Introduction

This package is intended to provide a simple way to post-process CFD simulation data for medical devices with the objective of predicting hemolysis, i.e., red blood cell damage. In particular, it supports the following functionalities:

  1. Reading in CFD results

  2. Computing velocity gradients

  3. Determining trajectories (pathlines) of red blood cells (RBCs)

  4. Interpolating arbitrary fields to the RBC pathlines, e.g., the velocity gradient

  5. Computing various stress-based and strain-based measures of RBC strain

  6. Integrating various empirical hemolysis correlations along the RBC pathlines using the aforementioned measures

The package is written in Python and uses VTK to process the CFD data. As a result, the package can import CFD data from any solver that can export to VTK. For all other solvers, the package supports the import of pathline data from a CSV file. The only requirement is that velocity gradient data is available along the pathlines. Then the functionalities 5 and 6 are available.

I am currently using the package to post-process Eulerian flow data from our in-house finite-element solver XNS. I am further planning on extending compatibility to .vtu files (COMSOL) and .cgns files (ANSYS Fluent, Star-CCM). If you are interested in using the package with a different solver, please let me know and I will try to accommodate your needs.

The package is meant for steady flow data. Rotational flows such as those present in most blood pumps are meant to be handled using the MRF approach to enable a steady solution. Transient flow data needs to be averaged in time to obtain quasi-MRF data before being processed by this package. Please contact me if you need help with the workflow for unsteady data.

How to cite

More information on the underlying theory can be found in Dirkes et al. [4]. If you use this package in your work, please reference this paper.

Corrigendum

There is a small mistake in eq. (15) of the above paper. The coefficients \(a\) and \(b\) should be defined the other way around, i.e.,

\[a(\boldsymbol{\Lambda}, \mathbf{E}) = \frac{f_2}{2f_3} \frac{\lambda_1 + \lambda_2}{\lambda_1 - \lambda_2} (E_{22} - E_{11}) \, , \qquad b(\boldsymbol{\Lambda}, \mathbf{E}) = \frac{f_2}{f_3} \frac{\lambda_1 + \lambda_2}{\lambda_1 - \lambda_2} E_{12} \, .\]

In the underlying code used in the paper as well as in HemTracer, this has always been implemented correctly. Any results generated previously as well as those presented in the paper are thus not affected. This discrepancy only has to be considered if one tries to implement the model from scratch using the equations given in the paper.