Dataset Information

Bayesian Optimization of Machine Learning Classification of Resting-State EEG Microstates in Schizophrenia: A Proof-of-Concept Preliminary Study Based on Secondary Analysis.

ABSTRACT: Resting-state electroencephalography (EEG) microstates reflect sub-second, quasi-stable states of brain activity. Several studies have reported alterations of microstate features in patients with schizophrenia (SZ). Based on these findings, it has been suggested that microstates may represent neurophysiological biomarkers for the classification of SZ. To explore this possibility, machine learning approaches can be employed. Bayesian optimization is a machine learning approach that selects the best-fitted machine learning model with tuned hyperparameters from existing models to improve the classification. In this proof-of-concept preliminary study based on secondary analysis, 20 microstate features were extracted from 14 SZ patients and 14 healthy controls' EEG signals. These parameters were then ranked as predictors based on their importance, and an optimized machine learning approach was applied to evaluate the performance of the classification. SZ patients had altered microstate features compared to healthy controls. Furthermore, Bayesian optimization outperformed conventional multivariate analyses and showed the highest accuracy (90.93%), AUC (0.90), sensitivity (91.37%), and specificity (90.48%), with reliable results using just six microstate predictors. Altogether, in this proof-of-concept study, we showed that machine learning with Bayesian optimization can be utilized to characterize EEG microstate alterations and contribute to the classification of SZ patients.

SUBMITTER: Keihani A

PROVIDER: S-EPMC9688063 | biostudies-literature | 2022 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Bayesian Optimization of Machine Learning Classification of Resting-State EEG Microstates in Schizophrenia: A Proof-of-Concept Preliminary Study Based on Secondary Analysis.

Keihani Ahmadreza A Sajadi Seyed Saman SS Hasani Mahsa M Ferrarelli Fabio F

Brain sciences 20221104 11

Resting-state electroencephalography (EEG) microstates reflect sub-second, quasi-stable states of brain activity. Several studies have reported alterations of microstate features in patients with schizophrenia (SZ). Based on these findings, it has been suggested that microstates may represent neurophysiological biomarkers for the classification of SZ. To explore this possibility, machine learning approaches can be employed. Bayesian optimization is a machine learning approach that selects the be ...[more]

PMID: 36358423

Similar Datasets

Project description:BackgroundThe "Avolition-apathy" domain of the negative symptoms was found to include different symptoms by factor analytic studies on ratings derived by different scales. In particular, the relationship of anhedonia with this domain is controversial. Recently introduced negative symptom rating scales provide a better assessment of anhedonia, allowing the distinction of anticipatory and consummatory aspects, which might be related to different psychopathological dimensions. The study of associations with external validators, such as electrophysiological, brain imaging or cognitive indices, might shed further light on the status of anhedonia within the Avolition-apathy domain.ObjectivesWe used brain electrical microstates (MSs), which represent subsecond periods of quasi-stable scalp electrical field, associated with resting-state neural networks (and thus with global patterns of functional connectivity), to test whether the component symptoms of Avolition-apathy share the same correlates.MethodWe analyzed multichannel resting EEGs in 142 individuals with schizophrenia (SCZ) and in 64 healthy controls (HC), recruited within the add-on EEG study of the Italian Network for Research on Psychoses. Relative time contribution, duration and occurrence of four MS classes (MS-A/-B/-C/-D) were calculated. Group differences on MS parameters (contribution and duration) and their associations with negative symptom domains (assessed using the Brief Negative Symptoms Scale) were investigated.ResultsSCZ, in comparison to HC, showed increased contribution and duration of MS-C. The contribution of MS-A positively correlated with Avolition-apathy, but not with Expressive deficit. Within the Avolition-apathy domain, anticipatory anhedonia, avolition and asociality, but not consummatory anhedonia, showed the same correlations with MS-A contribution.ConclusionOur findings support the existence of distinct electrophysiological correlates of Avolition-apathy with respect to Expressive deficit, and lend support to the hypothesis that only the anticipatory component of anhedonia shares the same pathophysiological underpinnings of the Avolition-apathy domain.

Project description:Resting-state brain activities have been extensively investigated to understand the macro-scale network architecture of the human brain using non-invasive imaging methods such as fMRI, EEG, and MEG. Previous studies revealed a mechanistic origin of resting-state networks (RSNs) using the connectome dynamics modeling approach, where the neural mass dynamics model constrained by the structural connectivity is simulated to replicate the resting-state networks measured with fMRI and/or fast synchronization transitions with EEG/MEG. However, there is still little understanding of the relationship between the slow fluctuations measured with fMRI and the fast synchronization transitions with EEG/MEG. In this study, as a first step toward evaluating experimental evidence of resting state activity at two different time scales but in a unified way, we investigate connectome dynamics models that simultaneously explain resting-state functional connectivity (rsFC) and EEG microstates. Here, we introduce empirical rsFC and microstates as evaluation criteria of simulated neuronal dynamics obtained by the Larter-Breakspear model in one cortical region connected with those in other cortical regions based on structural connectivity. We optimized the global coupling strength and the local gain parameter (variance of the excitatory and inhibitory threshold) of the simulated neuronal dynamics by fitting both rsFC and microstate spatial patterns to those of experimental ones. As a result, we found that simulated neuronal dynamics in a narrow optimal parameter range simultaneously reproduced empirical rsFC and microstates. Two parameter groups had different inter-regional interdependence. One type of dynamics was synchronized across the whole brain region, and the other type was synchronized between brain regions with strong structural connectivity. In other words, both fast synchronization transitions and slow BOLD fluctuation changed based on structural connectivity in the two parameter groups. Empirical microstates were similar to simulated microstates in the two parameter groups. Thus, fast synchronization transitions correlated with slow BOLD fluctuation based on structural connectivity yielded characteristics of microstates. Our results demonstrate that a bottom-up approach, which extends the single neuronal dynamics model based on empirical observations into a neural mass dynamics model and integrates structural connectivity, effectively reveals both macroscopic fast, and slow resting-state network dynamics.

Project description:The error-related negativity (ERN) is a negative deflection in the electroencephalography (EEG) waveform at frontal-central scalp sites that occurs after error commission. The relationship between the ERN and broader patterns of brain activity measured across the entire scalp that support error processing during early childhood is unclear. We examined the relationship between the ERN and EEG microstates - whole-brain patterns of dynamically evolving scalp potential topographies that reflect periods of synchronized neural activity - during both a go/no-go task and resting-state in 90, 4-8-year-old children. The mean amplitude of the ERN was quantified during the -64 to 108 millisecond (ms) period of time relative to error commission, which was determined by data-driven microstate segmentation of error-related activity. We found that greater magnitude of the ERN associated with greater global explained variance (GEV; i.e., the percentage of total variance in the data explained by a given microstate) of an error-related microstate observed during the same -64 to 108 ms period (i.e., error-related microstate 3), and to greater anxiety risk as measured by parent-reported behavioral inhibition. During resting-state, six data-driven microstates were identified. Both greater magnitude of the ERN and greater GEV values of error-related microstate 3 associated with greater GEV values of resting-state microstate 4, which showed a frontal-central scalp topography. Source localization results revealed overlap between the underlying neural generators of error-related microstate 3 and resting-state microstate 4 and canonical brain networks (e.g., ventral attention) known to support the higher-order cognitive processes involved in error processing. Taken together, our results clarify how individual differences in error-related and intrinsic brain activity are related and enhance our understanding of developing brain network function and organization supporting error processing during early childhood.

Dataset Information

Bayesian Optimization of Machine Learning Classification of Resting-State EEG Microstates in Schizophrenia: A Proof-of-Concept Preliminary Study Based on Secondary Analysis.

Publications

Bayesian Optimization of Machine Learning Classification of Resting-State EEG Microstates in Schizophrenia: A Proof-of-Concept Preliminary Study Based on Secondary Analysis.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets