Continuous Mental State Estimation Using EEG Band Power Time Series Predictions for BCI Control

Long Short-Term Memory network (LSTM) can be used as a prediction model that reproduces time series dynamics based on training data. Here, LSTM was used to predict changes on EEG band power time series (BPts) trained with data from two different mental tasks, assuming that BPts dynamics should diffe...

Full description

Saved in:
Bibliographic Details
Published in2023 IEEE EMBS R9 Conference pp. 1 - 4
Main Authors Moedano-Atristain, Jose E., Yanez-Suarez, Oscar, Bojorges-Valdez, Erik
Format Conference Proceeding
LanguageEnglish
Published IEEE 05.10.2023
Subjects
Brain modeling
Data models
Electroencephalography
Modulation
Predictive models
Time series analysis
Training data
Online AccessGet full text
DOI10.1109/IEEECONF60929.2023.10525603

Cover

More Information
Summary:Long Short-Term Memory network (LSTM) can be used as a prediction model that reproduces time series dynamics based on training data. Here, LSTM was used to predict changes on EEG band power time series (BPts) trained with data from two different mental tasks, assuming that BPts dynamics should differ between those two tasks. Prediction error was used as a single feature to continuously identify mental state. AUROC values were calculated, achieving 0.672 ± 0.027 on f3 band as best result. Clinical relevance - Band power time series (BPts) have been used to explore dynamics of the autonomic branches, and also as a descriptor of their functional integrity. This work explores if such dynamics change by subject intention or specific task execution. This was a first attempt at using BPts as BCI control signals, but analysis shows that their dynamics are more related to intrinsic neural modulation. If the former is true, then BPts could be explored as biomarkers of autonomic alterations.
DOI:10.1109/IEEECONF60929.2023.10525603