Machine Learning-based Prediction of Temporal Velocity-Informatics (TVI) Variables for Accelerated Characterization
Mostafa Rezaeitaleshmahalleh1,2, Zonghan Lyu1,2, Nan Mu1,2,3
1Department of Biomedical Engineering, Michigan Technological University, 1400 Townsend Drive, Houghton, 49931, MI, USA.
Abstract:
Temporal velocity-informatics (TVI) is a novel technique utilizing spatial analysis of time-resolved 3D velocity fields to quantify flow disturbance in vascular aneurysms. Although it can improve the characterization of intracranial aneurysms' (IA) rupture status, calculation of time-resolved 3D velocity fields using computational fluid dynamics (CFD) simulations limits its clinical translation. This study aims to test the feasibility of using IA's geometrical information in conjunction with machine learning (ML)-based regression methods to predict TVI parameters. The effectiveness of these ML-predicted TVI parameters in predicting IA's rupture status was evaluated using one hundred twelve IAs with known rupture status. We found that predicting the IA's rupture status using predicted TVI can achieve an AUC of 0.88, and a total accuracy of 81.6%. Also, We found that the consistency between ML-predicted TVI variables and estimated TVI metrics calculated from CFD-simulated velocity data was higher than our ability to predict wall shear stress-based metrics.
