Computational fluid dynamics methods applied to intracranial stenosis imaging

Abstract

Background Stenosis of the major intracranial arteries is an important cause of stroke, accounting for 8% to 10% of ischemic strokes in US. It was demonstrated that Computational Fluid Dynamics (CFD) modelling may be a promising technique in evaluating and understanding the hemodynamic characteristics of ICAS lesions. Previously, it has been shown that creating CFD models from routinely acquired computed tomographic angiography (CTA) and digital subtraction angiography (DSA) source images is feasible. However, it has not been evaluated if the results from CFD analysis of CTA and DSA source images correlate. Methods This retrospective study evaluated 80 patients with symptomatic ICAS lesions treated at Tiantan Hospital in Beijing. All patients had a symptomatic stenosis of the M1 segment of the middle cerebral artery (MCA-M1) with DSA and CTA imaging acquired at the time of treatment. The Vascular Modelling Toolkit software was used to generate 3D models from CTA source images. A region of interest containing the stenotic vessel was manually selected on the CTA source images, and then segmented using a gradient-driven level set method to generate an initial vessel volume. 3D reconstructions were generated from DSA images using Coronary3D software. The reconstructions were generated by using an anteroposterior and a lateral view of the stenotic vessel. Blood flow simulation was performed on both DSA and CTA generated meshes using ANSYS FLUENT SOLVER (ANSYS Inc.). A two-sample t test for paired data was used to compare the pressure and velocity measurements across imaging modalities. Results CTA and DSA imaging were evaluated for 20 patients with symptomatic ICAS. Although 3D reconstruction was possible for all 80 CTA scans, faint signals and convoluted vasculature in 60 of the 80 DSA scans prevented 3D reconstruction and excluded these cases from the study. When it was plotted the change in pressures of CTA versus the change in pressures of DSA, we received a R2 of 0.001; meaning that the linear model explains almost none of the variability in between the data in terms of changes in velocity and pressure. Conclusion It was found no noticeable correlation between the information regarding pressure and velocity as given by DSA and by CTA. Therefore, we could not say that the information from DSA is on par with that of CTA. Improvements in software that can robustly reconstruct DSA scans of brain vasculature is needed to computationally analyse a larger sample size of DSA data.