Detection and Classification of Levels of Consciousness Using Parietal EEG-fNIRS During Anesthesia
Detection and Classification of Levels of Consciousness Using Parietal EEG-fNIRS During Anesthesia
The current evaluations of the levels of consciousness during anesthesia have limited precision. This can produce negative clinical consequences such as intraoperative awareness or neurological damage due to under- or over-infusion of anesthesia, respectively. The study's objective is to determine and classify biomarkers of electrical and hemodynamical brain activity associated with the levels of consciousness between wakefulness and anesthesia. For this purpose, a parietal electroencephalography (EEG) and a functional near-infrared spectroscopy (fNIRS) measurement paradigm will be used, as well as machine-learning. Volunteering patients (n = 25), who will be subject to an endoscopy procedure, will be measured during the infusion of anesthesia with propofol. EEG and fNIRS parameters will then be related to the Modified Ramsay clinical scale of consciousness.
Devices for data acquisition The brain electrical signals will be measured with four electrodes and a Cyton OpenBCI board with a sampling frequency of 250 Hz. The brain hemodynamical signals will be measured with a NIRSport with 16 optodes and a sampling frequency of 15.525 Hz and a wavelength of 760 nm. The electrodes and optodes will be positioned in the parietal zone, following the recommended 10-5 system for the multimodal EEG-fNIRS paradigm. The software Lab Streaming Layer will be used to synchronize the EEG and fNIRS data.
Data acquisition Patients will be summoned 30 minutes before their clinical appointment to put, adjust and calibrate the EEG and fNIRS systems. A 5-minute baseline measurement will be performed before the endoscopy procedure, with eyes closed. Afterward, patients will be anesthetized with a constant infusion rate of 20 mg/kg/h-1 of intravenous propofol until loss of consciousness, within 20 minutes. During infusion, the Modified Ramsay Scale will be used by the anesthetist in charge to measure the levels of consciousness, assessed by the patient's response to verbal or painful stimuli. The level of consciousness will be evaluated every two minutes until complete loss of consciousness, which is assumed after the loss of defensive or purposeful response to a second standard tetanic stimulation. After this, the endoscopy procedure will continue normally following standard clinical protocols. Before leaving, patients will be asked to answer the BRICE survey, used to evaluate the patient's experience during surgery or similar procedures. Patients will be measured throughout the whole endoscopy procedure. Any additional considerations will be managed in a case-by-case manner by the medical staff in charge.
Justification of the chosen sample size The sample size (n = 25) was determined by the Cohen Test, with a statistical power of 0.8 and an alfa power of 0.05, to determine significant intraspecific subject differences when comparing wakefulness, deep sedation, and intermediate levels of consciousness.
EEG data analysis The delta (0.1-3 Hz), theta (4-7 Hz), lower-alpha (8-12 Hz), upper-alpha (12-15 Hz), and beta/gamma (15-40 Hz) power bands will be used as features for the decoding model. The features will be calculated using a moving window of one minute. The levels of consciousness identified using the Modified Ramsay Scale will be paired with the corresponding window. The software Homer 2012: MNE in Python will be used for these analyses.
fNIRS data analysis The temporal reference of the oxygenated (HbO2) and deoxygenated (HHb) hemoglobin will be obtained from the optical signals, using the modified Beer-Lambert law. The regions of interest (ROIs) will be obtained in relation to regional local average activity. The average, maximal, and slope of the signal of each ROI will be obtained. Vector-phase analyses will also be implemented, with one-minute windows. The software Homer 2021: MNE in Python will be used for these analyses.
Classification using machine-learning For each one-minute window, the EEG and fNIRS features will be given to a Support Vector Machine (SVM) classifier using a basal radius. Each window will be attached to the level of consciousness according to the Modified Ramsay Scale. Three models will be tested: only EEG, only fNIRS, and fNIRS + EEG. Each model will try to decode the patient's level of consciousness using the aforementioned scale.
Inclusion Criteria:
Exclusion Criteria: