Digital 3D Brain MRI Arterial Territories Atlas

Chin-Fu Liu, Johnny Hsu, Xin Xu, Ganghyun Kim, Shannon M. Sheppard, Erin L. Meier, Michael I. Miller, Argye E. Hillis, Andréia V. FariabioRxiv 2021 · DOI

TL;DR

Researchers have developed the first publicly available, 3D, deformable digital atlas of brain arterial territories using MRI data from over 1,200 stroke patients. This atlas aims to standardize and automate the analysis of brain damage in stroke patients, offering a valuable resource for clinical and research applications.

What Problem Does This Paper Address?

Accurately pinpointing the location and extent of brain damage caused by a stroke is crucial for effective treatment planning and prognosis. While vascular atlases can help with this, existing digital atlases are often not readily available for automated analysis, lack detailed sub-territory definitions, or are not flexible enough for individual patient data. This gap hinders large-scale, reproducible research and clinical decision-making. The authors aim to fill this void by creating a comprehensive, digital, and deformable atlas of brain arterial territories.

Methods

The core of this work involves creating probabilistic maps of arterial territories based on lesion data from stroke patients. The process includes:

  • Lesion Delineation: Experienced evaluators manually traced stroke lesions on diffusion-weighted MRI (DWI) scans. This was followed by neuroradiologist review for consensus.
  • Image Mapping: Lesion masks were transformed into a standardized MNI space using a combination of linear and non-linear deformation techniques (specifically, the "Broccoli" software). This step accounts for variations in brain shape and size.
  • Probabilistic Map Generation:
    • Average Method: Binary lesion masks were summed and divided by the sample size for each arterial territory to create probability maps. This method provides a basic estimate of voxel likelihood for each territory.
  • Bernoulli Mixture Model (BMM): A more advanced approach was used to capture voxel-level correlations and spatial features within lesion clusters. This method models lesion maps as a mixture of clusters, allowing for more nuanced probability estimations.
  • Territory Definition: The boundaries of arterial territories and their sub-territories were defined by analyzing the ratios of these probabilistic maps and, in some cases, by incorporating anatomical knowledge.

Datasets

The study utilized a dataset of 1,298 diffusion-weighted MRI (DWI) scans from patients admitted to the Johns Hopkins Stroke Center between 2009 and 2019, who had experienced acute or early subacute ischemic strokes affecting exclusively one major arterial territory.

Key Insights and Ideas

  • Leveraging Stroke Data for Atlas Creation: Instead of relying solely on healthy brain atlases, this study uses actual stroke lesion data to define arterial territories. This approach is more representative of how these territories are affected in stroke patients.
  • Hierarchical Atlas Structure: The atlas is built in two hierarchical levels, defining both major arterial territories (ACA, MCA, PCA, VB) and their sub-territories (e.g., thalamoperforating, lenticulostriate), providing a more detailed anatomical representation.
  • BMM for Improved Probability Mapping: The Bernoulli Mixture Model (BMM) approach demonstrated superior performance in generating probabilistic maps compared to simple averaging, particularly in capturing more uniformly distributed probabilities and better representing peripheral areas.
  • Addressing "Border Zones": The study explicitly addresses the challenge of defining "border zones" between arterial territories by using probability ratios and the BMM "territory-claimed" maps, leading to sharper and more defined boundaries.
  • Public Availability is Key: The decision to make the atlas publicly available is a significant contribution, aiming to foster widespread adoption and further research.

Key Results

  • A digital, 3D, deformable atlas of brain arterial territories has been created, covering four major territories and their sub-territories in two hierarchical levels.
  • The atlas was developed using 1,298 patient MRIs, representing a substantial dataset for this type of atlas.
  • The BMM method resulted in more uniformly distributed and higher probability maps compared to the average method, especially for the MCA territory.
  • The atlas boundaries were validated against existing anatomical atlases and digital reconstructions of vasculature, showing good agreement.
  • The atlas is available in Nifti format and includes detailed file descriptions for ease of use.

Limitations and Future Work

  • Regional Bias: The study acknowledges a potential regional bias due to the data originating from a single "Comprehensive Stroke Center," and hopes that future data contributions and user feedback will help identify and correct such biases.
  • Exclusion Criteria: Strokes classified as "watershed," cardioembolic, or multifocal were excluded. While this helped focus on large vascular areas, it means these specific stroke types are not directly represented in the atlas's derivation.
  • ACA Sample Size: The relatively smaller sample size for ACA strokes may have affected the precision of its territory definition and border zone delineation. The authors suggest that future inclusion of more ACA stroke cases will improve this.
  • Future Work: The authors plan to share individual images and lesion maps to enable researchers to create their own specialized atlases. They also aim to redefine less-stroke-prone territory boundaries using probabilistic maps as more data becomes available.

Summary updated

Abstract

ABSTRACT The locus and extent of brain damage in the event of vascular insult can be quantitatively established quickly and easily with vascular atlases. Although highly anticipated by clinicians and clinical researchers, no digital MRI arterial atlas is readily available for automated data analyses. We created a digital arterial territory atlas based on lesion distributions in 1,298 patients with acute stroke. The lesions were manually traced in the diffusion-weighted MRIs, binary stroke masks were mapped to a common space, probability maps of lesions were generated and the boundaries for each arterial territory was defined based on the ratio between probabilistic maps. The atlas contains the definition of four major supra- and infra-tentorial arterial territories: Anterior, Middle, Posterior Cerebral Arteries and Vertebro-Basilar, and sub-territories (thalamoperforating, lenticulostriate, basilar and cerebellar arterial territories), in two hierarchical levels. This study provides the first publicly-available, digital, 3D deformable atlas of arterial brain territories, which may serve as a valuable resource for large-scale, reproducible processing and analysis of brain MRIs of patients with stroke and other conditions.
BibTeX
@misc{liu2021digital,
  author = {Chin-Fu Liu and Johnny Hsu and Xin Xu and Ganghyun Kim and Shannon M. Sheppard and Erin L. Meier and Michael I. Miller and Argye E. Hillis and Andréia V. Faria},
  title = {{Digital 3D Brain MRI Arterial Territories Atlas}},
  year = {2021},
  doi = {10.1101/2021.05.03.442478},
  howpublished = {bioRxiv},
}
Figure