Deep Learning In a 3D Brain Cancer Twin for Advancing Radioimmunotherapy Against Brain Cancers
Abstract
Purpose
Glioblastoma is highly aggressive and the most prevalent brain tumor in adults. The median survival of patients is 15 months, despite the current standard of care comprising surgery, radiotherapy, and chemotherapy. Hence, new therapeutic approaches are urgently needed. On new approach is the combination of radiotherapy with immunotherapy. Here, we move from 2D to 3D radioimmunotherapy (RIT) to better mimic the in vivo tumor environment. Furthermore, we develop a MATLAB-based digital twin model that, when given therapeutic signatures, will accurately predict cell survival after specific treatment.
Methods
A convolutional neural network (CNN) is designed in Deep Network Designer to define the structure and features of a cell spheroid. The CNN is trained using input therapeutic signatures extracted using our in-house developed Unsupervised Machine Learning (UML) algorithms. The model predicts the therapeutic outcomes after radioimmunotherapy treatment based on the feature data obtained by the CNN and optimizes treatment from therapeutic signature.
Results
Our UML results involving RIT and 3D spheroid showed dramatic demonstration of machine learning capability. At 24 hours post-treatment, brightfield morphology achieved near-perfect treatment separation: 95% of RIT spheroids clustered together, distinct from 100% of lenalidomide-only spheroids. These results provide therapeutic signatures for our CNN under development. New results of the model will be presented and hopefully demonstrate the ability to promptly predict therapeutic outcomes.
Conclusion
A deep learning-driven 3D twin model of glioblastoma cell spheroids is being developed for in vitro prognosis that will guide clinical trials in the short term and personalized therapy using patient samples, in the long term.