Automatic clustering and population analysis of white matter tracts using maximum density paths

Deep fiber clustering: Anatomically informed fiber clustering with self-supervised deep learning for fast and effective tractography parcellation

White matter fiber clustering is an important strategy for white matter parcellation, which enables quantitative analysis of brain connections in health and disease. In combination with expert neuroanatomical labeling, data-driven white matter fiber clustering is a powerful tool for creating atlases that can model white matter anatomy across individuals. While widely used fiber clustering approaches have shown good performance using classical unsupervised machine learning techniques, recent advances in deep learning reveal a promising direction toward fast and effective fiber clustering. In this work, we propose a novel deep learning framework for white matter fiber clustering, Deep Fiber Clustering (DFC), which solves the unsupervised clustering problem as a self-supervised learning task with a domain-specific pretext task to predict pairwise fiber distances. This process learns a high-dimensional embedding feature representation for each fiber, regardless of the order of fiber points reconstructed during tractography. We design a novel network architecture that represents input fibers as point clouds and allows the incorporation of additional sources of input information from gray matter parcellation. Thus, DFC makes use of combined information about white matter fiber geometry and gray matter anatomy to improve the anatomical coherence of fiber clusters. In addition, DFC conducts outlier removal naturally by rejecting fibers with low cluster assignment probability. We evaluate DFC on three independently acquired cohorts, including data from 220 individuals across genders, ages (young and elderly adults), and different health conditions (healthy control and multiple neuropsychiatric disorders). We compare DFC to several state-of-the-art white matter fiber clustering algorithms. Experimental results demonstrate superior performance of DFC in terms of cluster compactness, generalization ability, anatomical coherence, and computational efficiency.

White matter degradation near cerebral microbleeds is associated with cognitive change after mild traumatic brain injury

To explore how cerebral microbleeds (CMBs) accompanying mild traumatic brain injury (mTBI) reflect white matter (WM) degradation and cognitive decline, magnetic resonance images were acquired from 62 mTBI adults (imaged ∼7 days and ∼6 months post-injury) and 203 matched healthy controls. On average, mTBI participants had a count of 2.7 ± 2.6 traumatic CMBs in WM, located 6.1 ± 4.4 mm from cortex. At ∼6-month follow-up, 97% of CMBs were associated with significant reductions (34% ± 11%, q < 0.05) in the fractional anisotropy of WM streamlines within ∼1 cm of CMB locations. Male sex and older age were significant risk factors for larger reductions (q < 0.05). For CMBs in the corpus callosum, cingulum bundle, inferior and middle longitudinal fasciculi, fractional anisotropy changes were significantly and positively associated with changes in cognitive functions mediated by these structures (q < 0.05). Our findings distinguish traumatic from non-traumatic CMBs by virtue of surrounding WM alterations and challenge the assumption that traumatic CMBs are neurocognitively silent. Thus, mTBI with CMB findings can be described as a clinical endophenotype warranting longitudinal cognitive assessment.

Deep white matter analysis (DeepWMA): Fast and consistent tractography segmentation

White matter tract segmentation, i.e. identifying tractography fibers (streamline trajectories) belonging to anatomically meaningful fiber tracts, is an essential step to enable tract quantification and visualization. In this study, we present a deep learning tractography segmentation method (DeepWMA) that allows fast and consistent identification of 54 major deep white matter fiber tracts from the whole brain. We create a large-scale training tractography dataset of 1 million labeled fiber samples, and we propose a novel 2D multi-channel feature descriptor (FiberMap) that encodes spatial coordinates of points along each fiber. We learn a convolutional neural network (CNN) fiber classification model based on FiberMap and obtain a high fiber classification accuracy of 90.99% on the training tractography data with ground truth fiber labels. Then, the method is evaluated on a test dataset of 597 diffusion MRI scans from six independently acquired populations across genders, the lifespan (1 day - 82 years), and different health conditions (healthy control, neuropsychiatric disorders, and brain tumor patients). We perform comparisons with two state-of-the-art tract segmentation methods. Experimental results show that our method obtains a highly consistent tract segmentation result, where on average over 99% of the fiber tracts are successfully identified across all subjects under study, most importantly, including neonates and patients with space-occupying brain tumors. We also demonstrate good generalization of the method to tractography data from multiple different fiber tracking methods. The proposed method leverages deep learning techniques and provides a fast and efficient tool for brain white matter segmentation in large diffusion MRI tractography datasets.

Test-retest reproducibility of white matter parcellation using diffusion MRI tractography fiber clustering

There are two popular approaches for automated white matter parcellation using diffusion MRI tractography, including fiber clustering strategies that group white matter fibers according to their geometric trajectories and cortical-parcellation-based strategies that focus on the structural connectivity among different brain regions of interest. While multiple studies have assessed test-retest reproducibility of automated white matter parcellations using cortical-parcellation-based strategies, there are no existing studies of test-retest reproducibility of fiber clustering parcellation. In this work, we perform what we believe is the first study of fiber clustering white matter parcellation test-retest reproducibility. The assessment is performed on three test-retest diffusion MRI datasets including a total of 255 subjects across genders, a broad age range (5-82 years), health conditions (autism, Parkinson's disease and healthy subjects), and imaging acquisition protocols (three different sites). A comprehensive evaluation is conducted for a fiber clustering method that leverages an anatomically curated fiber clustering white matter atlas, with comparison to a popular cortical-parcellation-based method. The two methods are compared for the two main white matter parcellation applications of dividing the entire white matter into parcels (i.e., whole brain white matter parcellation) and identifying particular anatomical fiber tracts (i.e., anatomical fiber tract parcellation). Test-retest reproducibility is measured using both geometric and diffusion features, including volumetric overlap (wDice) and relative difference of fractional anisotropy. Our experimental results in general indicate that the fiber clustering method produced more reproducible white matter parcellations than the cortical-parcellation-based method.

Combining white matter diffusion and geometry for tract-specific alignment and variability analysis

We present a framework for along-tract analysis of white matter (WM) fiber bundles based on diffusion tensor imaging (DTI) and tractography. We introduce the novel concept of fiber-flux density for modeling fiber tracts' geometry, and combine it with diffusion-based measures to define vector descriptors called Fiber-Flux Diffusion Density (FFDD). The proposed model captures informative features of WM tracts at both the microscopic (diffusion-related) and macroscopic (geometry-related) scales, thus enabling improved sensitivity to subtle structural abnormalities that are not reflected by either diffusion or geometrical properties alone. A key step in this framework is the construction of an FFDD dissimilarity measure for sub-voxel alignment of fiber bundles, based on the fast marching method (FMM). The obtained aligned WM tracts enable meaningful inter-subject comparisons and group-wise statistical analysis. Moreover, we show that the FMM alignment can be generalized in a straight forward manner to a single-shot co-alignment of multiple fiber bundles. The proposed alignment technique is shown to outperform a well-established, commonly used DTI registration algorithm. We demonstrate the FFDD framework on the Human Connectome Project (HCP) diffusion MRI dataset, as well as on two different datasets of contact sports players. We test our method using longitudinal scans of a basketball player diagnosed with a traumatic brain injury, showing compatibility with structural MRI findings. We further perform a group study comparing mid- and post-season scans of 13 active football players exposed to repetitive head trauma, to 17 non-player control (NPC) subjects. Results reveal statistically significant FFDD differences (p-values<0.05) between the groups, as well as increased abnormalities over time at spatially-consistent locations within several major fiber tracts of football players.

An anatomically curated fiber clustering white matter atlas for consistent white matter tract parcellation across the lifespan

This work presents an anatomically curated white matter atlas to enable consistent white matter tract parcellation across different populations. Leveraging a well-established computational pipeline for fiber clustering, we create a tract-based white matter atlas including information from 100 subjects. A novel anatomical annotation method is proposed that leverages population-based brain anatomical information and expert neuroanatomical knowledge to annotate and categorize the fiber clusters. A total of 256 white matter structures are annotated in the proposed atlas, which provides one of the most comprehensive tract-based white matter atlases covering the entire brain to date. These structures are composed of 58 deep white matter tracts including major long range association and projection tracts, commissural tracts, and tracts related to the brainstem and cerebellar connections, plus 198 short and medium range superficial fiber clusters organized into 16 categories according to the brain lobes they connect. Potential false positive connections are annotated in the atlas to enable their exclusion from analysis or visualization. In addition, the proposed atlas allows for a whole brain white matter parcellation into 800 fiber clusters to enable whole brain connectivity analyses. The atlas and related computational tools are open-source and publicly available. We evaluate the proposed atlas using a testing dataset of 584 diffusion MRI scans from multiple independently acquired populations, across genders, the lifespan (1 day-82 years), and different health conditions (healthy control, neuropsychiatric disorders, and brain tumor patients). Experimental results show successful white matter parcellation across subjects from different populations acquired on multiple scanners, irrespective of age, gender or disease indications. Over 99% of the fiber tracts annotated in the atlas were detected in all subjects on average. One advantage in terms of robustness is that the tract-based pipeline does not require any cortical or subcortical segmentations, which can have limited success in young children and patients with brain tumors or other structural lesions. We believe this is the first demonstration of consistent automated white matter tract parcellation across the full lifespan from birth to advanced age.

Mapping population-based structural connectomes

Advances in understanding the structural connectomes of human brain require improved approaches for the construction, comparison and integration of high-dimensional whole-brain tractography data from a large number of individuals. This article develops a population-based structural connectome (PSC) mapping framework to address these challenges. PSC simultaneously characterizes a large number of white matter bundles within and across different subjects by registering different subjects' brains based on coarse cortical parcellations, compressing the bundles of each connection, and extracting novel connection weights. A robust tractography algorithm and streamline post-processing techniques, including dilation of gray matter regions, streamline cutting, and outlier streamline removal are applied to improve the robustness of the extracted structural connectomes. The developed PSC framework can be used to reproducibly extract binary networks, weighted networks and streamline-based brain connectomes. We apply the PSC to Human Connectome Project data to illustrate its application in characterizing normal variations and heritability of structural connectomes in healthy subjects.

Neighborhood resolved fiber orientation distributions (NRFOD) in automatic labeling of white matter fiber pathways

Accurate digital representation of major white matter bundles in the brain is an important goal in neuroscience image computing since the representations can be used for surgical planning, intra-patient longitudinal analysis and inter-subject population connectivity studies. Reconstructing desired fiber bundles generally involves manual selection of regions of interest by an expert, which is subject to user bias and fatigue, hence an automation is desirable. To that end, we first present a novel anatomical representation based on Neighborhood Resolved Fiber Orientation Distributions (NRFOD) along the fibers. The resolved fiber orientations are obtained by generalized q-sampling imaging (GQI) and a subsequent diffusion decomposition method. A fiber-to-fiber distance measure between the proposed fiber representations is then used in a density-based clustering framework to select the clusters corresponding to the major pathways of interest. In addition, neuroanatomical priors are utilized to constrain the set of candidate fibers before density-based clustering. The proposed fiber clustering approach is exemplified on automation of the reconstruction of the major fiber pathways in the brainstem: corticospinal tract (CST); medial lemniscus (ML); middle cerebellar peduncle (MCP); inferior cerebellar peduncle (ICP); superior cerebellar peduncle (SCP). Experimental results on Human Connectome Project (HCP)'s publicly available "WU-Minn 500 Subjects + MEG2 dataset" and expert evaluations demonstrate the potential of the proposed fiber clustering method in brainstem white matter structure analysis.

3D-SSF: A bio-inspired approach for dynamic multi-subject clustering of white matter tracts

There is growing interest in the study of white matter (WM) variation across subjects, and in particular the analysis of specific WM bundles, to better understand brain development and aging, as well as to improve early detection of some diseases. Several WM multi-subject clustering methods have been proposed to study WM bundles. These methods aim to overcome the complexity of the problem, which includes the huge size of the WM tractography datasets generated from multiple subjects, the existence of various streamlines with different positions, lengths and geometric forms, as well as the presence of outliers. However, the current methods are not sufficiently flexible to address all of these constraints. Here we introduce a novel dynamic multi-subject clustering framework based on a distributed multiagent implementation of the Multiple Species Flocking model, that we name 3D-Streamlines Stream Flocking (3D-SSF). Specifically, we consider streamlines from different subjects as data streams, and each streamline is assigned to a mobile agent. Agents work together following flocking rules in order to form a flock. Thanks to a similarity function, the agents that are associated with similar streamlines form a flock, whereas the agents that are associated with dissimilar streamlines are considered outliers. We use various experiments performed on noisy synthetic and real human brain data to validate 3D-SSF and demonstrate that it is more efficient and robust to outliers compared to other classical approaches. 3D-SSF is able to extract WM bundles at a population level, while considering WM variation across subjects and eliminating outlier streamlines.

An automated string-based approach to extracting and characterizing White Matter fiber-bundles

In this paper, we propose an automated approach to extracting White Matter (WM) fiber-bundles through clustering and model characterization. The key novelties of our approach are: a new string-based formalism, allowing an alternative representation of WM fibers, a new string dissimilarity metric, a WM fiber clustering technique, and a new model-based characterization algorithm. Thanks to these novelties, the complex problem of WM fiber-bundle extraction and characterization reduces to a much simpler and well-known string extraction and analysis problem. Interestingly, while several past approaches extract fiber-bundles by grouping available fibers on the basis of provided atlases (and, therefore, cannot capture possibly existing fiber-bundles nor represented in the atlases), our approach first clusters available fibers once and for all, and then tries to associate obtained clusters with models provided directly and dynamically by users. This more dynamic and interactive way of proceeding can help the detection of fiber-bundles autonomously proposed by our approach and not present in the initial models provided by experts.

White Matter Injury Susceptibility via Fiber Strain Evaluation Using Whole-Brain Tractography

Microscale brain injury studies suggest axonal elongation as a potential mechanism for diffuse axonal injury (DAI). Recent studies have begun to incorporate white matter (WM) structural anisotropy in injury analysis, with initial evidence suggesting improved injury prediction performance. In this study, we further develop a tractography-based approach to analyze fiber strains along the entire lengths of fibers from voxel- or anatomically constrained whole-brain tractography. This technique potentially extends previous element- or voxel-based methods that instead utilize WM fiber orientations averaged from typically coarse elements or voxels. Perhaps more importantly, incorporating tractography-based axonal structural information enables assessment of the overall injury risks to functionally important neural pathways and the anatomical regions they connect, which is not possible with previous methods. A DAI susceptibility index was also established to quantify voxel-wise WM local structural integrity and tract-wise damage of individual neural pathways. This "graded" injury susceptibility potentially extends the commonly employed treatment of injury as a simple binary condition. As an illustration, we evaluate the DAI susceptibilities of WM voxels and transcallosal fiber tracts in three idealized head impacts. Findings suggest the potential importance of the tractography-based approach for injury prediction. These efforts may enable future studies to correlate WM mechanical responses with neuroimaging, cognitive alteration, and concussion, and to reveal the relative vulnerabilities of neural pathways and identify the most vulnerable ones in real-world head impacts.

Maturation Along White Matter Tracts in Human Brain Using a Diffusion Tensor Surface Model Tract-Specific Analysis

Previous diffusion tensor imaging tractography studies have demonstrated exponential patterns of developmental changes for diffusion parameters such as fractional anisotropy (FA) and mean diffusivity (MD) averaged over all voxels in major white matter (WM) tracts of the human brain. However, this assumes that the entire tract is changing in unison, which may not be the case. In this study, a surface model based tract-specific analysis was applied to a cross-sectional cohort of 178 healthy subjects (83 males/95 females) aged from 6 to 30 years to spatially characterize the age-related changes of FA and MD along the trajectory of seven major WM tracts - corpus callosum (CC) and six bilateral tracts. There were unique patterns of regions that showed different exponential and linear rates of increasing FA or decreasing MD and age at which FA or MD levels off along each tract. Faster change rate of FA was observed in genu of CC and frontal-parietal part of superior longitudinal fasciculus (SLF). Inferior corticospinal tract (CST), posterior regions of association tracts such as inferior longitudinal fasciculus, inferior frontal occipital fasciculus and uncinate fasciculus also displayed earlier changing patterns for FA. MD decreases with age also exhibited this posterior-to-anterior WM maturation pattern for most tracts in females. Both males and females displayed similar FA/MD patterns of change with age along most large tracts; however, males had overall reached the FA maxima or MD minima later compared with females in most tracts with the greater differences occurring in the CST and frontal-parietal part of SLF for MD. Therefore, brain WM development has spatially varying trajectories along tracts that depend on sex and the tract.

GENETIC ANALYSIS OF STRUCTURAL BRAIN CONNECTIVITY USING DICCCOL MODELS OF DIFFUSION MRI IN 522 TWINS

Genetic and environmental factors affect white matter connectivity in the normal brain, and they also influence diseases in which brain connectivity is altered. Little is known about genetic influences on brain connectivity, despite wide variations in the brain's neural pathways. Here we applied the "DICCCOL" framework to analyze structural connectivity, in 261 twin pairs (522 participants, mean age: 21.8 y ± 2.7SD). We encoded connectivity patterns by projecting the white matter (WM) bundles of all "DICCCOLs" as a tracemap (TM). Next we fitted an A/C/E structural equation model to estimate additive genetic (A), common environmental (C), and unique environmental/error (E) components of the observed variations in brain connectivity. We found 44 "heritable DICCCOLs" whose connectivity was genetically influenced (a²>1%); half of them showed significant heritability (a²>20%). Our analysis of genetic influences on WM structural connectivity suggests high heritability for some WM projection patterns, yielding new targets for genome-wide association studies.

Sequential Hierarchical Agglomerative Clustering of White Matter Fiber Pathways

OBJECTIVE

We consider the problem of clustering white matter fiber pathways, extracted from diffusion MRI data via tractography, into bundles that are consistent with the neuroanatomy.

METHODS

We cast this problem as clustering streams of data, and use a sequential framework to process one fiber at a time. Our method, named as sequential hierarchical agglomerative clustering (HAC), represents the clusters with parametric models, performs HAC of relatively small number of fibers only when the parameters need to be initialized and/or updated, and assigns the labels to the following streams of data according to the current models.

RESULTS

Experiments on phantom data evaluate the sensitivity of our method to initialization and parameter tuning, and show its advantages over alternative techniques. Experiments on real data demonstrate its efficacy and speed in clustering white matter fiber pathways into anatomically distinct bundles.

CONCLUSION

Sequential HAC is a fast method that benefits from having a predefined number of clusters, and rapidly assigns labels to incoming data with high accuracy. It can be thought of as a mechanism that does clustering, while simultaneously accepting newly computed fibers; thereby, alleviating the burden of computing the distances between every pair of fibers in a tractogram.

SIGNIFICANCE

Sequential HAC is a practical tool that can interactively cluster fiber pathways and can be integrated into fiber tracking, which will be very useful for clinical researchers and neuroanatomists.

Diffusion weighted imaging-based maximum density path analysis and classification of Alzheimer's disease

Characterizing brain changes in Alzheimer's disease (AD) is important for patient prognosis and for assessing brain deterioration in clinical trials. In this diffusion weighted imaging study, we used a new fiber-tract modeling method to investigate white matter integrity in 50 elderly controls (CTL), 113 people with mild cognitive impairment, and 37 AD patients. After clustering tractography using a region-of-interest atlas, we used a shortest path graph search through each bundle's fiber density map to derive maximum density paths (MDPs), which we registered across subjects. We calculated the fractional anisotropy (FA) and mean diffusivity (MD) along all MDPs and found significant MD and FA differences between AD patients and CTL subjects, as well as MD differences between CTL and late mild cognitive impairment subjects. MD and FA were also associated with widely used clinical scores. As an MDP is a compact low-dimensional representation of white matter organization, we tested the utility of diffusion tensor imaging measures along these MDPs as features for support vector machine based classification of AD.

Structural abnormality of the corticospinal tract in major depressive disorder

BACKGROUND

Scientists are beginning to document abnormalities in white matter connectivity in major depressive disorder (MDD). Recent developments in diffusion-weighted image analyses, including tractography clustering methods, may yield improved characterization of these white matter abnormalities in MDD. In this study, we acquired diffusion-weighted imaging data from MDD participants and matched healthy controls. We analyzed these data using two tractography clustering methods: automated fiber quantification (AFQ) and the maximum density path (MDP) procedure. We used AFQ to compare fractional anisotropy (FA; an index of water diffusion) in these two groups across major white matter tracts. Subsequently, we used the MDP procedure to compare FA differences in fiber paths related to the abnormalities in major fiber tracts that were identified using AFQ.

RESULTS

FA was higher in the bilateral corticospinal tracts (CSTs) in MDD (p's < 0.002). Secondary analyses using the MDP procedure detected primarily increases in FA in the CST-related fiber paths of the bilateral posterior limbs of the internal capsule, right superior corona radiata, and the left external capsule.

CONCLUSIONS

This is the first study to implicate the CST and several related fiber pathways in MDD. These findings suggest important new hypotheses regarding the role of CST abnormalities in MDD, including in relation to explicating CST-related abnormalities to depressive symptoms and RDoC domains and constructs.

Skull-stripping with machine learning deformable organisms

BACKGROUND

Segmentation methods for medical images may not generalize well to new data sets or new tasks, hampering their utility. We attempt to remedy these issues using deformable organisms to create an easily customizable segmentation plan. We validate our framework by creating a plan to locate the brain in 3D magnetic resonance images of the head (skull-stripping).

NEW METHOD

Our method borrows ideas from artificial life to govern a set of deformable models. We use control processes such as sensing, proactive planning, reactive behavior, and knowledge representation to segment an image. The image may have landmarks and features specific to that dataset; these may be easily incorporated into the plan. In addition, we use a machine learning method to make our segmentation more accurate.

RESULTS

Our method had the least Hausdorff distance error, but included slightly less brain voxels (false negatives). It also had the lowest false positive error and performed on par to skull-stripping specific method on other metrics.

COMPARISON WITH EXISTING METHOD(S)

We tested our method on 838 T1-weighted images, evaluating results using distance and overlap error metrics based on expert gold standard segmentations. We evaluated the results before and after the learning step to quantify its benefit; we also compare our results to three other widely used methods: BSE, BET, and the Hybrid Watershed algorithm.

CONCLUSIONS

Our framework captures diverse categories of information needed for brain segmentation and will provide a foundation for tackling a wealth of segmentation problems.

Interoperable atlases of the human brain

The last two decades have seen an unprecedented development of human brain mapping approaches at various spatial and temporal scales. Together, these have provided a large fundus of information on many different aspects of the human brain including micro- and macrostructural segregation, regional specialization of function, connectivity, and temporal dynamics. Atlases are central in order to integrate such diverse information in a topographically meaningful way. It is noteworthy, that the brain mapping field has been developed along several major lines such as structure vs. function, postmortem vs. in vivo, individual features of the brain vs. population-based aspects, or slow vs. fast dynamics. In order to understand human brain organization, however, it seems inevitable that these different lines are integrated and combined into a multimodal human brain model. To this aim, we held a workshop to determine the constraints of a multi-modal human brain model that are needed to enable (i) an integration of different spatial and temporal scales and data modalities into a common reference system, and (ii) efficient data exchange and analysis. As detailed in this report, to arrive at fully interoperable atlases of the human brain will still require much work at the frontiers of data acquisition, analysis, and representation. Among them, the latter may provide the most challenging task, in particular when it comes to representing features of vastly different scales of space, time and abstraction. The potential benefits of such endeavor, however, clearly outweigh the problems, as only such kind of multi-modal human brain atlas may provide a starting point from which the complex relationships between structure, function, and connectivity may be explored.