A new machine learning (ML) model that predicts the glioblastoma multiforme (GBM) brain tumor was developed by the researchers at the University of Waterloo and the University of Toronto in collaboration with St. Michael’s Hospital in Toronto.
GBM is a very aggressive form of brain cancer. It is fast growing and very difficult to surgically remove.
Despite the modern therapies applied to the patients diagnosed with GBM, the average survival is only slightly over one year.
The model
The research team created a new ML model that was tested and validated on two datasets provided by Magnetic Resonance Imaging (MRI). The first dataset was generated from synthetic tumors and the second dataset was extracted from the MRI of five patients with untreated GBM.
The approach used the proliferation-invasion (PI) mathematical model that relies on two key parameters, the tumor cell diffusivity (D) and proliferation rate (r).
During the tests, the model was capable of making accurate and personalized predictions of tumor growth.
This new model’s ability to solely rely on the patient’s data whose GBM forecasts are being made is one of its many advantages.
Conclusion, future research
This research proved that ML models can learn to identify patterns and make accurate predictions about GBM brain tumor.
They may enable faster diagnoses and help doctors to choose the most effective treatment for each particular patient.
For future research, the model may be expanded to include the impact of treatment on GBM tumors. The pipeline used to estimate the parameters of this model can also be extended to other diseases.
Learn more
- Research paper: “Deep Learning Characterization of Brain Tumours With Diffusion Weighted Imaging” (on bioRxiv)
- Story source: “Using Machine Learning to Predict Brain Tumor Progression” (on Neuroscience)