Classification of cognitive reserve in healthy older adults based on brain activity using support vector machine

Objective: Cognitive reserve (CR) refers to the capacity of the brain to actively cope with damage via the implementation of remedial cognitive processes. Traditional CR measurements focus on static proxies, which may not be able to appropriately estimate dynamic changes in CR. This study therefore investigated the cognitive performance and characteristics of brain activity of low- and high-CR healthy adults during resting and n-back task states and categorized subjects according to magnetoencephalographic (MEG) information using a support vector machine (SVM) classifier. Approach: Forty-one volunteers were divided into groups with low and high CR indexes based on their education, occupational attainment, leisure and social activities. Main results: The results can be summarized as follows. First, subjects with a higher CR had higher accuracies and faster reaction times in the task. Second, subjects with a lower CR had a higher M300 intensity but a constant M300 latency. Third, subjects with a higher CR had a higher beta intensity in the parietal and occipital regions during the task, whereas subjects with a higher CR had a higher gamma intensity in the right temporal region in the resting state. Finally, subjects with a higher CR had negative gamma asymmetry between the right and left occipital regions, whereas subjects with a lower CR had positive values in the resting state. Significance: These MEG results were subsequently used to classify subjects into high-/low-CR subjects using an SVM classifier, and a mean accuracy of 88.89% was obtained. This objective and nonstatic method for determining CR level warrants further research for a wider variety of future clinical applications.