{
  "abstract": "Introduction Machine learning (ML) is a powerful tool for clinical prediction that can improve patient outcomes. However, no current ML model reliably predicts cerebral aneurysm formation from baseline anatomy. Training such a model requires longitudinal neuroimaging data pre- and post-aneurysm formation from large patient cohorts, and is limited by the lack of a fully automated method for extracting patient-specific vascular geometries from scans. This work established an automated, end-to-end pipeline for generating high-fidelity computational fluid dynamics (CFD) data from vascular geometries derived from the circle of Willis (CW). By enabling scalable, consistent extraction of vascular geometry data from patient scans, this pipeline overcomes a key barrier that has previously limited ML model development in this space. Ultimately, this pipeline will guide future ML model development linking patient-specific anatomy to hemodynamics and enabling identification of anatomy-based predictors of aneurysm formation.Methods Neuroimaging data were used to quantify vascular geometry and generate patient-specific digital models of the cerebral arteries. Patient-specific vasculature was segmented using state-of-the-art machine learning models (nnUNetv2), followed by generation of high-quality hexahedral meshes for CFD simulations. To ensure physiological relevance, physically accurate 3D-printed vessel models were fabricated and evaluated in NAU’s Bioengineering Devices Laboratory Simulated Surgical Suite (BDL-S3). In vitro flow measurements obtained from these benchtop models were used to calibrate and validate the CFD simulations. This integrated, end-to-end pipeline enabled reproducible extraction of hemodynamic parameters, including wall shear stress, pressure, and flow rates from patient-specific anatomy. Resulting datasets were used to identify regions of elevated hemodynamic stress (‘hot spots’) associated with aneurysm-prone conditions and establish a foundation for future ML models linking vascular anatomy to hemodynamics.Results The performance of Aneumetrics was evaluated on a dataset of 125 patients with both CTA and MRA imaging (250 images total). Aneumetrics accurately reconstructed the arterial network in all cases, representing a substantial improvement over traditional algorithms that often fail to accommodate the inherent imprecision of ML-segmented images. These reconstructed vascular geometries were then incorporated into CFD simulations that are calibrated using 3D-printed CW models.Conclusion Aneumetrics enables automated generation of accurate vascular geometries and CFD-ready mesh models directly from raw MRA and CTA images. The resulting datasets provide a consistent and scalable foundation for characterizing complex hemodynamic patterns across diverse patient anatomies. This pipeline supports future development of ML models for clinical prediction of aneurysm formation.Disclosures T. Carver: None.Abstract E-234 Figure 13D printed vascular model in benchtop simulated surgical suite (left) and CFD-derived blood flow analysis (right)",
  "authors": [
    {
      "affiliations": [
        "Northern Arizona University, Flagstaff, AZ"
      ],
      "name": "T Carver"
    }
  ],
  "title": "E-234 Aneumetrics: an automated pipeline for extracting, modeling, and analyzing cerebral arteries and aneurysms",
  "uid": "128718fb-95d4-5d73-b475-3a50c839ef5c"
}