INFERENCE OF BRAIN STATES UNDER ANESTHESIA WITH META LEARNING BASED DEEP LEARNING MODELS
Abstract : Monitoringthe depth of unconsciousnessduring anesthesia is beneficial in both clinical settings and neuroscience investigations to understand brain mechanisms. Electroencephalogram(EEG) has been used as an objective means of characterizing brain altered arousal and/or cognition states induced by anesthetics in real-time. Different general anesthetics affect cerebral electrical activities in different ways. However, the performance of conventional machine learning models on EEG data is unsatisfactory due to the low Signal to Noise Ratio (SNR) in the EEG signals, especially in the office-based anesthesia EEG setting. Deep learning models have been used widely in the field of Brain Computer Interface (BCI) to perform classificationand pattern recognition tasks due to their capability of good generalization and handling noisesCompared to other BCI applications, where deep learning has demonstrated encouraging results, the deep learning approach for classifying different brain consciousnessstates under anesthesia has been much less investigated. In this paper, we propose a new framework based on meta-learning using deep neural networks, named Anes-MetaNet, to classify brain states under anesthetics The Anes-MetaNet is composed of ConvolutionalNeural Networks (CNN) to extract power spectrum features, and a time consequence model based on Long Short-Term Memory (LSTM) networks to capture the temporal dependencies, and a meta-learning framework to handle large cross-subject variability We use a multi-stage training paradigm to improve the performance, which is justified by visualizing the high-level feature mapping. Experiments on the office-based anesthesia EEG dataset demonstrate the effectivenessof our proposedAnes-MetaNet by comparison of existing methods. Index Terms—Brain state estimation, meta learning, anesthesia EEG, deep learning
? Experiments on the office-based anesthesia EEG dataset demonstrate the effectivenessof our proposedAnes-MetaNet by comparison of existing methods.
? As for the DoA task, only a few researches based on deep learning have been conducted. For instance, Park et al. proposed a real-time DoA monitoring system based on a convolutional neural network framework
? Although deep learning frameworks proposed in the previous literature typically outperform the existing classical machine learning algorithms, these studies are designed to analyze the anesthetic EEG data collected from the hospital-based environment.
? EEG signals vary significantly from subject to subject. The classification function f obtained using CNN training from EEG data of some subjects may not be applicable to other subjects. Meta-learning is used in this paper to address this problem.
? LSTM network is a special instance of RNN which mitigates the problems of gradient vanishing and gradient explosion in long sequence training of RNN.
? we randomly extract the same amount of data for training, which can effectively prevent the overfitting problem caused by the problem of class imbalance
? Experiments on the office-based anesthesia EEG dataset demonstrate the effectivenessof our proposedAnes-MetaNet by comparison of existing methods.
? Although deep learning frameworks proposed in the previous literature typically outperform the existing classical machine learning algorithms, these studies are designed to analyze the anesthetic EEG data collected from the hospital-based environment.
? Even though the CLA model performs much better than the other existing models in when applying it to an HBA dataset, such results indicate this model is not suitable for the OBA condition
? Electroencephalogram(EEG) has been used as an objective means of characterizing brain altered arousal and/or cognition states induced by anesthetics in real-time
? Measurements through the observation of heart rate, breathing pattern, blood pressure and other factors, have been used to measure DoA
? Currently, the Bispectral Index (BIS, Aspect Medical Systems, USA) is the most widely used DoA monitor index in clinical practice; however, previous researchers have reported a poor connection between the DoA and the BIS
|