Prediction of an fMRI-based schizophrenia biomarker from EEG using dynamic functional connectivity: a simultaneous EEG-fMRI study.

Objective.Recent advances in functional magnetic resonance imaging (fMRI) have identified brain functions associated with psychiatric disorders using machine-learning classifiers that serve as potential fMRI-derived biomarkers (fMRI-BMs). Resting-state fMRI functional connectivity (FC) is gaining attention as a potential target for neurofeedback (NF) therapy. However, using electroencephalography (EEG) instead of fMRI remains an important research direction for clinical implementation. Here, we propose a novel approach for estimating an fMRI-BM from EEG signals using dynamic FC (dFC) to enable biomarker-informed EEG-based neurofeedback.Approach.We targeted a schizophrenia (SCZ) fMRI-BM defined as a weighted linear summation (WLS) of 16 FCs, previously developed as a classifier distinguishing patients with SCZ and healthy control participants (HC). We first used a resting state-fMRI dataset including 100 SCZ patients and 554 HC to examine the correlation between the original SCZ-BM and a dynamic BM computed using dFC. For the estimation, we additionally used newly acquired simultaneous EEG-fMRI data from 54 control participants. The dFC time window was set to approximately 30-60 s to support integration into an EEG-NF system. Sparse logistic regression (SLR) was applied to address overfitting caused by the high dimensionality of input features. Prediction accuracy was assessed using the area under the receiver operating characteristic curve (AUC) with leave-one-run-out cross-validation and individual-level permutation testing.Main results.The proportion of biomarker-defined states derived from dFC was strongly correlated with the original SCZ-BM across both patients with SCZ and HC (Spearman's rho = 0.77 - 0.86). Permutation testing showed significant AUC values in 19 out of 21 participants, and specificity against MDD-BM was confirmed.Significance.These findings demonstrate a methodological proof-of-concept for estimating fMRI-defined connectivity biomarkers from EEG signals, paving the way for future biomarker-informed EEG-based NF applications, while clinical applicability remains to be established.