The hidden Markov model reveals the changes in brain dynamics among patients with end-stage renal disease under different dialysis methods

Abstract

End-stage renal disease patients often exhibit impairments in attention, memory, and executive function, reducing quality of life and increasing mortality risk.  Hemodialysis and peritoneal dialysis are the main treatment modalities, yet their effects on brain functional dynamics remain unclear.  In this study, resting-state functional magnetic resonance imaging data were collected from healthy controls and end-stage renal disease patients undergoing hemodialysis or peritoneal dialysis.  Group differences were then investigated using dynamic functional connectivity, a hidden Markov model, and support vector machine classification to investigate alterations in end-stage renal disease patients.  Dynamic analysis showed that hemodialysis patients exhibited more frequent state switching and unstable network patterns, reflecting widespread disruption of brain functional organization.  In contrast, peritoneal dialysis patients displayed state occupancy and dwell times closer to healthy controls, indicating relatively preserved dynamic stability, though minor connectivity abnormalities persisted.  Functional analysis revealed abnormalities in frontoparietal network and dorsal attention network in hemodialysis patients, while peritoneal dialysis patients showed imbalances between the default mode and other systems.  Support vector machine classification further confirmed these patterns, with the highest accuracy in distinguishing healthy controls from hemodialysis patients and intermediate network alterations in peritoneal dialysis patients.  These findings demonstrate that dynamic brain network analysis combined with machine learning can reveal modality-specific alterations in end-stage renal disease and provide potential imaging biomarkers for understanding dialysis-related cognitive impairment.