Digital Twins for Advancing Nanoparticle Mediated Radiotherapy Against Brain Cancers
Abstract
Purpose
Nanoparticle-mediated radiation therapy (NPRT) is an emerging clinical research frontier which seeks to improve radiotherapy outcomes through local tumor dose enhancement and radiosensitization. We recently showed improved NPRT outcomes for 2D glioblastoma (GBM) cells treated with graphene quantum dots (GQD) in vitro. To better mimic the in vivo tumor response in 3D microenvironments, we use Digital Twin-driven experimental-computational framework that integrates machine learning with advanced 3D in-vitro tumor models to characterize response to NPRT.
Methods
3D spheroids of GBM were produced and used to replicate physiologically relevant tumor architecture and microenvironmental conditions. Spheroids were treated with biocompatible GQD to effect both the increased release of electrons that promote reactive oxygen species (ROS) generation and dose enhancement. We implemented time-resolved imaging and biophysical measurements that provide readouts of cell viability, morphology, growth, migration from tumor and clonogenic cell survival. Unsupervised Machine Learning (UML) algorithms were developed in-house, in MATLAB, to extract therapeutic signatures from the 2D and 3D treatment results. Convolutional Neural Networks (CNNs) were trained based on extracted therapeutic signatures and used for automated image analysis and feature extraction. A digital twin model continuously coupled experimental observations with computational predictions, enabling iterative model updates.
Results
Our UML algorithms have successfully clustered images of cells treated with GQD and Temozolomide, without radiation, in a manner distinct from clustering patterns when radiotherapy is introduced. Deep learning models are being optimized to enable comparison with prior 2D NPRT data. Correlations between tumor viability, migration from spheroids/clonogenic survival and UML image clusters are being established and results will be presented.
Conclusion
This combined experimental-computational approach offers a scalable path toward patient-specific modeling and optimization of GBM radiotherapy. This framework will hopefully accelerate clinical trials involving NPRT, especially in cases of highly radioresistant brain cancers such as GBM.