The White Matter Query Language: A Novel Approach for Describing Human White Matter Anatomy

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.

💡 Research Summary

The paper introduces the White Matter Query Language (WMQL), a novel, near‑English textual framework for formally describing human white‑matter anatomy and automatically extracting the corresponding fiber tracts from diffusion‑MRI (dMRI) data. Traditional approaches to white‑matter tract identification rely heavily on expert manual labeling or on complex clustering algorithms that are difficult to reproduce and scale to large cohorts. WMQL addresses these limitations by encoding expert neuroanatomical knowledge into a set of human‑readable queries that specify the gray‑matter endpoints, intermediate white‑matter regions, and spatial relationships (e.g., “passes between region A and region B”, “contacts region C”, “avoids region D”).

The authors first constructed a dictionary of WMQL definitions for 10 association tracts (including the arcuate fasciculus, inferior fronto‑occipital fasciculus, uncinate fasciculus, etc.), 15 projection tracts (corticospinal tract, thalamic radiations, optic radiation, etc.), and 7 commissural tracts (corpus callosum, anterior commissure, etc.). Each definition combines region‑of‑interest (ROI) masks derived from standard atlases with logical operators that capture the tract’s anatomical constraints.

A matching engine was then implemented to evaluate every streamline in a whole‑brain tractogram against the WMQL definitions. The engine computes intersections, inclusions, and distance metrics between streamline points and ROI masks, allowing rapid, fully automated labeling of thousands of subjects. The pipeline is fully scripted, requiring only a pre‑processed dMRI dataset and the WMQL dictionary as inputs.

Validation was performed on a set of 20 subjects for which expert neuroanatomists manually labeled the same tracts. WMQL’s automatic labels achieved an average Dice coefficient of 0.85 (range 0.78‑0.91) and comparable mean surface distances, demonstrating that the query‑based approach matches expert performance while eliminating inter‑rater variability. Notably, WMQL performed especially well on complex commissural and crossing fibers where manual delineation is notoriously inconsistent.

To showcase scalability, the authors applied WMQL to a cohort of 77 healthy adults (age 20‑70, balanced gender). For each tract, diffusion metrics such as fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) were extracted, and a population‑based white‑matter atlas was generated. The atlas captures normative tract geometry, inter‑subject variability, and age‑related trends, providing a reference framework for future studies.

Finally, a proof‑of‑concept clinical application examined 15 patients with schizophrenia and 15 matched controls. Using WMQL‑derived tract labels, the authors identified significant FA reductions in the left arcuate fasciculus and bilateral uncinate fasciculi (p < 0.01), consistent with prior reports of disrupted frontotemporal connectivity in schizophrenia. No significant differences were observed in major projection pathways, suggesting that the disease‑related alterations are more pronounced in association fibers. This pilot demonstrates that WMQL can be deployed directly in clinical research pipelines to detect subtle white‑matter changes.

In summary, WMQL offers three major advances: (1) a reproducible, text‑based representation of white‑matter anatomy that can be shared across laboratories; (2) an automated, high‑throughput labeling engine that scales to large dMRI databases without sacrificing anatomical fidelity; and (3) a validated normative atlas and a demonstrable clinical utility in a psychiatric disorder. The authors envision extending WMQL to incorporate cerebellar pathways, subcortical loops, and multimodal imaging data (e.g., functional MRI, PET) to build a comprehensive, query‑driven connectome framework for both basic neuroscience and precision medicine.