Glioblastoma (GBM) ranks among the most lethal of all human cancers (14.6 months survival rate). Typical treatment for GBM occurs in two steps: (1) tumor resection (removal of the tumor mass), and (2) radiation therapy (applying radiation to the margin to kill residual tumor cells). Advances of sensing and computer technologies have produced immense amounts of data that can be used to improve glioblastoma treatment outcomes, where some data (medical imaging) is more available than other data (tumor biopsies). Biopsies provide a direct measure of tumor cell density, but due to their invasiveness, there are a limited number available. Magnetic resonance images (MRI) of the patient's brain can only provide an indirect measure of tumor cell density but are noninvasive and already integrated with GBM treatment. Additionally, there is opportunity to use the proliferation-invasion (PI) mechanistic model, which utilizes pre-existing biological knowledge of (1) invasion of cells into nearby tissue and (2) proliferation of tumor cells.
The proposed method, ML-PI, integrates machine learning (ML) trained on empirical information of the MRI as well as biological knowledge of the PI mechanistic model to generate an accurate tumor cell density prediction map (ML-PI). Imaging data and biopsy measurements are integrated, while data-driven semi-supervised learning and mechanistic models are fused to maximally utilize all available information.
ML-PI's quantification of tumor cell density facilities precision tumor resection and precision radiotherapy for each GBM tumor, which can result in higher quality patient treatment and improved clinical outcomes.
Gaw, N., Hawkins-Daarud, A., Hu, L. S., Yoon, H., Wang, L., Xu, Y., ... & Gonzales, A. (2019). Integration of machine learning and mechanistic models accurately predicts variation in cell density of glioblastoma using multiparametric MRI. Scientific reports, 9(1), 1-9. https://doi.org/10.1038/s41598-019-46296-4
