Connectome-based model predicts episodic memory performance in individuals with subjective cognitive decline and amnestic mild cognitive impairment.

OBJECTIVE

To explore whether the whole brain resting-state functional connectivity (rs-FC) could predict episodic memory performance in individuals with subjective cognitive decline and amnestic mild cognitive impairment.

METHOD

This study included 33 cognitive normal (CN), 26 subjective cognitive decline (SCD) and 27 amnestic mild cognitive impairment (aMCI) patients, and all the participants completed resting-state fMRI (rs-fMRI) scan and neuropsychological scale test data. Connectome-based predictive modeling (CPM) based on the rs-FC data was used to predict the auditory verbal learning test-delayed recall (AVLT-DR) scores, which measured episodic memory in individuals. Pearson correlation between each brain connection in the connectivity matrices and AVLT-DR scores was computed across the patients in predementia stages of Alzheimer's disease (AD). The Pearson correlation coefficient values separated into a positive network and a negative network. Predictive networks were then defined and employed by calculating positive and negative network strengths. CPM with leave-one-out cross-validation (LOOCV) was conducted to train linear models to respectively relate positive and negative network strengths to AVLT-DR scores in the training set. During the testing procedure, each left-out testing subject's strengths of positive and negative network was normalized using the parameters acquired during training procedure, and then the trained models were used to predict the testing participant's AVLT-DR score.

RESULTS

The negative network predictive model tested LOOCV significantly predicted individual differences in episodic memory from rs-FC. Key nodes that brain regions contributed to the prediction model were mainly located in the prefrontal cortex, frontal cortex, parietal cortex and temporal lobe.

CONCLUSION

Our findings demonstrated that rs-FC among multiple neural systems could predict episodic memory at the individual level.