorespiratory fitness through inverse transformation-a key capability of 3D transport-based morphometry. A strong statistical correlation was found between brain tissue changes related to ageing and those associated with lower cardiorespiratory fitness (white matter, r = 0.62, P < 0.001; grey matter, r = 0.74, P < 0.001). In both cases, frontotemporal regions shifted the most while basal ganglia shifted the least. Our results highlight the importance of cardiorespiratory fitness in maintaining brain health later in life. Furthermore, this work demonstrates 3D transport-based morphometry as a novel neuroinformatic technology that may aid assessment of therapeutic approaches for brain ageing and neurodegenerative diseases.

Radon Cumulative Distribution Transform Subspace Modeling for Image Classification

We present a new supervised image classification method applicable to a broad class of image deformation models. The method makes use of the previously described Radon Cumulative Distribution Transform (R-CDT) for image data, whose mathematical properties are exploited to express the image data in a form that is more suitable for machine learning. While certain operations such as translation, scaling, and higher-order transformations are challenging to model in native image space, we show the R-CDT can capture some of these variations and thus render the associated image classification problems easier to solve. The method - utilizing a nearest-subspace algorithm in the R-CDT space - is simple to implement, non-iterative, has no hyper-parameters to tune, is computationally efficient, label efficient, and provides competitive accuracies to state-of-the-art neural networks for many types of classification problems. In addition to the test accuracy performances, we show improvements (with respect to neural network-based methods) in terms of computational efficiency (it can be implemented without the use of GPUs), number of training samples needed for training, as well as out-of-distribution generalization. The Python code for reproducing our results is available at [1].

PARTITIONING SIGNAL CLASSES USING TRANSPORT TRANSFORMS FOR DATA ANALYSIS AND MACHINE LEARNING

A relatively new set of transport-based transforms (CDT, R-CDT, LOT) have shown their strength and great potential in various image and data processing tasks such as parametric signal estimation, classification, cancer detection among many others. It is hence worthwhile to elucidate some of the mathematical properties that explain the successes of these transforms when they are used as tools in data analysis, signal processing or data classification. In particular, we give conditions under which classes of signals that are created by algebraic generative models are transformed into convex sets by the transport transforms. Such convexification of the classes simplify the classification and other data analysis and processing problems when viewed in the transform domain. More specifically, we study the extent and limitation of the convexification ability of these transforms under an algebraic generative modeling framework. We hope that this paper will serve as an introduction to these transforms and will encourage mathematicians and other researchers to further explore the theoretical underpinnings and algorithmic tools that will help understand the successes of these transforms and lay the groundwork for further successful applications.

A Heteromodal Word-Meaning Binding Site in the Visual Word Form Area under Top-Down Frontoparietal Control

The integral capacity of human language together with semantic memory drives the linkage of words and their meaning, which theoretically is subject to cognitive control. However, it remains unknown whether, across different language modalities and input/output formats, there is a shared system in the human brain for word-meaning binding and how this system interacts with cognitive control. Here, we conducted a functional magnetic resonance imaging experiment based on a large cohort of subjects (50 females, 50 males) to comprehensively measure the brain responses evoked by semantic processing in spoken and written word comprehension and production tasks (listening, speaking, reading, and writing). We found that heteromodal word input and output tasks involved distributed brain regions within a frontal-parietal-temporal network and focally coactivated the anterior lateral visual word form area (VWFA), which is located in the basal occipitotemporal area. Directed connectivity analysis revealed that the VWFA was invariably under significant top-down modulation of the frontoparietal control network and interacts with regions related to attention and semantic representation. This study reveals that the VWFA is a key site subserving general semantic processes linking words and meaning, challenging the predominant emphasis on this area's specific role in reading or more general visual processes. Our findings also suggest that the dynamics between semantic memory and cognitive control mechanisms during word processing are largely independent of the modalities of input or output.SIGNIFICANCE STATEMENT Binding words and their meaning into a coherent whole during retrieval requires accessing semantic memory and cognitive control, allowing our thoughts to be expressed and comprehended through mind-external tokens in multiple modalities, such as written or spoken forms. However, it is still unknown whether multimodal language comprehension and production share a common word-meaning binding system in human brains and how this system is connected to a cognitive control mechanism. By systematically measuring brain activity evoked by spoken and written verbal input and output tasks tagging word-meaning binding processes, we demonstrate a general word-meaning binding site within the visual word form area (VWFA) and how this site is modulated by the frontal-parietal control network.

Irregular Respiratory Motion Compensation for Liver Contrast-Enhanced Ultrasound via Transport-Based Motion Estimation

Contrast-enhanced ultrasound (CEUS) imaging has been widely applied for the detection and characterization of focal liver lesions (FLLs), for its ability to visualize the blood flow in real time. However, cyclic liver motion poses a great challenge to the recovery of perfusion curves as well as quantitative kinetic parameters estimation. Recently, a few gating methods have been proposed to eliminate unexpected intensity fluctuations by the breathing phase estimation, with the assumption that each breathing phase corresponds to a specific lesion position strictly. While practical liver motion tends to be irregular due to changes in the patient's underlying physiologic status, that is, the same phase might not correspond to the same position. To tackle the challenge of motion irregularity, we present a novel motion estimation-based respiratory compensation method, named RCME, which first estimates salient motion information through the framework of optimal transport (OT) by jointly modeling pixel intensity as well as their locations and then employs sparse subspace clustering (SSC) to identify the subset of frames acquired at the same position. Our proposed method is evaluated on 15 clinical CEUS sequences in both qualitative and quantitative manners. Experimental results demonstrate good performance on irregular liver motion compensation.

Signal-Processing Framework for Ultrasound Compressed Sensing Data: Envelope Detection and Spectral Analysis

Acquisition times and storage requirements have become increasingly important in signal-processing applications, as the sizes of datasets have increased. Hence, compressed sensing (CS) has emerged as an alternative processing technique, as original signals can be reconstructed using fewer data samples collected at frequencies below the Nyquist sampling rate. However, further analysis of CS data in both time and frequency domains requires the reconstruction of the original form of the time-domain data, as traditional signal-processing techniques are designed for uncompressed data. In this paper, we propose a signal-processing framework that extracts spectral properties for frequency-domain analysis directly from under-sampled ultrasound CS data, using an appropriate basis matrix, and efficiently converts this into the envelope of a time-domain signal, avoiding full reconstruction. The technique generates more accurate results than the traditional framework in both time- and frequency-domain analyses, and is simpler and faster in execution than full reconstruction, without any loss of information. Hence, the proposed framework offers a new standard for signal processing using ultrasound CS data, especially for small and portable systems handling large datasets.

Trajectories from Distribution-valued Functional Curves: A Unified Wasserstein Framework★

Temporal changes in medical images are often evaluated along a parametrized function that represents a structure of interest (e.g. white matter tracts). By attributing samples along these functions with distributions of image properties in the local neighborhood, we create distribution-valued signatures for these functions. We propose a novel and comprehensive framework which models their temporal evolution trajectories. This is achieved under the unifying scheme of Wasserstein distance metric. The regression problem is formulated as a constrained optimization problem and solved using an alternating projection algorithm. The solution simultaneously preserves the functional characteristics of the curve, models the temporal change in distribution profiles and forces the estimated distributions to be valid. Hypothesis testing is applied in two ways using Wasserstein based test statistics. Validation is presented on synthetic data. Detection of delayed growth is shown on DTI tracts, for a pediatric subject with respect to a healthy population of infants.

Enabling early detection of osteoarthritis from presymptomatic cartilage texture maps via transport-based learning

Many diseases have no visual cues in the early stages, eluding image-based detection. Today, osteoarthritis (OA) is detected after bone damage has occurred, at an irreversible stage of the disease. Currently no reliable method exists for OA detection at a reversible stage. We present an approach that enables sensitive OA detection in presymptomatic individuals. Our approach combines optimal mass transport theory with statistical pattern recognition. Eighty-six healthy individuals were selected from the Osteoarthritis Initiative, with no symptoms or visual signs of disease on imaging. On 3-y follow-up, a subset of these individuals had progressed to symptomatic OA. We trained a classifier to differentiate progressors and nonprogressors on baseline cartilage texture maps, which achieved a robust test accuracy of 78% in detecting future symptomatic OA progression 3 y prior to symptoms. This work demonstrates that OA detection may be possible at a potentially reversible stage. A key contribution of our work is direct visualization of the cartilage phenotype defining predictive ability as our technique is generative. We observe early biochemical patterns of fissuring in cartilage that define future onset of OA. In the future, coupling presymptomatic OA detection with emergent clinical therapies could modify the outcome of a disease that costs the United States healthcare system $16.5 billion annually. Furthermore, our technique is broadly applicable to earlier image-based detection of many diseases currently diagnosed at advanced stages today.

Machine intelligence in healthcare-perspectives on trustworthiness, explainability, usability, and transparency

Machine Intelligence (MI) is rapidly becoming an important approach across biomedical discovery, clinical research, medical diagnostics/devices, and precision medicine. Such tools can uncover new possibilities for researchers, physicians, and patients, allowing them to make more informed decisions and achieve better outcomes. When deployed in healthcare settings, these approaches have the potential to enhance efficiency and effectiveness of the health research and care ecosystem, and ultimately improve quality of patient care. In response to the increased use of MI in healthcare, and issues associated when applying such approaches to clinical care settings, the National Institutes of Health (NIH) and National Center for Advancing Translational Sciences (NCATS) co-hosted a Machine Intelligence in Healthcare workshop with the National Cancer Institute (NCI) and the National Institute of Biomedical Imaging and Bioengineering (NIBIB) on 12 July 2019. Speakers and attendees included researchers, clinicians and patients/ patient advocates, with representation from industry, academia, and federal agencies. A number of issues were addressed, including: data quality and quantity; access and use of electronic health records (EHRs); transparency and explainability of the system in contrast to the entire clinical workflow; and the impact of bias on system outputs, among other topics. This whitepaper reports on key issues associated with MI specific to applications in the healthcare field, identifies areas of improvement for MI systems in the context of healthcare, and proposes avenues and solutions for these issues, with the aim of surfacing key areas that, if appropriately addressed, could accelerate progress in the field effectively, transparently, and ethically.

An Emerging Model for Cancer Development from a Tumor Microenvironment Perspective in Mice and Humans

In the past, cancer development was studied in terms of genetic mutations acquired in cancer cells at each stage of the development. We present an emerging model for cancer development in which the tumor microenvironment (TME) plays an integral part. In this model, the tumor development is initiated by a slowly growing nearly homogeneous colony of cancer cells that can evade detection by the cell's innate mechanism of immunity such as natural killer (NK) cells (first stage; colonization). Subsequently, the colony develops into a tumor filled with lymphocytes and stromal cells, releasing pro-inflammatory cytokines, growth factors, and chemokines (second stage; lymphocyte infiltration). Cancer progression proceeds to a well-vesiculated silent tumor releasing no inflammatory signal, being nearly devoid of lymphocytes (third stage; silenced). Eventually some cancer cells within a tumor undertake epithelial-to-mesenchymal transition (EMT), which leads to cancer metastasis (fourth stage; EMT). If a circulating metastasized cancer cell finds a niche in a new tissue and evades detection by NK cells, it can establish a new colony in which very few stromal cells are present (fifth stage; metastasis), which is much like a colony at the first stage of development. At every stage, cancer cells influence their own TME, and in turn, the TME influences the cancer cells contained within, either by direct interaction between cancer cells and stromal cells or through exchange of cytokines. In this article, we examine clinical findings and animal experiments pertaining to this paradigm-shifting model and consider if, indeed, some aspects of cancer development are governed solely by the TME.

Brain Morphological Dynamics of Procrastination: The Crucial Role of the Self-Control, Emotional, and Episodic Prospection Network

Globally, about 17% individuals are suffering from the maladaptive procrastination until now, which impacts individual's financial status, mental health, and even public policy. However, the comprehensive understanding of neuroanatomical understructure of procrastination still remains gap. 688 participants including 3 independent samples were recruited for this study. Brain morphological dynamics referred to the idiosyncrasies of both brain size and brain shape. Multilinear regression analysis was utilized to delineate brain morphological dynamics of procrastination in Sample 1. In the Sample 2, cross-validation was yielded. Finally, prediction models of machine learning were conducted in Sample 3. Procrastination had a significantly positive correlation with the gray matter volume (GMV) in the left insula, anterior cingulate gyrus (ACC), and parahippocampal gyrus (PHC) but was negatively correlated with GMV of dorsolateral prefrontal cortex (dlPFC) and gray matter density of ACC. Furthermore, procrastination was positively correlated to the cortical thickness and cortical complexity of bilateral orbital frontal cortex (OFC). In Sample 2, all the results were cross-validated highly. Predication analysis demonstrated that these brain morphological dynamic can predict procrastination with high accuracy. This study ascertained the brain morphological dynamics involving in self-control, emotion, and episodic prospection brain network for procrastination, which advanced promising aspects of the biomarkers for it.

The quandary of covarying: A brief review and empirical examination of covariate use in structural neuroimaging studies on psychological variables

Although covarying for potential confounds or nuisance variables is common in psychological research, relatively little is known about how the inclusion of covariates may influence the relations between psychological variables and indices of brain structure. In Part 1 of the current study, we conducted a descriptive review of relevant articles from the past two years of NeuroImage in order to identify the most commonly used covariates in work of this nature. Age, sex, and intracranial volume were found to be the most commonly used covariates, although the number of covariates used ranged from 0 to 14, with 37 different covariate sets across the 68 models tested. In Part 2, we used data from the Human Connectome Project to investigate the degree to which the addition of common covariates altered the relations between individual difference variables (i.e., personality traits, psychopathology, cognitive tasks) and regional gray matter volume (GMV), as well as the statistical significance of values associated with these effect sizes. Using traditional and random sampling approaches, our results varied widely, such that some covariate sets influenced the relations between the individual difference variables and GMV very little, while the addition of other covariate sets resulted in a substantially different pattern of results compared to models with no covariates. In sum, these results suggest that the use of covariates should be critically examined and discussed as part of the conversation on replicability in structural neuroimaging. We conclude by recommending that researchers pre-register their analytic strategy and present information on how relations differ based on the inclusion of covariates.

Assessing Postconcussive Reaction Time Using Transport-Based Morphometry of Diffusion Tensor Images

BACKGROUND AND PURPOSE

Cognitive deficits are among the most commonly reported post-concussive symptoms, yet the underlying microstructural injury is poorly understood. Our aim was to discover white matter injury underlying reaction time in mild traumatic brain injury DTI by applying transport-based morphometry.

MATERIALS AND METHODS

In this retrospective study, we performed DTI on 64 postconcussive patients (10-28 years of age; 69% male, 31% female) between January 2006 and March 2013. We measured the reaction time percentile by using Immediate Post-Concussion Assessment and Cognitive Testing. Using the 3D transport-based morphometry technique we developed, we mined fractional anisotropy maps to extract the common microstructural injury associated with reaction time percentile in an automated manner. Permutation testing established statistical significance of the extracted injuries. We visualized the physical substrate responsible for reaction time through inverse transport-based morphometry transformation.

RESULTS

The direction in the transport space most correlated with reaction time was significant after correcting for covariates of age, sex, and time from injury (Pearson r = 0.44, P < .01). Inverting the computed direction using transport-based morphometry illustrates physical shifts in fractional anisotropy in the corpus callosum (increase) and within the optic radiations, corticospinal tracts, and anterior thalamic radiations (decrease) with declining reaction time. The observed shifts are consistent with biologic pathways underlying the visual-spatial interpretation and response-selection aspects of reaction time.

CONCLUSIONS

Transport-based morphometry discovers complex white matter injury underlying postconcussive reaction time in an automated manner. The potential influences of edema and axonal loss are visualized in the visual-spatial interpretation and response-selection pathways. Transport-based morphometry can bridge the gap between brain microstructure and function in diseases in which the structural basis is unknown.

Exploratory Population Analysis with Unbalanced Optimal Transport

The plethora of data from neuroimaging studies provide a rich opportunity to discover effects and generate hypotheses through exploratory data analysis. Brain pathologies often manifest in changes in shape along with deterioration and alteration of brain matter, i.e., changes in mass. We propose a morphometry approach using unbalanced optimal transport that detects and localizes changes in mass and separates them from changes due to the location of mass. The approach generates images of mass allocation and mass transport cost for each subject in the population. Voxelwise correlations with clinical variables highlight regions of mass allocation or mass transfer related to the variables. We demonstrate the method on the white and gray matter segmentations from the OASIS brain MRI data set. The separation of white and gray matter ensures that optimal transport does not transfer mass between different tissues types and separates gray and white matter related changes. The OASIS data set includes subjects ranging from healthy to mild and moderate dementia, and the results corroborate known pathology changes related to dementia that are not discovered with traditional voxel-based morphometry. The transport-based morphometry increases the explanatory power of regression on clinical variables compared to traditional voxel-based morphometry, indicating that transport cost and mass allocation images capture a larger portion of pathology induced changes.