High-speed predictions of post-burn contraction using a neural network trained on 2D-finite element simulations

Ginger Egberts*, Fred Vermolen, Paul van Zuijlen

*Corresponding author for this work

Research output: Contribution to journalArticleScientificpeer-review

24 Downloads (Pure)

Abstract

Severe burn injuries often lead to skin contraction, leading to stresses in and around the damaged skin region. If this contraction leads to impaired joint mobility, one speaks of contracture. To optimize treatment, a mathematical model, that is based on finite element methods, is developed. Since the finite element-based simulation of skin contraction can be expensive from a computational point of view, we use machine learning to replace these simulations such that we have a cheap alternative. The current study deals with a feed-forward neural network that we trained with 2D finite element simulations based on morphoelasticity. We focus on the evolution of the scar shape, wound area, and total strain energy, a measure of discomfort, over time. The results show average goodness of fit (R2) of 0.9979 and a tremendous speedup of 1815000X. Further, we illustrate the applicability of the neural network in an online medical app that takes the patient's age into account.
Original languageEnglish
Article number1098242
Pages (from-to)1-9
Number of pages9
JournalFrontiers in Applied Mathematics and Statistics
Volume9
DOIs
Publication statusPublished - 2023

Keywords

  • machine learning
  • post-burn scar contraction
  • morphoelasticity
  • eed–forward neural network
  • online application
  • Monte Carlo simulations

Fingerprint

Dive into the research topics of 'High-speed predictions of post-burn contraction using a neural network trained on 2D-finite element simulations'. Together they form a unique fingerprint.

Cite this