Predicting superagers by machine learning classification based on the functional brain connectome using resting-state...

Area of Science:

• Neuroscience
• Gerontology
• Machine Learning

Background:

• Superagers exhibit memory performance comparable to younger adults.
• Understanding the neurobiological basis of successful aging is crucial.
• Brain connectome analysis offers insights into cognitive aging.

Purpose of the Study:

• Investigate unique functional brain connectome patterns in superagers.
• Develop machine learning models to differentiate superagers from typical agers.
• Identify brain regions and networks associated with superaging.

Main Methods:

• Collected resting-state functional magnetic resonance imaging (rsfMRI) data from 32 superagers and 58 typical agers.
• Utilized machine learning classifiers: Support Vector Machine (SVM), Random Forest (RF), and Logistic Regression (LR).
• Employed an ensemble learning method combining multiple classifiers.

Main Results:

• Machine learning models achieved high prediction accuracy for superagers (SVM, RF, LR all ~0.944).
• Random Forest classifier showed the highest Area Under the Curve (AUC) of 0.979.
• An ensemble model reached the highest AUC of 0.986.
• Key predictive brain regions included the precuneus, posterior cingulate gyrus, insular cortex, and frontal gyri within default, salient, and multiple-demand networks.

Conclusions:

• Resting-state fMRI data accurately predicts superager status.
• Functional brain connectome patterns effectively differentiate superagers from typical agers.
• This approach captures unique neurophysiological characteristics of successful cognitive aging.