Bayesian Aneurysm Growth Detection via Surface Displacement Modeling

Area of Science:

• Medical Imaging
• Radiology
• Computational Biology

Background:

• Clinical management of unruptured intracranial aneurysms relies on detecting growth via serial magnetic resonance angiography (MRA).
• Current methods using manual 2D diameters or volumetric changes are subjective, prone to variability, and may miss subtle growth.
• Distinguishing true aneurysm growth from imaging or processing artifacts remains a challenge.

Purpose of the Study:

• To develop and validate a novel Bayesian displacement-based model for accurate and reliable assessment of intracranial aneurysm growth from longitudinal MRA.
• To improve the objectivity and reproducibility of aneurysm growth detection compared to existing methods.
• To provide calibrated probabilistic outputs to aid clinical decision-making.

Main Methods:

• A Bayesian displacement model was developed, utilizing the surrounding vasculature as an internal reference to correct for imaging and processing variability.
• Longitudinal MRA surface data were registered, and normal-directed displacements were computed.
• Aneurysm growth was quantified by comparing mean aneurysm displacement to mean displacement of the reference vessel segment.
• The model was evaluated on two independent MRA cohorts with varying imaging sequences and rater expertise.

Main Results:

• The Bayesian displacement model demonstrated strong discrimination of aneurysm growth, achieving an Area Under the Curve (AUC) of 0.86-0.87.
• The model significantly improved agreement with expert labels (Cohen's kappa up to 0.66) compared to volumetric criteria (kappa 0.35).
• Performance remained robust across different MRA sequences and was preserved when trained on labels from less experienced raters.

Conclusions:

• The proposed Bayesian displacement model offers an interpretable, probabilistic framework for assessing intracranial aneurysm growth from longitudinal MRA.
• This method reduces dependence on subjective manual measurements and clinician expertise, enhancing diagnostic accuracy and consistency.
• Calibrated uncertainty estimates can guide clinical decisions, potentially optimizing surveillance strategies and reducing unnecessary interventions.