@Article{cmes.2026.082664,
AUTHOR = {Pasquale De Luca, Livia Marcellino},
TITLE = {Physics-Informed Neural Networks for Osteosarcoma Tumor-Immune Dynamics},
JOURNAL = {Computer Modeling in Engineering \& Sciences},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/CMES/online/detail/26999},
ISSN = {1526-1506},
ABSTRACT = {Osteosarcoma is the most common primary malignant bone tumor in pediatric populations. This work presents an extended Physics-Informed Neural Network framework that incorporates interferon-gamma (IFN-<mml:math id="mml-ieqn-1"><mml:mi>γ</mml:mi></mml:math>) as a fifth biological variable, complementing previous four-variable formulations with an explicit cytokine-mediated macrophage activation pathway. The model couples five biological fields with mechanical tissue response through Biot’s poroelastic theory over a two-dimensional domain. Four distinct initial macrophage distributions were investigated. Numerical stability was achieved across all scenarios, with total loss values between 0.056 and 0.158 and mechanical residuals below <mml:math id="mml-ieqn-2"><mml:mn>3.2</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn>10</mml:mn><mml:mrow><mml:mo>−</mml:mo><mml:mn>5</mml:mn></mml:mrow></mml:msup></mml:math>. The boundary-concentrated configuration yielded the lowest biological loss. Predicted dynamics are biologically consistent, exhibiting initial immune-mediated suppression followed by progressive macrophage depletion. Comparison of the four scenarios suggests that spatial co-localization between macrophages and tumor boundaries enhances early immune-tumor contact via pressure-driven advection, while sustained immune engagement leads to measurable macrophage exhaustion. Temporal stiffness introduced by the rapid interferon-gamma decay was managed through curriculum learning and adaptive loss weighting.},
DOI = {10.32604/cmes.2026.082664}
}