Adaptive Tumor Growth Forecasting via Neural and Universal ODEs

Learn how to enhance traditional tumor growth prediction models using Scientific Machine Learning techniques in this conference talk that demonstrates the application of Neural Ordinary Differential Equations (Neural ODEs) and Universal Differential Equations (UDEs) to overcome limitations of conventional mathematical models. Discover how traditional ODE-based models like Gompertz and Bertalanffy equations, while providing theoretical foundations, often fail to capture the complexity of real-world tumor progression influenced by factors such as microenvironment, chemotherapy, genetics, and immune response. Explore the implementation of a SciML approach using Julia packages including DifferentialEquations.jl for numerical ODE solving, DiffEqFlux.jl for neural network integration, Lux.jl for neural architecture construction, and Optimization.jl with OptimizationOptimisers.jl for parameter optimization using gradient-based methods like Adam and BFGS. Understand how UDE models replace fixed growth parameters with neural networks, enabling better adaptation to experimental data and improved prediction accuracy compared to traditional models. Examine the potential for personalized medical treatment through tumor growth predictions tailored to individual clinical histories, and learn how this data-driven approach can optimize treatment planning and enhance clinical decision-making in oncology while advancing precision medicine applications.