Deep Learning Prior Speeds Up Tumor Modeling
A new study in IEEE Transactions on Medical Imaging presents a hybrid method combining deep learning and evolutionary sampling to model glioblastoma growth more efficiently. The approach, with contributions from the Menze Lab, uses a deep learning ensemble to estimate initial tumor growth parameters, constraining the search space for evolutionary sampling. This integration achieved a fivefold acceleration in convergence while maintaining a 95% Dice-score overlap with ground truth tumor data. The method enables accurate predictions of tumor infiltration beyond MRI-visible regions, offering a robust and computationally feasible solution for individualized radiotherapy planning in glioblastoma patients.
Publication Link: https://doi.org/10.1109/TMI.2024.3494022
