Related Experiment Video
Updated: Sep 28, 2025
12:03
A Method for Tracking the Time Evolution of Steady-State Evoked Potentials
Published on: May 25, 2019
8.6K
Spontaneous State Detection Using Time-Frequency and Time-Domain Features Extracted From
Huanpeng Ye1, Zhen Fan2, Guangye Li1
1State Key Laboratory of Mechanical System and Vibration, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China.
Frontiers in Neuroscience
|April 4, 2022
Summary
This study explored new features from stereo-EEG (SEEG) recordings for brain-computer interfaces (BCI). High-gamma band power and event-related potentials (ERP) showed promise for detecting brain activity with high accuracy.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Stereo-electroencephalography (SEEG) is a minimally invasive technique for recording intracranial brain signals.
- Current SEEG-based brain-computer interface (BCI) studies primarily use event-related potential (ERP) amplitude or band power features.
- The decoding potential of other time-frequency and time-domain features in SEEG remains underexplored.
Purpose of the Study:
- To validate the utility of various time-domain and time-frequency features derived from SEEG signals for BCI applications.
- To identify the most informative features for distinguishing between active and idle brain states.
- To assess the classification performance of these features in detecting neural activity.
Main Methods:
- SEEG signals were recorded during intermittent auditory stimuli, including responses to specific names.
- Features such as average amplitude, root mean square, linear regression slope, and line-length were extracted from ERP and band power (high-gamma, beta, alpha) traces.
- A hidden Markov model (HMM) was employed to detect the onset and offset of brain activation states.
Main Results:
- Valid time-domain and time-frequency features were identified across cortical and subcortical regions, including the temporal lobe, parietal lobe, and insula.
- The most effective features included average amplitude, root mean square, and line-length from high-gamma (60-140 Hz) power, and line-length from ERP.
- The HMM achieved high sensitivity (95.7 ± 1.3%) and precision (91.7 ± 1.6%) in detecting active brain states.
Conclusions:
- This study demonstrates the effectiveness of novel time-domain and time-frequency features for SEEG-based BCI.
- High-gamma band power and ERP features offer significant potential for improving BCI performance.
- The findings provide valuable insights for feature selection in future SEEG-BCI research and applications.
