Finite Element Validation Using Digital Volume Correlation in FEM3 – A High-Performance, Memory Efficient Micro-FE Solver for Large Scale Heterogeneous Image-Based Models on Consumer Hardware
Speakers:
Adam Gorski (University of Waterloo)
Jonathan Kusins (University of Waterloo)
Louis Ferreira (Western University)
Nikolas Knowles (University of Waterloo)
Image-based finite element (FE) modelling has become an essential tool for investigating the mechanical behaviour of biological tissues and engineered materials. However, validating large-scale heterogeneous models remains computationally demanding and often requires access to high-performance computing resources. This workshop will introduce FEM3, a high-performance, memory-efficient micro-FE solver designed to enable simulation of large image-based models on standard consumer hardware. Participants will learn how Digital Volume Correlation (DVC) can be used to generate full-field experimental displacement and strain measurements for quantitative validation of FE predictions. Through a combination of lectures and practical demonstrations, attendees will explore workflows for image acquisition, model generation, DVC analysis, FE simulation, and validation. Case studies from musculoskeletal biomechanics and bone mechanics will illustrate best practices for comparing experimental and computational results, assessing model accuracy, and identifying sources of uncertainty in image-based finite element analyses.
Introduction to the Neuromusculoskeletal Modeling Pipeline: From Code to Clinic
Speaker:
Prof. B.J. Fregly (Rice University)
Neuromusculoskeletal computer modeling has reached the point of potential clinical utility. However, a remaining barrier is the lack of easy-to-use computational software that allows researchers working with clinicians to 1) personalize a neuromusculoskeletal computer model using an individual patient’s movement data, and then 2) design the best treatment for the patient by optimizing the function of the patient’s personalized model. The Matlab-based open-source Neuromusculoskeletal Modeling (NMSM) Pipeline software seeks to fill this void by adding Model Personalization and Treatment Optimization toolsets to OpenSim. The Model Personalization toolset uses the patient’s pre-treatment movement data to personalize joint, muscle-tendon, neural control, and foot-ground contact properties, while the Treatment Optimization toolset uses the patient’s personalized model to design an optimal clinical treatment via predictive simulations. The workshop will discuss the design of the NMSM Pipeline, review the software’s functionality, and provide an interactive hands-on training tutorial (no coding required!) involving both toolsets.
Learning and using Python for Human Movement Biomechanics using Kinetics Toolkit
Speakers:
Félix Chénier (Université du Québec à Montréal)
Samuel Hybois (Université Paris Saclay)
This workshop introduces Comodo.jl, an open source Julia package for computational (bio)mechanics and computational design. Comodo offers functionality for geometry processing, meshing, automated design, image-based modelling, and finite element analysis. Comodo.jl started out as a modern re-implementation in Julia of the MATLAB toolbox GIBBON. MATLAB is an interpreted language and hence does not scale well for high performance applications. In the context of biomedical engineering such applications may include medical device design optimisation, in-silico clinical trials, and clinical software deployment. To address this gap we here introduce Comodo, which loosely stands for Computational MOdelling for Design Optimisation. Comodo is implemented in Julia, which is a high performance and open source scientific programming language. Although its syntax is similar to MATLAB or Python its performance can match Fortran or C++.
FEBio Workshop
Speakers:
Jeffrey A. Weiss (University of Utah)
Gerard A. Ateshian (Columbia University)
Steve A. Maas (University of Utah)
The FEBio workshop will provide an overview of the FEBio project, an introduction to FEBio Studio, a demonstration of exciting new features in the 3.0 release of FEBio Studio, and an overview of recent developments in FEBio. We plan to demonstrate the new FEBio FUSE plugin, the plugin repository, the febcode domain-specific language for extending FEBio, the FEBio Monitor for monitoring and controlling execution of FEBio jobs, new capabilities for working with Python scripting, and support for automatic differentiation within FEBio and FEBio Studio. We will reserve 45 minutes at the end of the workshop to answer questions from users and to discuss specific modeling challenges with the FEBio developers. Participants are encouraged to bring their models and questions to the workshop.
Models for the Prediction of In Vivo Thrombus Composition
Speakers:
Frank Gijsen (TU Delft)
Alfons Hoekstra (University of Amsterdam)
There is an enigmatic observation in several clinical studies regarding the distribution of components in human thrombi retrieved from acute ischemic stroke patients. The distribution, at least as far as the balance between red blood cells and fibrin and platelets is nearly flat (see figure)1,2,3,4. This is -of course- a rather unusual but very consistent finding. Unusual, because generally it is assumed that thrombi from acute ischemic stroke patients either stem from the heart, from the carotid artery, or form an ‘other possible source’. One would therefor expect a bi- or tri-modal distribution in the thrombus composition. How can this flat distribution be explained? We propose a workshop in which we want to present results of the 4 most likely explanations for this finding. Any modelling approach can be used. We will have presentations from contributors to the workshop1, followed by a 30 minutes consensus discussion. If successful, a scientific paper can be the output of this workshop.
Automated Model Discovery with Constitutive Neural Networks
Speakers:
Uzair Manack (TU Delft / KU Leuven)
Rogier Krijnen (TU Delft)Â
Mathias Peirlinck (TU Delft)
Constitutive modelling remains a fundamental challenge in biomedical engineering, with researchers often having to choose between competing material models and manually calibrate their parameters. This hands-on workshop introduces constitutive neural networks, a physics-informed machine learning framework that combines classical constitutive modelling with modern neural network techniques.
Participants will learn how to discover constitutive laws directly from experimental data while enforcing thermodynamic and physical constraints. Using interactive Jupyter notebooks, attendees will implement and train constitutive neural networks on benchmark datasets from soft biological tissues and explore how automated model discovery can identify appropriate constitutive models and parameters from data. The workshop bridges biomechanics, machine learning, and computational modelling, and provides practical tools that participants can apply to their own research problems.
LumpedFlux: Hands-on 0D Biofluids Modelling
Speakers:
Irene Vignon-Clementel
Sylvain Freud
Ramdane Bessaid
Friederike Schäfer
Aseem Pradhan
(Inria Saclay ÃŽle-de-France)
This workshop introduces LumpedFlux, an open-source library for zero-dimensional modeling, that can be readily coupled to other solvers (https://codeberg.org/twin4life-team-inria/lumpedflux). Users can assemble closed-loop systems from a library of elementary building blocks, such as resistances, capacitances, and inductances, seamlessly define and integrate arbitrary custom equations, and solve complex dynamical systems using LumpedFlux’s robust solver library. Models can be coupled with user-defined models or external solvers, and can be easily exchanged, reused, and combined across research groups. LumpedFlux offers a user-friendly, high-level Python interface, a C++ library, and a graphical user interface also in development, to broaden accessibility for clinicians and modelers alike. A built-in analysis toolkit supports sensitivity analysis, calibration, identifiability analysis, and uncertainty quantification, all within a single environment. The session combines a presentation of LumpedFlux’s design philosophy with a hands-on tutorial in which participants build and analyze a simple cardiovascular model, providing a foundation for developing their own models.
Creating Combined Multibody/Finite Element Models in ArtiSynth from Existing OpenSim Models
Speakers:
John Lloyd (University of British Columbia)
Benedikt Sagl (Medical University of Vienna)
Robin Remus (Ruhr University Bochum)
Eva Herbst (Imperial College)
Alexander Graf (University of Duisburg)
ArtiSynth (www.artisynth.org) is a free, open-source, biomechanics modelling platform that supports the combined simulation of multibody and finite element (FE) models in a single application, removing the need to interface between separate multibody and FE programs. It is well-suited for embedding one or more FE components within a larger multibody musculoskeletal model to run forward and inverse dynamic simulations and has been used in a wide range of applications involving the spine, lower limb, foot, arm, shoulder, and head and neck regions.
To facilitate model creation, we have implemented a feature that allows ArtiSynth to import OpenSim models, allowing users to jumpstart development by leveraging the large OpenSim model base. Imported models can then be edited to add FE models within a specific region of interest.
This workshop will provide an overview of ArtiSynth, together with a hands-on coding exercise showing how to create and run an OpenSim based model.
Comodo: Open Source Software for Computational Biomechanics and Design
Speaker:
Dr. Kevin Moerman (University of Galway)
This workshop provides an introduction to Comodo.jl, an open source Julia package for computational (bio)mechanics and computational design. Comodo offers functionality for geometry processing, meshing, automated design, image-based modelling, and finite element analysis. Comodo.jl started out as a modern re-implementation in Julia of the MATLAB toolbox GIBBON. MATLAB is an interpreted language and hence does not scale well for high performance applications. In the context of biomedical engineering such applications may include medical device design optimisation, in-silico clinical trials, and clinical software deployment. To address this gap we here introduce Comodo, which loosely stands for Computational MOdelling for Design Optimisation. Comodo is implemented in Julia, which is a high performance and open source scientific programming language. Although its syntax is similar to MATLAB or Python its performance can match Fortran or C++.
The workshop will treat Comodo based geometry processing and meshing as well as image segmentation (using Imago.jl) and coupling with FEBio.jl for finite element analysis.
Building Trustworthy Statistical Shape Models: Hands-on Coding with Scalismo
Speaker:
Vahid Arbabi, PhD (Utrecht University)
Constitutive modelling remains a fundamental challenge in biomedical engineering, with researchers often having to choose between competing material models and manually calibrate their parameters. This hands-on workshop introduces constitutive neural networks, a physics-informed machine learning framework that combines classical constitutive modelling with modern neural network techniques.
Participants will learn how to discover constitutive laws directly from experimental data while enforcing thermodynamic and physical constraints. Using interactive Jupyter notebooks, attendees will implement and train constitutive neural networks on benchmark datasets from soft biological tissues and explore how automated model discovery can identify appropriate constitutive models and parameters from data. The workshop bridges biomechanics, machine learning, and computational modelling, and provides practical tools that participants can apply to their own research problems.