Gaining Brain Insights By Tapping Into The Black Box: Linking Structural MRI Features To Age And Cognition Using Shapley-Based Interpretation Methods
New research is leveraging the power of artificial intelligence (AI) to unlock secrets hidden within brain imaging data, offering a clearer picture of how our brains change with age and influence our cognitive abilities. Traditional machine learning models, often called “black boxes” due to their complex internal workings, can predict outcomes with high accuracy but don’t always reveal why they made a particular prediction. To address this, scientists are now using specialized methods, such as Shapley-based interpretation techniques, to peek inside these black boxes and understand which brain features are most important for different predictions.
By applying these “explainable AI” methods to structural magnetic resonance imaging (MRI) scans from a large population dataset, researchers trained AI models to predict a person’s age and their fluid intelligence (the ability to solve new problems and adapt to new situations). They found that specific brain regions play distinct roles. For instance, the average signal intensity in subcortical regions—areas located deep within the brain—was consistently and significantly associated with how the brain ages. Meanwhile, the ability to predict an individual’s fluid intelligence was strongly influenced by contributions from the hippocampus (crucial for memory), the cerebellum (involved in coordination and cognitive functions), and well-known cognitive centers in the frontal and temporal lobes. These findings highlight how interpretable AI can provide valuable, data-driven insights into the intricate relationship between brain structure, aging, and cognitive function.