Detecting Connected Consciousness During Propofol-Induced Anesthesia Using EEG Based Brain Decoding

Self-Supervised Learning for Electroencephalography

Decades of research have shown machine learning superiority in discovering highly nonlinear patterns embedded in electroencephalography (EEG) records compared with conventional statistical techniques. However, even the most advanced machine learning techniques require relatively large, labeled EEG repositories. EEG data collection and labeling are costly. Moreover, combining available datasets to achieve a large data volume is usually infeasible due to inconsistent experimental paradigms across trials. Self-supervised learning (SSL) solves these challenges because it enables learning from EEG records across trials with variable experimental paradigms, even when the trials explore different phenomena. It aggregates multiple EEG repositories to increase accuracy, reduce bias, and mitigate overfitting in machine learning training. In addition, SSL could be employed in situations where there is limited labeled training data, and manual labeling is costly. This article: 1) provides a brief introduction to SSL; 2) describes some SSL techniques employed in recent studies, including EEG; 3) proposes current and potential SSL techniques for future investigations in EEG studies; 4) discusses the cons and pros of different SSL techniques; and 5) proposes holistic implementation tips and potential future directions for EEG SSL practices.

EEG spectral slope: A reliable indicator for continuous evaluation of consciousness levels during propofol anesthesia

The level of consciousness undergoes continuous alterations during anesthesia. Prior to the onset of propofol-induced complete unconsciousness, degraded levels of behavioral responsiveness can be observed. However, a reliable index to monitor altered consciousness levels during anesthesia has not been sufficiently investigated. In this study, we obtained 60-channel EEG data from 24 healthy participants during an ultra-slow propofol infusion protocol starting with an initial concentration of 1 μg/ml and a stepwise increase of 0.2 μg/ml in concentration. Consecutive auditory stimuli were delivered every 5 to 6 s, and the response time to the stimuli was used to assess the responsiveness levels. We calculated the spectral slope in a time-resolved manner by extracting 5-second EEG segments at each auditory stimulus and estimated their correlation with the corresponding response time. Our results demonstrated that during slow propofol infusion, the response time to external stimuli increased, while the EEG spectral slope, fitted at 15-45 Hz, became steeper, and a significant negative correlation was observed between them. Moreover, the spectral slope further steepened at deeper anesthetic levels and became flatter during anesthesia recovery. We verified these findings using an external dataset. Additionally, we found that the spectral slope of frontal electrodes over the prefrontal lobe had the best performance in predicting the response time. Overall, this study used a time-resolved analysis to suggest that the EEG spectral slope could reliably track continuously altered consciousness levels during propofol anesthesia. Furthermore, the frontal spectral slope may be a promising index for clinical monitoring of anesthesia depth.

Selective corticocortical connectivity suppression during propofol-induced anesthesia in healthy volunteers

We comprehensively studied directional feedback and feedforward connectivity to explore potential connectivity changes that underlie propofol-induced deep sedation. We further investigated the corticocortical connectivity patterns within and between hemispheres. Sixty-channel electroencephalographic data were collected from 19 healthy volunteers in a resting wakefulness state and propofol-induced deep unconsciousness state defined by a bispectral index value of 40. A source analysis was employed to locate cortical activity. The Desikan-Killiany atlas was used to partition cortices, and directional functional connectivity was assessed by normalized symbolic transfer entropy between higher-order (prefrontal and frontal) and lower-order (auditory, sensorimotor and visual) cortices and between hot-spot frontal and parietal cortices. We found that propofol significantly suppressed feedforward connectivity from the left parietal to right frontal cortex and bidirectional connectivity between the left frontal and left parietal cortex, between the frontal and auditory cortex, and between the frontal and sensorimotor cortex. However, there were no significant changes in either feedforward or feedback connectivity between the prefrontal and all the lower-order cortices and between the frontal and visual cortices or in feedback connectivity from the frontal to parietal cortex. Propofol anesthetic selectively decreased the unidirectional interaction between higher-order frontoparietal cortices and bidirectional interactions between the higher-order frontal cortex and lower-order auditory and sensorimotor cortices, which indicated that both feedback and feedforward connectivity were suppressed under propofol-induced deep sedation. Our findings provide critical insights into the connectivity changes underlying the top-down mechanism of propofol anesthesia at deep sedation.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11571-021-09775-x.

A Vector Quantization-Based Spike Compression Approach Dedicated to Multichannel Neural Recording Microsystems

Implantable high-density multichannel neural recording microsystems provide simultaneous recording of brain activities. Wireless transmission of the entire recorded data causes high bandwidth usage, which is not tolerable for implantable applications. As a result, a hardware-friendly compression module is required to reduce the amount of data before it is transmitted. This paper presents a novel compression approach that utilizes a spike extractor and a vector quantization (VQ)-based spike compressor. In this approach, extracted spikes are vector quantized using an unsupervised learning process providing a high spike compression ratio (CR) of 10-80. A combination of extracting and compressing neural spikes results in a significant data reduction as well as preserving the spike waveshapes. The compression performance of the proposed approach was evaluated under variant conditions. We also developed new architectures such that the hardware blocks of our approach can be implemented more efficiently. The compression module was implemented in a 180-nm standard CMOS process achieving a SNDR of 14.49[Formula: see text]dB and a classification accuracy (CA) of 99.62% at a CR of 20, while consuming 4[Formula: see text][Formula: see text]W power and 0.16[Formula: see text]mm² chip area per channel.