Source-linked AI summary
The White Matter Query Language: A Novel Approach for Describing Human White Matter Anatomy
Demian Wassermann, Makris Nikos, Yogesh Rathi, Shenton Martha, Ron Kikinis, Marek Kubicki, Carl-Fredrik Westin
TL;DR
The paper addresses the gap between textual neuroanatomical descriptions of white matter tracts and their extraction from dMRI. It introduces WMQL, a near-to-English query language interpreted to automatically dissect tracts, and reports agreement with manual delineation alongside atlas generation from 77 healthy subjects. The approach is intended to support tract-specific analysis while remaining extensible across tract definitions.
Problem
Existing tractography workflows and anatomical conventions make tract isolation, definition comparison, and reproducible tract-specific analysis difficult.
Method
WMQL encodes white matter anatomy as near-to-English textual queries that software interprets to extract tracts from dMRI tractographies.
Results
WMQL automatic extraction showed substantial agreement with manual delineation, with κ>0.76 for all five evaluated tracts.
Takeaways & Limitations
WMQL enabled an extensive white matter atlas and tract-specific analyses across association, commissural, and projection tracts.
Takeaways & Limitations
Results depend on tractography quality, including preprocessing, algorithm choice, false-positive streamlines, sparse curves, and stopping criteria.
Abstract
from arXiv · showhide
We have developed a novel method to describe human white matter anatomy using an approach that is both intuitive and simple to use, and which automatically extracts white matter tracts from diffusion MRI volumes. Further, our method simplifies the quantification and statistical analysis of white matter tracts on large diffusion MRI databases. This work reflects the careful syntactical definition of major white matter fiber tracts in the human brain based on a neuroanatomist's expert knowledge. The framework is based on a novel query language with a near-to-English textual syntax. This query language makes it possible to construct a dictionary of anatomical definitions that describe white matter tracts. The definitions include adjacent gray and white matter regions, and rules for spatial relations. This novel method makes it possible to automatically label white matter anatomy across subjects. After describing this method, we provide an example of its implementation where we encode anatomical knowledge in human white matter for 10 association and 15 projection tracts per hemisphere, along with 7 commissural tracts. Importantly, this novel method is comparable in accuracy to manual labeling. Finally, we present results applying this method to create a white matter atlas from 77 healthy subjects, and we use this atlas in a small proof-of-concept study to detect changes in association tracts that characterize schizophrenia.
1. Introduction
The paper introduces WMQL to narrow the gap between neuroanatomical descriptions of white matter tracts and their isolation from dMRI. It targets reproducible, extensible tract-specific analysis across large populations.
- Motivation: Current tractography methods often require manual placement of multiple regions of interest to isolate white matter tracts.Approaches use predefined seed and target regions or retain streamlines satisfying ROI conjunctions, disjunctions, or exclusions.
- Motivation: Tract-specific quantitative analysis can improve statistical power, sensitivity, and interpretability but requires precise, consistent delineation across subjects.This requirement is complicated because some fascicles and fascicle subdivisions remain anatomically debated.
- Motivation: Three challenges hinder reproducible tractography-based analysis: debated fascicle anatomy, incompatible atlas definitions, and semi-automated systems that are difficult to extend.The third challenge reflects the technical expertise required by fixed fascicle sets.
- Contribution: WMQL represents white matter tract anatomy as readable text that software interprets to extract described tracts from dMRI tractographies.The language is designed to be modifiable and extensible without an engineering background.
- Contribution: The framework addresses automated virtual dissection while supporting comparison and extension of tract definitions across anatomical literature.Its near-to-English syntax preserves explicit textual tract descriptions rather than encoding them only in software-specific procedures.
2. Materials and Methods
WMQL formalizes tract anatomy through region-based streamline sets, spatial relations, and logical operations, then evaluates textual queries with efficient indexing and parsing procedures.
- WMQL representation: WMQL describes tracts using anatomical terms, relative positions, and logical operations over gyri, sulci, subcortical structures, and white matter regions.These operations encode whether streamlines traverse or end in structures and whether they lie in specified spatial relations.
- WMQL representation: Each anatomical region is associated with streamline sets representing endpoints, traversal, and six relative positions.For the amygdala, these include endpoints_in, traversal, and anterior, posterior, medial, lateral, superior, and inferior relations.
- Query operations: Logical WMQL clauses are formalized as set operations, including union, exclusion, and selection of streamlines belonging only to a specified set.The formalization gives tract definitions the expressive power of set theory and propositional logic.
- Query operations: The .side and .opposite suffixes allow one assignment to define corresponding tracts in both hemispheres.A bihemispheric definition applies side-specific anatomical regions while excluding the opposite hemisphere.
- Implementation: An AABB-tree indexes more than a million streamlines to accelerate candidate-region detection before exact WMQL evaluation.The index recursively subdivides bounding boxes until each contains one streamline, balancing indexing resources with intersection testing.
- Implementation: WMQL queries are parsed into abstract syntax trees and evaluated with hash-table set representations, with each operation requiring at most linear time in set size.The grammar is specified in Backus normal form and processed using an LL(1) parser.
2.1. Generation of a WMQL Atlas of White Matter Tract Definitions
The atlas formalizes white matter tract descriptions from classic and current neuroanatomy literature, covering association, commissural, and projection systems.
- Atlas definitions: WMQL queries encode tract descriptions from neuroanatomy literature and are processed with MRI images to extract tracts automatically.The atlas construction reviews current descriptions and formalizes them as reusable queries.
- Association tracts: The association-tract definitions include the cingulum, superior longitudinal fascicle variants, arcuate fascicle, inferior longitudinal fascicle, middle longitudinal fascicle, uncinate fascicle, and extreme capsule.These descriptions specify courses, relative positions, and cortical or subcortical connections.
- Association tracts: The atlas distinguishes SLF I, SLF II, SLF III, and AF as separate fiber tracts in one literature-based classification.This separation addresses the recurring interchange of the arcuate fascicle and superior longitudinal fascicle names.
- Commissural tracts: The corpus callosum is represented as seven sections following Witelson's proposed subdivision.Diffusion imaging permits in vivo delineation of a considerable portion of callosal fibers.
- Projection tracts: Projection-tract descriptions cover the cortico-spinal, cortico-thalamical, and cortico-striatal systems.The cortico-spinal tract is described as a prominent descending fiber system, while the other systems connect cortex with thalamic nuclei or striatum.
2.2. Data Acquisition and Preprocessing
The study acquired diffusion-weighted and T1 MRI data from healthy and schizophrenia groups using a 3.0T scanner and corrected diffusion data for eddy currents and head motion.
- Participants: Diffusion-weighted images were acquired from 77 healthy subjects and 20 age-matched male right-handed schizophrenia subjects.Healthy subjects averaged 32.9 ± 12.4 years of age; 64 were male and all were right-handed.
- MRI acquisition: DWI acquisition used a GE SignaHDxT 3.0T scanner with 51 diffusion directions, b=900 s/mm2, 8 b=0 images, and 1.7 mm isotropic voxels.A T1 MRI acquisition was also performed.
- Preprocessing: Diffusion data were corrected for eddy currents and head motion using FSL.The preprocessing was applied before subsequent tractography-based analyses.
2.3. Experiments
The experiments evaluated WMQL against manual tract delineation, generated a volumetric atlas from 57 tracts across 77 healthy subjects, and applied the atlas in a schizophrenia proof-of-concept analysis.
- Validation: WMQL definitions were tested against manual segmentation based on an independent ROI protocol applied to diffusion MRI tractography.The manual protocol used subject-specific FA-image ROIs, whereas WMQL used subject-independent anatomical definitions.
- Volumetric atlas: The experiment measured across-subject tract overlap and generated a volumetric atlas for reference or population studies.
- Volumetric atlas: WMQL combined with multi-tensor tractography generalized previous atlases and added tracts such as the MdLF and three SLF subdivisions.
- Volumetric atlas: 57 white matter tracts were extracted for each healthy subject using WMQL definitions to construct the volumetric tract atlas.The atlas pipeline also normalized FA maps to MNI space and computed voxel-wise group effect maps from tract visitation maps.
- Schizophrenia analysis: A proof-of-concept analysis compared selected WMQL-atlas tracts between 20 schizophrenia subjects and 20 age- and gender-matched healthy controls.The analysis focused on MdLF and SLF I, II, and III, which were not included in other atlases described in the passage.
3. Results
WMQL tract extraction agreed substantially with manual delineation and produced consistent across-subject tract maps, while selected atlas-only association tracts showed group differences in fractional anisotropy.
- Validation: κ>0.76 across five tracts indicated substantial agreement between WMQL extraction and manual delineation by two experts.The five tracts were IOFF, UF, ILF, CST, and AF, evaluated in 10 subjects.
- Validation: κ=0.76 for the right AF and κ=0.90 for the left ILF marked the worst and best inter-rater agreements, respectively.
- Validation: κ>0.7 for all tracts showed substantial overlap between manually segmented tracts and WMQL-extracted tracts.
- Validation: κ=0.71 for the left AF and κ=0.89 for the left CST were the lowest and highest overlaps between manual segmentation and WMQL extraction.
- Validation: The WMQL scores were slightly lower than inter-rater scores, with a mean difference of 0.033±0.020.The largest difference was 0.08 for the left AF, while the smallest was 0.01 for the left CST and IOFF.
- Volumetric atlas: 57 tracts extracted from 2 million whole-brain streamlines in less than 3:40 minutes per subject showed high across-subject overlap in MNI space.The overlap was significant at p-value <0.0001 after correction for multiple comparisons, and the resulting surfaces formed a healthy-subject volumetric atlas.
- Schizophrenia analysis: Bilateral FA decreases were found in the MdLF, SLF II, and SLF III in the schizophrenia group comparison.Reported p-values were .010 and 0.17 for right and left MdLF, 0.009 and 0.015 for right and left SLF II, and 0.017 and 0.018 for right and left SLF III.
4. Discussion
WMQL improves reproducibility and flexibility in tract-specific analysis by formalizing anatomical definitions, while its outputs remain constrained by operator expertise, tractography quality, and biological and methodological variability.
- Advantages: WMQL addresses reproducibility problems by translating anatomical tract descriptions into readable queries that can be modified and extended without engineering expertise.The approach also supports tract definitions that change or were not catalogued in the original system.
- Advantages: WMQL can automatically extract newly described or previously uncatalogued tracts from dMRI images and remains independent of the atlas or brain parcellation choice.This flexibility supports interactive and batch use across study-specific settings.
- Limitations: WMQL results depend on tractography quality, preprocessing, algorithm choice, and operator proficiency, while streamline terminations remain anatomically debatable.False positives can make fascicles noisier, whereas too few curves can make them sparse.
- Validation: κ>0.7 was observed for all five evaluated tracts, with WMQL overlap close to but always lower than inter-rater overlap.The worst consistency was for the left AF and the best was for the left CST.
- Atlas and population findings: 77 healthy subjects showed variability in tract locations and trajectories despite every reported tract being present in all subjects.Higher variability was found for the right MdLF and EC, the left SLF I, and cortico-thalamic tracts.
- Atlas and population findings: The schizophrenia pilot detected characteristic tract differences in SLF I, II, III, and MdLF, including tracts absent from other available atlases.The study computed one FA value per tract per subject from group effect maps.
5. Conclusion
WMQL represents anatomical knowledge of white matter tracts and extracts them from full-brain tractographies, supporting tract-specific studies. The paper demonstrates this utility through a population atlas and a schizophrenia pilot study.
- WMQL represents anatomical knowledge of white matter tracts and extracts them from full-brain tractographies.
- Tract-specific analyses can increase sensitivity and specificity relative to full-brain methodologies, improving interpretability and reducing the sample size needed for significant results.
- The authors demonstrate WMQL by producing an extensive atlas containing association, commissural, and projection tracts from a healthy population.
- A pilot study on a schizophrenic sample illustrates tract-specific analyses using WMQL.
- WMQL descriptions make tract extraction more transparent and flexible, including for tracts not catalogued in the study.
Ethical Approval
The paper states that formal consent is not required for this type of study.
- Formal consent is not required for this type of study.