/FEOpt

Optimization framework for automated patient-specific tuning of cardiovascular finite element models

Primary LanguagePythonMIT LicenseMIT

FEOpt

FEOpt is an optimization framework for the tuning of patient/subject-specific msterial parameters with finite element models. It is implemented in Python, Bash, and Matlab and uses Continuity for model evaluations.

Meshes of patient/subject heart scans are passively inflated to the measured pressure. Since the scans are of a loaded (inflated) ventricle, an iterative process is used to fetermine the unloaded geometry. The quality of differrent material parameter values is assessed via fitment with the Klotz curve.

Features

This project contains a number of features that support the automation of model tuning:

  • Seperate submission scripts manage optimization and evaluation. Status and result updates are passed via worker-specific folders.
  • The iterative process - consisting of a variable number of inflation and deflation steps - is handled via a Bash script. Stopping criteria are also implemented.
  • Model parameters to be tuned are listed in a struct. The initialization of reference values and scaling is automatically handled.
  • Models are evaluated in parallel during the optimization. Additionally, each model is evaluated on up to 8 cores; more cores don't provide significant runtime improvement.
  • Results are logged and queried to prevent duplicate model evaluations.

Dependencies

The core of the cardiovascular finite element model is Continuity, which can be downloaded directly from their website.

This program also uses solvers from the Matlab Optimization Toolbox.

Installation

  1. Clone the FEOpt repository
  2. Download and install Continuity
  3. Set correct home and Continuity directories in Python, Bash, and Matlab files

Using FEOpt

  1. Submission of 'evaluate.script' and 'optimize.script' to a SLURM scheduler will tune the material parameters for the subject in that folder
  2. Running 'clear' manually resets the folder for a new optimization run
  3. 'PaperInflatePlot.m' can be run to produce the same plot style in the paper

File Organization

  • Each folder contains the files for one of four canine subjects
  • '...cont6': Contain the geometric ventricle models and Continuity settings for each canine subject
  • '...xls': Contains the node connectivity for a given subject
  • 'optimize.script': SLURM job submission script for the optimizer
  • 'InflationOpt...m': Matlab optimization manager and solver
  • 'Inflate...vec.m': Vectorized model evaluation - sets flags and logs results of Continuity model
  • 'evaluate.script': SLURM job submission script for the Continuity workers
  • 'runCase.script': Bash script managing a specific Continuity evaluation
  • 'Full_...py': Python script for inflation and deflation of the ventricle

Publications

  1. Joshua Mineroff, Andrew D. McCulloch, David Krummen, Baskar Ganapathysubramanian, and Adarsh Krishnamurthy (Accepted 2019). Optimization Framework for Patient-Specific Cardiac Modeling. Cardiovascular Engineering and Technology
  2. Joshua Mineroff (2018). An Optimization and Uncertainty Quantification Framework for Patient-Specific Cardiac Modeling. Graduate Theses and Dissertations

Licensing

This code is licensed under MIT license.