Median Nerve Stimulation Based BCI: A New Approach to Detect Intraoperative Awareness During General Anesthesia
The research addresses Accidental Awareness during General Anesthesia (AAGA), a rare but distressing phenomenon where patients unexpectedly wake up during surgery under general anesthesia. It affects approximately 0.1-0.2% of cases, leading to severe trauma and potential long-term psychological effects such as PTSD. Current methods to monitor anesthesia depth, including clinical observation and EEG-based techniques, often fall short in preventing AAGA. To mitigate this issue, the study proposes an innovative passive Brain-Computer Interface (BCI) based on Motor Imagery (MI) triggered by Median Nerve Stimulation (MNS). This approach leverages patients' attempts to move when experiencing AAGA, despite being curarized (neuromuscular blockage), which typically inhibits physical movement. The research investigates how MNS modulates motor cortex activity and explores the feasibility of detecting MI through EEG signals. Initial experiments on healthy participants suggest that this BCI could potentially detect AAGA with high accuracy. In summary, the study aims to develop a novel BCI system that utilizes MI triggered by MNS to detect AAGA, offering a promising approach to enhance patient safety during anesthesia.
Enhancing Patient Safety: Leveraging Median Nerve Stimulation and Motor Imagery for AAGA Detection
The study explores methods to detect Accidental Awareness during General Anesthesia (AAGA) using a passive Brain-Computer Interface (BCI). Current techniques often fail to prevent AAGA, where patients wake unexpectedly during surgery. The research proposes using median nerve stimulation (MNS) combined with motor imagery (MI) to detect intention of movement, crucial during AAGA, when patients attempt to move despite being paralyzed. Challenges include developing a BCI without explicit time markers and achieving high accuracy in detecting movement intentions. Previous studies suggest MNS induces specific EEG patterns (ERD during stimulation, ERS post-stimulation), modulated by MI. Experiments involved 16 healthy subjects performing motor tasks under MNS to analyze EEG responses. Results indicate MI significantly modulates ERD and ERS induced by MNS, supporting the potential for a more effective AAGA detection system compared to conventional methods. The study aims to enhance patient safety by leveraging these findings to develop a reliable BCI for hospitals.
Condition 1 (C1) : Real Movement (RM)
Real Movement (RM), involved participants performing an isometric grasp between the thumb and index finger on a pointer button. A low-frequency beep signaled the start of the movement, which lasted 2 seconds. Another beep indicated the end of the task. Trigger signals from the pointer button recorded when the participant initiated and ceased the RM task. This straightforward movement reliably elicited observable EEG changes, as documented in previous studies (Shibasaki et al., 1993).
2.2.1. Condition 1: Real Movement (RM)
Participants performed an isometric grasp between the thumb and index finger on a pointer button. A low-frequency beep signaled the start and another beep indicated the end of the 2-second task. Trigger signals from the pointer button were recorded to precisely time the execution and cessation of the movement. This task reliably induced observable EEG changes, as previously documented.
2.2.2. Condition 2: Motor Imagery (MI)
In this condition, participants imagined performing the same grasping movement without any physical movement. Similar to RM, a low-frequency beep initiated the 2-second motor imagery task, with another beep marking its conclusion
2.2.3. Condition 3 : Motor Imagery With Median Nerve Stimulation (MI + MNS)
Subjects performed motor imagery while their median nerve was stimulated 750 ms after task onset. This timing was chosen based on average reaction times to maximize stimulation during Event-Related Desynchronization (ERD) of motor imagery. Stimulation intensity ranged between 8 and 15 mA and lasted 100 ms.
2.2.4. Condition 4: Median Nerve Stimulation Only (MNS)
Participants received median nerve stimulation without concurrent motor imagery. Stimulation parameters were adjusted individually (8-15 mA) to elicit a slight thumb-index finger movement without causing pain.
2.3. Experimental Design
Each participant underwent a 120-minute session divided into four phases: EEG cap setup, stimulation intensity calibration, execution of RM, MI, MI + MNS, and MNS trials, and session conclusion. Trials were randomized across two runs to minimize bias and fatigue. Trials lasted 8 ± 1 seconds each.
Data AcquisitionData Acquisition
EEG signals were recorded using the OpenViBE platform with a Biosemi Active Two 128-channel system. Electrode placement covered key cortical areas, ensuring comprehensive observation of motor cortex activity during tasks.
2.5. Data Pre-Processing
Offline analysis involved EEG signal processing using EEGLAB and Matlab2015b. Signals were re-referenced, resampled, and segmented into epochs to analyze ERD/ERS patterns. Trials with muscle artifacts or significant deviations from expected ERD/ERS patterns were excluded from analysis to ensure data quality. This methodological approach facilitated the investigation of motor patterns during RM, MI, MI + MNS, and MNS conditions, offering insights into EEG modulations relevant to developing a BCI for AAGA detection.
2.6. Time-Frequency Analysis
Event-Related Spectral Perturbation (ERSP) analysis was conducted between 8 and 35 Hz using EEGLAB to compare all four experimental conditions. A sliding Fast Fourier Transform (FFT) window of 256 points with a padratio of 4 was applied to compute mean ERSPs from 2 seconds before the task onset to 7 seconds after. Baseline activity was determined from 1.5 seconds before the auditory cue for Conditions 1 and 2, and 2 seconds before stimulation for Conditions 3 and 4. Statistical significance (p < 0.05) was confirmed using a surrogate permutation test with 2,000 permutations.
2.7. Topographies
Brain topographies were used to visualize changes over different scalp electrodes, particularly to distinguish MI + MNS from MNS conditions and determine optimal time parameters for classification. ERSPs in the merged mu+beta band (8–30 Hz) were computed and statistically validated using surrogate permutation tests (p < 0.05, 2,000 permutations). False Discovery Rate (FDR) correction was applied to control for multiple comparisons, enhancing the accuracy of localization differences between conditions.
2.8. ERD/ERS Quantification
ERD/ERS percentages were computed using the band power method to quantify modulation in mu (7–13 Hz), beta (15–30 Hz), and mu+beta (8–30 Hz) bands. ERD indicates desynchronization (negative percentage), and ERS indicates synchronization (positive percentage). Averaging techniques were employed to enhance accuracy in detecting power modulations during MI, MI + MNS, and MNS conditions.
2.9. Classification
Classification tasks were performed to distinguish between RM vs. Rest, MI vs. Rest, and MI + MNS vs. MNS conditions. Trials were segmented into motor task and rest periods (2.5 seconds each for RM and MI, 3 seconds for MI + MNS and MNS) for feature extraction and classification. EEG signals were bandpass filtered (8–30 Hz) with a 5th-order Butterworth filter. Each condition comprised 52 trials, and classification accuracy was evaluated to assess the effectiveness of detecting motor imagery and stimulation-induced responses. These analytical methods enabled comprehensive exploration of EEG dynamics during motor tasks and stimulation, contributing to the development of a robust BCI for detecting intraoperative awareness.
3. Result ,3.1. Behavioral Results ,3.1.1. Reaction Time
Real Movement (RM) Condition: The reaction time between the first beep and the start of movement was 0.5948 seconds ± 0.1929. The reaction time between the second beep and the cessation of movement was 0.5038 seconds ± 0.1174. These reaction times are consistent with findings in the literature (Jain et al., 2015).
3.1.2. Removed Trials
Artifacts: A total of 832 trials were initially acquired (52 per subject per condition). Due to artifact presence, trials were removed as follows: RM Condition: 125 trials (15%) Motor Imagery (MI) Condition: 119 trials (14.3%) Median Nerve Stimulation (MNS) Condition: 114 trials (13.7%) MI + MNS Condition: 138 trials (16.6%) Artifacts were evenly distributed across subjects.
3.2. Time-Frequency Maps
Time-frequency maps were generated to illustrate the evolution of signal power, aiding in identifying frequency and time windows where Event-Related Spectral Perturbations (ERSP) occurred (see Figure 4).
Software
Signal Recording and Processing: OpenViBE software was utilized for EEG and EMG signal recording, median nerve stimulator synchronization, and auditory cue generation (Renard et al., 2010). Data Analysis: MATLAB 2015b was used for processing and analyzing ERD/ERS modulations, implementing custom scripts and algorithms. Classification Algorithms: All classification algorithms were implemented using Scikit Learn Python 2.7, including Linear Discriminant Analysis (LDA) with Common Spatial Pattern (CSP) features, and Riemannian Geometry based methods such as Minimum Distance to Riemannian Mean (MDM) and Tangent Space Linear Regression (TS + LR) (Pedregosa et al., 2011). These results and methods contribute to understanding motor imagery and stimulation-induced responses, crucial for developing effective Brain-Computer Interfaces (BCIs) in clinical settings.