Machine Learning-based Prediction of Temporal Velocity-Informatics (TVI) Variables for Accelerated Characterization...

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Fluid Dynamics

Background:

• Temporal velocity-informatics (TVI) quantifies flow disturbance in vascular aneurysms using time-resolved 3D velocity fields.
• Computational fluid dynamics (CFD) simulations for TVI are computationally intensive, limiting clinical application for intracranial aneurysms (IA).

Purpose of the Study:

• To assess the feasibility of using machine learning (ML) and IA geometrical data to predict TVI parameters.
• To evaluate the efficacy of ML-predicted TVI parameters in determining IA rupture status.

Main Methods:

• Developed ML regression models to predict TVI parameters from IA geometric information.
• Evaluated the predictive performance of ML-derived TVI parameters on a dataset of 112 IAs with known rupture status.

Main Results:

• ML-predicted TVI parameters achieved an AUC of 0.88 and 81.6% total accuracy in predicting IA rupture status.
• The consistency between ML-predicted TVI variables and CFD-derived TVI metrics surpassed that of predicting wall shear stress-based metrics.

Conclusions:

• ML-based prediction of TVI parameters from IA geometry is feasible and effective for rupture status assessment.
• This ML approach provides a computationally efficient alternative to CFD for TVI analysis in clinical settings.