Machine learning-based software can reproduce knee joint cartilage stresses analyzed using finite element modeling during gait

  • Nematollahi, Mahan (University of Eastern Finland)
  • Kosonen, Joonas (University of Eastern Finland)
  • Kuronen, Joni (University of Eastern Finland)
  • Esrafilian, Amir (University of Eastern Finland)
  • Arokoski, Jari (University Hospital and Helsinki University)
  • Turunen, Mikael (University of Eastern Finland and Kuopio Univ)
  • Mononen, Mika (University of Eastern Finland)
  • K Korhonen, Rami (University of Eastern Finland)

Please login to view abstract download link

Introduction: Finite element (FE) modeling is widely used to estimate knee joint cartilage mechanics [1]. Machine learning (ML)-based approaches have also shown promise to assist FE modeling [2]. However, both methods still require technical expertise, which limits their translational potential. Here, we present knee joint cartilage stress results obtained by a user-friendly software that is based on ML, and compare the results with those obtained by independent FE modeling. Methods: Magnetic resonance images (MRIs) of eight healthy adults were obtained. For the software (Algoa v1.0), knee joint dimensions were first defined from the subjects’ MRIs (Fig. 1). These dimensions together with other patient data were then used as inputs in the software. The ML model of the software was originally trained by ~1000 atlas-based FE modeling results [3]. For the FE model of this study (Abaqus), tibial and femoral cartilage geometries were obtained from MRIs by using an automatic segmentation algorithm (Fig. 1), with the same mesh and material model of cartilage as presented earlier [4]. Axial joint contact force and generic knee flexion angle were used as inputs in the FE models, as they were in line with the models used to train the ML model of the software [3]. The mean maximum principal stresses over the contact area of medial tibial cartilage were compared between the two approaches. Results: Average maximum principal stresses over the stance phase estimated with the software were primarily in line with the FE predictions (Fig. 1). Conclusion: The presented user-friendly software can replicate knee joint mechanics predicted by FE modeling. The software could benefit researchers for easy stress and strain analyses of the knee joint cartilage as well as healthcare professionals for abnormal stress estimation and developing new rehabilitation strategies. References: [1] Song et al. J Biomechanics, 2023; [2] Paz et al. Ann Biomed Eng, 2024; [3] Mononen et al Scientific Report, 2023 [4] Esrafilian et al. IEEE Trans Biomed. Eng, 2025