Scientists at the University of Michigan have developed an innovative machine learning model that will assist doctors in selecting the most effective treatments for patients with glioma.
Glioma is the most common type of brain tumor, originating from glial cells—the supporting structures of the nervous system that ensure the functioning of neurons. This tumor is characterized by its aggressiveness and metabolic diversity, which makes its treatment particularly challenging.
According to a study published in the journal Cell Metabolism, researchers have created a “digital twin”—a computer model that reflects the tumor’s metabolism in real-time. This technology allows for the pre-determination of how a specific patient’s tumor will respond to various types of diets or medications.
Previously, scientists knew that certain diets (such as restricting specific amino acids) could slow the growth of glioma. However, a significant problem was that some tumor subjects are capable of producing these substances themselves, rendering the diet ineffective. The new technology will help medics determine exactly which patients will benefit from dietary restrictions and which tumors will prove resistant.
The model is based on Deep Learning, specifically Convolutional Neural Networks (CNNs). It analyzes three primary components:
-
The patient’s blood analysis;
-
The metabolic indicators of the tumor tissue;
-
The tumor’s genetic profile.
By integrating this data, the AI determines the “metabolic flux”—the rate at which cancer cells consume and process nutrients.
The researchers validated the model’s accuracy through experiments involving human data and subsequently through testing on mice. The AI accurately predicted not only the effectiveness of the diet but also the impact of specific medications.
Scientists hope that in the future, this method will be applicable not only to brain cancer but also to various other types of tumors, further advancing the development of personalized medicine.
