Three-Dimensional Time-resolved Cardiovascular Magnetic Resonance Imaging based Particle Tracing: Recommendations

Abstract

IntroductionCardiovascular magnetic resonance imaging (CMR) is an imaging modality from which the structure, function, perfusion, and metabolism of the cardiovascular system can be evaluated, which is essential in acquired and congenital cardiovascular disease (CVD). CMR can improve the outcomes of CVD during long term follow-up. 4D Flow MRI is a three-dimensional time-resolved CMR technique that uses phase contrast sequences with three-directional velocity-encoding and allows evaluation of the blood flow. 4D Flow MRI data can be analyzed with particle tracing.Particle tracing is a technique that releases virtual particles in flow data (seeding) and traces the path the particles follow throughout the vasculature by integrating the flow data. The particle traces can subsequently be quantified with flow distribution analyses and visualized with pathlines. Important settings for particle tracing are the integration method, integration timestep and data interpolation method. The aim of this research is to investigate the effect of different particle tracing settings on the accuracy and computation time and provide a recommendation on the use of particle tracing in 4D Flow MRI data and which particle tracing settings to aim for.MethodsA particle tracing algorithm with a graphical user interface was developed for an existing inhouse developed 4D Flow post-processing workspace. Two 4D Flow MRI datasets from previous studies were acquired. One dataset was of a total cavopulmonary connection of a patient with Fontan circulation and the other was an intracardiac dataset of a healthy volunteer. Particle tracing with common settings and variations in integration method, integration timestep and data interpolation method was performed in both datasets and computation time, flow distributions (pulmonic distribution and four-component analysis), and number of particles that left the structure (missing particles fraction) and number of integration errors were quantified. The latter three were a measure of accuracy of the algorithm. Particle traces were subsequently visualized with pathlines.ResultsParticle with common settings gave a good accuracy for both datasets (missing particles fraction <20%). No differences in visualization, flow distributions and missing particles fraction were found for varying integration methods, interpolation methods and timesteps of 20ms or shorter. For timesteps of 30ms or longer, flow distributions changed compared to results with other settings and number of integration errors increased. Computation time increased almost linearly with the order of the integration method and almost inverted linearly proportional to the timestep.DiscussionFor 4D Flow MRI based particle tracing we would recommend using a low order interpolator (e.g. a third order Runge-Kutta integrator), a maximum timestep of half the temporal resolution of the data, and quadrilinear data interpolation. The results of the particle tracing tests with common settings were comparable to literature, except for the direct flow and retained inflow components of the four-component analysis. To enable accurate particle tracing, the particle seeding method must be chosen adequately and the MRI data and segmentations must be of good quality. Additional techniques such as segmentation shape interpolation and adaptive timestep control might be beneficial and their effect on the accuracy can be investigated in further research.