Disrupted network communication predicts mild cognitive impairment in end-stage renal disease: an individualized machine learning study based on resting-state fMRI

Abstract End-Stage Renal Disease (ESRD) is known to be associated with a range of brain injuries, including cognitive decline. The purpose of this study is to investigate the functional connectivity (FC) of the resting-state networks (RSNs) through resting state functional magnetic resonance imaging (MRI), in order to gain insight into the neuropathological mechanism of ESRD. A total of 48 ESRD patients and 49 healthy controls underwent resting-state functional MRI and neuropsychological tests, for which Independent Components Analysis and graph-theory (GT) analysis were utilized. With the machine learning results, we examined the connections between RSNs abnormalities and neuropsychological test scores. Combining intra/inter network FC differences and GT results, ESRD was optimally distinguished in the testing dataset, with a balanced accuracy of 0.917 and area under curve (AUC) of 0.942. Shapley additive explanations results revealed that the increased functional network connectivity between DMN and left frontoparietal network (LFPN) was the most critical predictor for ESRD associated mild cognitive impairment diagnosis. Moreover, hypoSN (salience network) was positively correlated with Attention scores, while hyperLFPN was negatively correlated with Execution scores, indicating correlations between functional disruption and cognitive impairment measurements in ESRD patients. This study demonstrated that both the loss of FC within the SN and compensatory FC within the lateral frontoparietal network coexist in ESRD. This provides a network basis for understanding the individual brain circuits and offers additional noninvasive evidence to comprehend the brain networks in ESRD.