ESR 13: Identifying Patient Specific Material Models and Parameters using Adjoint-based Inverse Modelling Approaches: Model Selection and Parameter Identification on Phantom Gels

Early Stage Rsearcher: Milad Zeraatpisheh
Host institution: University of Luxembourg (UL, LU)
Supervisor: Prof. Stéphane Bordas, UL
Co-supervisor: Frank Vogel, Managing Director, InuTech Gmbh (DE)
Clinical expertise: Dr. Frank Hertel, Centre Hospitalier de Luxembourg, LU
Further institution involved: Cardiff University (UK), and InSimo (FR)

Objectives

There exists no tool to quantitatively and systematically discriminate between constitutive biomechanics models, to identify the associated parameters and quantify uncertainty. This makes it difficult to create patient-specific material models for biomedical simulations. Based on our POC for 3D heterogeneous hyper-elastic materials, we will devise a general, adjoint-based data assimilation algorithm to identify the most suitable material model from our modeling database, and the corresponding parameters. This computational research will be performed in parallel with a simple experimental protocol on phantom gels, using digital image correlation, ultrasound and MRI. We will investigate the role of initial and boundary conditions and data sparsity on model selection and parameter identification. The fellow will drive innovation by: developing tailored model-order reduction approaches to make the statistical inversion efficient enough for clinical time scales, developing parallel mesh-scalable solution strategies, and quantifying parameter, structural and algorithmic uncertainty. Prediction of the algorithms will be tested against elastography data from existing projects at UL and AnatoScope.

Expected Results

Adjoint-based data assimilation algorithms for high-dimensional parameter estimation in geometrically nonlinear hyperelastic materials. Parameter data will be exported/coupled into the SOFA simulation framework.

Planned research stays

  • UL (7 months): Training in adjoint approaches for inverse problems.
  • Cardiff University (Mathematics Department and the Department of Engineering) (5 months): gather real data and training in hyper-elastic models with expert Angela Mihai
  • UL (6 months): continued training in adjoint approaches for inverse problems
  • InSimo (5 months): train mastering real-time data acquisition on phantom gels and set up an experiment.
  • UL (13 months): Complete validation, software deliverables and thesis.