Medical News Observer

• The study involved 48 participants aged 21-65, using fMRI to analyze brain response to a sensorimotor task for predicting cognitive health.
• The machine learning model utilized hemodynamic response functions, including amplitude and dispersion changes, to forecast BrainHealth Index gains.
• Achieving 90% accuracy, the research underscores the potential of neuroimaging and machine learning in tracking and predicting cognitive health.

Researchers from the Center for BrainHealth at The University of Texas at Dallas trained a machine learning model to forecast changes in cognitive brain health using neural biomarkers from the brain’s blood flow response. The key focus of the research was to predict improvements in the BrainHealth Index using the neural markers obtained from the fMRI sessions.

The study involved 48 participants aged 21-65 who performed a sensorimotor task during two functional magnetic resonance imaging (fMRI) sessions six months apart. The hemodynamic response functions (HRFs) were analyzed using traditional metrics (like amplitude, dispersion, and latency) and novel ones (curvature, canonicality) and used as inputs for a neural network model.

The machine learning model showed a 90% accuracy in predicting significant gains in the BrainHealth Index. This prediction was based on three hemodynamic response function parameters: changes in amplitude, dispersion, and similarity to a canonical hemodynamic response function shape at the baseline. 

In summary, this research demonstrates the potential of neuroimaging measures to track and predict cognitive health in healthy individuals. It highlights how machine learning can significantly contribute to precision brain health, especially with novel metrics.

Ref: Spence, Jeffrey S, Monroe P Turner, Bart Rypma, Mark D’Esposito, and Sandra Bond Chapman. 2023. “Toward Precision Brain Health: Accurate Prediction of a Cognitive Index Trajectory Using Neuroimaging Metrics.” Cerebral Cortex, November, bhad435. https://doi.org/10.1093/cercor/bhad435.