Classifying schizophrenia subtypes via resting-state EEG complexity networks

Schizophrenia (SZ) is increasingly recognized as a network disorder marked by abnormal functional connectivity, yet the clinical utility of fMRI remains limited. Electroencephalography (EEG) provides a more practical alternative, though conventional complexity measures, such as sample entropy (SampEn), often fail to capture spatiotemporal network dynamics and yield inconsistent results. Here, we propose a novel EEG-based complexity network approach to investigate functional alterations in SZ subtypes, deficient (DS) and non-deficient (NDS), and to differentiate them from healthy controls (HCs). Resting-state EEG (64-channel, 500 Hz) was recorded from 19 DS patients, 19 NDS patients, and 30 HCs. Sample entropy and fuzzy entropy were computed, and complexity networks were constructed using Spearman and Pearson correlation coefficients. Key topological features, including global efficiency, local efficiency, and strength, were extracted and subjected to machine learning classification. While traditional SampEn differentiated groups only in the delta band, our network-based approach revealed distinct topological patterns: DS showed the highest local efficiency (δ, θ, α) but the lowest global efficiency (δ, α), whereas NDS exhibited lower global efficiency (θ) and higher local efficiency (β). SVM-based classification achieved an overall accuracy of 96.3%, with optimal performance in the δ and θ bands. These results underscore the utility of EEG complexity networks in distinguishing SZ subtypes from HCs and provide compelling evidence for aberrant connectivity in SZ. This method holds considerable promise for clinical applications, particularly in outpatient settings, though further validation in larger cohorts and task-based paradigms is warranted.