No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
A Combinatorial Deep Learning Structure for Precise Depth of Anesthesia Estimation From EEG Signals
Abstract
Electroencephalography (EEG) is commonly used to measure the depth of anesthesia (DOA) because EEG reflects surgical pain and state of the brain. However, precise and real-time estimation of DOA index for painful surgical operations is challenging due to problems such as postoperative complications and accidental awareness. To tackle these problems, we propose a new combinatorial deep learning structure involving convolutional neural networks (inspired by the inception module), bidirectional long short-term memory, and an attention layer. The proposed model uses the EEG signal to continuously predicts the bispectral index (BIS). It is trained over a large dataset, mostly from those under general anesthesia with few cases receiving sedation/analgesia and spinal anesthesia. Despite the imbalance distribution of BIS values in different levels of anesthesia, our proposed structure achieves convincing root mean square error of 5.59 1.04 and mean absolute error of 4.3 0.87, as well as improvement in area under the curve of 15% on average, which surpasses state-of-the-art DOA estimation methods. The DOA values are also discretized into four levels of anesthesia and the results demonstrate strong inter-subject classification accuracy of 88.7% that outperforms the conventional methods.
... This study aims to develop an accurate estimation of the LoH index and LoH state. In the previous studies [32]- [35], although the LoH classification algorithms have predicted the LoH state very accurately, this still does not solve the problem for the LoH indexes which are on the borderline of the LoH states. As a result of these values, the LoH state can be misinterpreted and result in inaccurate drug administrating during different states which can cause serious inter-operative and postoperative health conditions [36]. ...
... Various machine learning and deep learning-based approaches for estimating the LoH have been proposed in the literature [24], [35], and [44][42]- [46]. By utilizing the Bagged Tree ML algorithm [20], the highest classification accuracy achieved was 95% while using a feature set consisting of 12 features. ...
... By utilizing the Bagged Tree ML algorithm [20], the highest classification accuracy achieved was 95% while using a feature set consisting of 12 features. While in deep learning-based systems, several studies have utilized long short-term memory (LSTM) and CNN [35] which have reported an overall classification accuracy of 87.36% and an Area under the Curve (AUC) of 79% for regressively predicted values of BIS. However, the CNN-based proposed systems for EEG classification still face the issue of: (i) limited big datasets which are required to train the CNN. ...
Objective:
Classifying the patient's depth of anesthesia (LoH) level into a few distinct states may lead to inappropriate drug administration. To tackle the problem, this paper presents a robust and computationally efficient framework that predicts a continuous LoH index scale from 0-100 in addition to the LoH state.
Methods:
This paper proposes a novel approach for accurate LoH estimation based on Stationary Wavelet Transform (SWT) and fractal features. The deep learning model adopts an optimized temporal, fractal, and spectral feature set to identify the patient sedation level irrespective of age and the type of anesthetic agent. This feature set is then fed into a multilayer perceptron network (MLP), a class of feed-forward neural networks. A comparative analysis of regression and classification is made to measure the performance of the chosen features on the neural network architecture.
Results:
The proposed LoH classifier outperforms the state-of-the-art LoH prediction algorithms with the highest accuracy of 97.1% while utilizing minimized feature set and MLP classifier. Moreover, for the first time, the LoH regressor achieves the highest performance metrics (R2 = 0.9, MAE = 1.5) as compared to previous work.
Significance:
This study is very helpful for developing highly accurate monitoring for LoH which is important for intraoperative and postoperative patients' health.
... Parameters such as frequency bands (delta, theta, alpha, beta, gamma), amplitude, and symmetry were analyzed. This comprehensive approach allowed for a detailed assessment of the brain's electrical activity across a wide range of frequencies, including delta (1-3 Hz), associated with deep sleep or anesthesia; theta (4-7 Hz), related to drowsiness and first stages of sleep; alpha (8)(9)(10)(11)(12)(13), seen in relaxed, yet awake states; beta (14-30 Hz), indicative of active, cognitive processing and alertness; and gamma (>30 Hz), associated with higher-order cognitive functions such as perception and consciousness. Analyzing these parameters provides insights into the neurophysiological effects of anesthetic agents on brain function. ...
... EEG offers a window into the cerebral activity, revealing how different anesthetic agents modulate brain function. [11,12] This study focuses on Fentanyl, a potent opioid, and Ketamine, a NMDA receptor antagonist, both widely used in anesthesia but with distinctly different mechanisms of action. [13] Fentanyl is known for its potent sedative effects, primarily acting on the mu-opioid receptors, leading to significant alterations in brain wave patterns, particularly in the Figure 1. ...
This study aimed to investigate and compare the neurophysiological impacts of two widely used anesthetic agents, Fentanyl and Ketamine, on EEG power spectra during different stages of anesthesia in adult patients undergoing minimally invasive surgery. EEG data were collected from patients undergoing anesthesia with either Fentanyl or Ketamine. The data were analyzed for relative power spectrum and fast-to-slow wave power ratios, alongside Spectral Edge Frequency 95% (SEF95), at 3 key stages: pre-anesthesia, during stable anesthesia, and post-anesthesia. EEG Relative Power Spectrum: Initially, both groups exhibited similar EEG spectral profiles, establishing a uniform baseline ( P > .05). Upon anesthesia induction, the Fentanyl group showed a substantial increase in delta band power ( P < .05), suggesting deeper anesthesia, while the Ketamine group maintained higher alpha and beta band activity ( P < .05), indicative of a lighter sedative effect. Fast and Slow Wave Power Ratios: The Fentanyl group exhibited a marked reduction in the fast-to-slow wave power ratio during anesthesia ( P < .05), persisting post-anesthesia ( P < .05) and indicating a lingering effect on brain activity. Conversely, the Ketamine group demonstrated a more stable ratio ( P > .05), conducive to settings requiring rapid cognitive recovery. Spectral Edge Frequency 95% (SEF95): Analysis showed a significant decrease in SEF95 values for the Fentanyl group during anesthesia ( P < .05), reflecting a shift towards lower frequency power. The Ketamine group experienced a less pronounced decrease ( P > .05), maintaining a higher SEF95 value that suggested a lighter level of sedation. The study highlighted the distinct impacts of Fentanyl and Ketamine on EEG power spectra, with Fentanyl inducing deeper anesthesia as evidenced by shifts towards lower frequency activity and a significant decrease in SEF95 values. In contrast, Ketamine’s preservation of higher frequency activity and more stable SEF95 values suggests a lighter, more dissociative anesthetic state. These findings emphasize the importance of EEG monitoring in anesthesia for tailoring anesthetic protocols to individual patient needs and optimizing postoperative outcomes.
... On the other hand, deep learning architectures have demonstrated promising results in capturing complex temporal and spatial dependencies within the EEG signals for DoA monitoring [17], [18]. In particular, employing hybrid deep learning methods showed promising results [19], [20]. Moreover, several other analyses have been explored for DoA monitoring, including functional connectivity [21], microstate analysis [22], unscented Kalman filter-based neural mass modeling [23], amplitude modulation [24], and intrinsic phaseamplitude coupling [25]. ...
... based on the use of four EEG channels, whereas ours are based on a single EEG channel, potentially reducing wearable complexity. Our proposed algorithm shows a higher mean CC for random sampling analysis compared to [20], which utilized two EEG channels. Additionally, our method exhibits lower computational complexity than [26], as we employ simple linear filters for extracting EEG sub-bands instead of MEMD. ...
Objectives: Commercial systems for monitoring the depth of anesthesia (DoA) are often financially inaccessible to developing countries. As an alternative, a wearable single frontal electroencephalogram (EEG) device can be utilized. Nonetheless, most studies addressing DoA monitoring utilizing just one frontal EEG channel rely on nonlinear features that require parameter tuning before computation, overlooking the potential interchangeability of such features across different databases.
Methods: Here, we present a parameter-free feature set for DoA monitoring using a single frontal EEG channel and evaluate its performance on two databases with different characteristics. First, the EEG signal is de-noised and split into its sub-bands. Second, several parameter-free features based on entropy, power and frequency, fractal, and variation are extracted from all sub-bands. Finally, the distinguished features are chosen and input into a random forest regressor to estimate the DoA index values.
Results: The reliability of the proposed feature set for the DoA monitoring is indicated by achieving a comparable correlation coefficient of 0.80 and 0.79 and mean absolute error of 7.1 and 9.0 between the reference and estimated DoA index values for Databases I and II, respectively.
Significance: The obtained results from this study confirm the possibility of affordable DoA monitoring using a portable EEG system. Given its simplicity and comparable results for both databases, the proposed feature set holds promise for practical application in real-world scenarios.
... Afshar S et al. proposed a combined deep learning architecture based on a detailed analysis of EEG in measuring anesthesia depth. This framework not only improved the accuracy of inter subject classification, but also improved the accuracy of anesthesia depth measurement [12]. ...
... Its value range is [0,+∞), and the derivative function value range is {0,1}. The expressions of the two are shown in equation (12). ...
Currently, brain-computer interface technology still poses hidden dangers in the complete control of ideas. Additionally, there are issues with the low sampling frequency and accuracy of EEG signal acquisition equipment. To address these concerns, this study proposes a combined model for EEG signal recognition and classification analysis by combining frequency division multi-feature complex brain networks with parallel convolutional neural networks. The effectiveness of this model has been verified. In model visualization analysis, the visualization results of t-distribution random neighborhood embedding in the third separable convolutional layer indicate that the two types of imagination have already experienced separation. There is a clear boundary between the two at position 0 on both the horizontal and vertical axes. This is a significant improvement compared to the comparative model. In the model performance verification, the full band classification accuracy in the synchronous network was maintained between 60% to 84%, and the μ-rhythm was maintained between 59% to 81%. The average classification accuracy of the combined model was 77.40% with higher performance, which was higher than 68.53% and 70.87% of the single scale convolutional neural network. In comparison with deep learning algorithms, the average classification accuracy of the combined model was 85.74%, much higher than the 66.20% and 76.69% of the comparative models. The composite model constructed has good performance in recognizing and classifying electroencephalogram signals. It can be effectively applied in practical brain-computer interface technology or electroencephalogram signal analysis.
... In recent works, Lee et al. proposed a deep learning model for investigating the influence of the infusion histories of propofol and remifentanil and patient physiological characteristics on the DOA [12]. In addition, Sara Afshar et al. developed a combinatorial deep learning structure that predicts the depth of anesthesia according to electroencephalography (EEG) signals, demonstrating good prediction performance [13][14][15]. However, EEG signals are susceptible to inotropic and cardiac interference and sensitive to electromagnetic interference [16]. ...
... This deep learning model was trained on 231 subjects who received TIVA during surgery. In contrast to previous work, Sara Afshar et al. [13] combined convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and an attention mechanism to develop a new framework for predicting the DOA according to EEG signals. However, in clinical surgery, anesthesiologists are more likely to control the DOA based on the effects of the drug. ...
In the target-controlled infusion (TCI) of propofol and remifentanil intravenous anesthesia, accurate prediction of the depth of anesthesia (DOA) is very challenging. Patients with different physiological characteristics have inconsistent pharmacodynamic responses during different stages of anesthesia. For example, in TCI, older adults transition smoothly from the induction period to the maintenance period, while younger adults are more prone to anesthetic awareness, resulting in different DOA data distributions among patients. To address these problems, a deep learning framework that incorporates domain adaptation and knowledge distillation and uses propofol and remifentanil doses at historical moments to continuously predict the bispectral index (BIS) is proposed in this paper. Specifically, a modified adaptive recurrent neural network (AdaRNN) is adopted to address data distribution differences among patients. Moreover, a knowledge distillation pipeline is developed to train the prediction network by enabling it to learn intermediate feature representations of the teacher network. The experimental results show that our method exhibits better performance than existing approaches during all anesthetic phases in the TCI of propofol and remifentanil intravenous anesthesia. In particular, our method outperforms some state-of-the-art methods in terms of root mean square error and mean absolute error by 1 and 0.8, respectively, in the internal dataset as well as in the publicly available dataset.
... Nevertheless, the sample size is small to generalize the algorithm. Afshar et al. [8] presented a hybrid deep-learning method based on the convolutional neural networks, bidirectional long short-term memory, and an attention layer. The authors reported the mean correlation of 0.8 between the reference and estimated BIS values. ...
... On one hand, compared to [7] that only 10 subjects were used for the analysis, the proposed algorithm achieved a similar mean of CC with 18 subjects (0.83 vs 0.85) for the intra-subject variability. On the other hand, when comparing our results versus [8], we obtained a higher mean of CC (0.87 vs. 0.80) and a lower mean of AE (5.5 vs. 6.03) for the inter-subject variability. ...
... Deep learning and machine learning methods are proposed to help anesthesiologists make decisions about anesthesia, such as inference of brain states under anesthesia [19] and ultrasound image guidance [20]. Various indicators of postoperative anesthesia help anesthesiologists better monitor patients' vital signs after surgery, ensure smooth and safe recovery of patient consciousness during the awakening period, and strive to reduce complications during the awakening period [21]. ...
... Firstly, previous work fails to predict the recovery time from anesthesia for anesthesiologists. Although previous methods inform decision-making for anesthesiologists, these approaches do not focus on anesthesia recovery time to inform decision-making [3,21]. This paper seeks an effective method to help anesthesiologists estimate the recovery time from anesthesia for each patient. ...
It is significant for anesthesiologists to have a precise grasp of the recovery time of the patient after anesthesia. Accurate prediction of anesthesia recovery time can support anesthesiologist decision-making during surgery to help reduce the risk of surgery in patients. However, effective models are not proposed to solve this problem for anesthesiologists. In this paper, we seek to find effective forecasting methods. First, we collect 1824 patient anesthesia data from the eye center and then performed data preprocessing. We extracted 85 variables to predict recovery time from anesthesia. Second, we extract anesthesia information between variables for prediction using machine learning methods, including Bayesian ridge, lightGBM, random forest, support vector regression, and extreme gradient boosting. We also design simple deep learning models as prediction models, including linear residual neural networks and jumping knowledge linear neural networks. Lastly, we perform a comparative experiment of the above methods on the dataset. The experiment demonstrates that the machine learning method performs better than the deep learning model mentioned above on a small number of samples. We find random forest and XGBoost are more efficient than other methods to extract information between variables on postoperative anesthesia recovery time.
... El objetivo principal actual del estudio de la IA en anestesia es mejorar el manejo clínico prediciendo con precisión posibles complicaciones y sugiriendo estrategias terapéuticas óptimas en tiempo real [4]. La investigación actual en medicina perioperatoria esta centrada principalmente en administración automática de anestésicos, también conocida como anestesia de circuito cerrado [5], modelos que puedan estimar con precisión la profundidad de la anestesia y emersión [6], predicción de vía aérea intubación endotraqueal difícil [7], hipotensión perioperatoria, complicaciones posoperatorias, como delirio posoperatorio [8], medicina transfusional perioperatoria y predicción del riesgo de hemorragia [9] (Tabla 1). ...
The development of advanced Artificial Intelligence (AI) models raises various challenges and doubts about applications, benefits, limitations and possible complications of the use of this disruptive technology. Rapid advances in AI have led to analysis, diagnosis, management and prediction applications in anesthesiology in various areas, mainly: preoperative assessment and perioperative risk prediction; intraoperative monitoring and drug administration; and postoperative, allowing the anesthesiologist to adopt a proactive approach in the prevention, approach, resolution of crisis situations in the intraoperative and monitoring systems in the postoperative. The adaptation of AI algorithms and tools in anesthesia offers great potential that must be exploited under ethical precepts and considerations, in data analysis with predictive capabilities, optimization of strategies and automated assistance focused on patient safety.
... ML encompasses numerous algorithms that can be employed to create a robust index for evaluating the DoA [35]. In several studies [36,37,38,39,40] assessment of DoA was performed with artificial neural networks using EEG features. Automated methods used in these studies outperformed the use of BIS. ...
... Moreover, flexibility is required to customize computed features and their relationships to patients' or animal species' specificities. Machine learning (ML) and deep learning (DL) strategies have been proposed to aid in this endeavor (Afshar et al., 2021;Schmierer et al., 2024). Crucially, feature selection is an initial step in model optimization and conservation of computational resources. ...
Introduction: In the medical and veterinary fields, understanding the significance of physiological signals for assessing patient state, diagnosis, and treatment outcomes is paramount. There are, in the domain of machine learning (ML), very many methods capable of performing automatic feature selection. We here explore how such methods can be applied to select features from electroencephalogram (EEG) signals to allow the prediction of depth of anesthesia (DoA) in pigs receiving propofol. Methods: We evaluated numerous ML methods and observed that these algorithms can be classified into groups based on similarities in selected feature sets explainable by the mathematical bases behind those approaches. We limit our discussion to the group of methods that have at their core the computation of variances, such as Pearson's and Spearman's correlations, principal component analysis (PCA), and ReliefF algorithms. Results: Our analysis has shown that from an extensive list of time and frequency domain EEG features, the best predictors of DoA were spectral power (SP), and its density ratio applied specifically to high-frequency intervals (beta and gamma ranges), as well as burst suppression ratio, spectral edge frequency and entropy applied to the whole spectrum of frequencies. Discussion: We have also observed that data resolution plays an essential role not only in feature importance but may impact prediction stability. Therefore, when selecting the SP features, one might prioritize SP features over spectral bands larger than 1 Hz, especially for frequencies above 14 Hz. KEYWORDS
... One of the advantages of using magnetic field rather than electrical signals like electroencephalogram (EEG) [2][3][4][5] is that it could be computing outside the head without any interference and distortion. This property of magnetic field provides the ability to accurately determining the location of dipoles inside the brain. ...
... 11 To this aim, several food and drug administration (FDA)-approved 12 systems are available to monitor DoA based on a straightforward 13 numerical scale. These systems mostly employ multichannel elec- ranging from 0 to 100, where 0 signifies the absence of brain activity, 17 and 100 corresponds to a state of complete awareness [3]. Nonetheless, 18 the current commercial DoA systems are typically expensive and 19 require using costly disposable electrodes for each subject [4] alone [9]. ...
While entropy metrics derived from electroencephalography (EEG) have shown significant promise in monitoring the depth of anesthesia (DoA), the applicability of fuzzy entropy (FuzzEn) initially proposed to address the limitations of conventional entropy metrics regarding the sample size and class boundaries, remains unexplored in this context. This paper addresses two primary objectives: proposing a new method for DoA monitoring using a fusion of FuzzEn with Gaussian and Exponential membership functions, specifically designed for a wearable single frontal EEG channel system, and evaluating its comparative effectiveness against other entropy metrics. First, the EEG signal undergoes denoising and is then decomposed into sub-bands. Second, seven entropy metrics, including FuzzEn, are extracted from each sub-band. Lastly, each set of individual entropy metrics obtained from all sub-bands is separately fed into a regressor. In direct comparison with other metrics like sample entropy and approximate entropy, the proposed fused FuzzEn exhibited clear superiority with a higher mean correlation coefficient (0.85 vs. 0.63, 0.61) and lower mean absolute error (5.4 vs. 8.7, 8.9) between the reference and estimated DoA index values. The obtained results underscore the potential of the proposed FuzzEn for DoA monitoring.
... In [38], the author employed NNs to acquire direct-to-reverberant ratio (DRR) for classifying various time-frequency (T-F) bins and extracted relevant information based on the obtained classifications. In general, combining DOA estimation with machine learning has become a research hotspot for achieving more accurate results [1,24,33,36,38]. ...
With the development of deep learning techniques, the field of direction of arrival (DOA) estimation has also made significant progress. However, the accuracy of DOA estimation using end-to-end neural networks (NNs) heavily relies on the classification step of the networks, which necessitates the use of large and representative datasets. Additionally, conventional speech presence probability (SPP) estimation methods based on the ideal ratio mask (IRM) may misclassify time-frequency (T-F) bins dominated by non-speech and noise, which hinders the accurate extraction of directional information. To improve the robustness of existing DOA estimation algorithms, this paper proposes a DOA estimation method with T-F bin selection. In terms of output, instead of using IRM-based SPP, our proposed approach focuses on the a posteriori SPP, a deliberate choice aimed at circumventing potential confusion. For input optimization, we construct features that encompass spatial, temporal, and directional information concurrently, and these are coupled with a frequency bin-wise recurrent neural network (RNN) model to attain precise multi-channel SPP estimation. Subsequently, these SPP estimates are utilized to extract local information for DOA estimation. Moreover, the cascaded structure ensures that the model has the ability to complete out-of-label tasks, effectively reducing the dataset requirements by training only a subset of direction information to achieve omnidirectional DOA estimation. Besides, this contributes to the algorithm’s ability to eliminate its reliance on the step size, setting it apart from other end-to-end methods. Simulation results validate that the proposed method achieves higher accuracy and lower error compared to both NN-based end-to-end approaches and traditional full-band approaches under various conditions of reverberation and signal-to-noise ratio.
... In other words, extracting feature vectors from diferent sources of data and arranging them into a long feature vector is not feature fusion. Feature concatenation methods highly afect the number of training parameters in several classifers such as Bayes classifer [10] or deep learning schemes [11,12]. In the case of a small sample size problem, the covariance of such a small dataset is underestimated. ...
There is growing interest in developing linear/nonlinear feature fusion methods that fuse the elicited features from two different sources of information for achieving a higher recognition rate. In this regard, canonical correlation analysis (CCA), cross-modal factor analysis, and probabilistic CCA (PCCA) have been introduced to better deal with data variability and uncertainty. In our previous research, we formerly developed the kernel version of PCCA (KPCCA) to capture both nonlinear and probabilistic relation between the features of two different source signals. However, KPCCA is only able to estimate latent variables, which are statistically correlated between the features of two independent modalities. To overcome this drawback, we propose a kernel version of the probabilistic dependent-independent CCA (PDICCA) method to capture the nonlinear relation between both dependent and independent latent variables. We have compared the proposed method to PDICCA, CCA, KCCA, cross-modal factor analysis (CFA), and kernel CFA methods over the eNTERFACE and RML datasets for audio-visual emotion recognition and the M2VTS dataset for audio-visual speech recognition. Empirical results on the three datasets indicate the superiority of both the PDICCA and Kernel PDICCA methods to their counterparts.
... It is considered more feasible, and cost-effective in terms of real-time BCI implementation (Kiranyaz et al., 2021). Several EEG time series applications like seizure detection (Anuragi et al., 2021(Anuragi et al., , 2022Mehla et al., 2021;Rajendra Acharya et al., 2018;Sharma et al., 2020), epilepsy diagnosis (Abdulhay et al., 2020;Nishad & Pachori, 2020;Serna et al., 2020), subject identification (Rathee et al., 2022;Sun et al., 2019), drowsing states during driving (Doniec et al., 2020), depth-of-anesthesia (Afshar et al., 2021;Altıntop et al., 2022), human activity recognition (Garima, Goel & Rathee, 2023;Lin & Jianning, 2020), and so forth, of 1D-CNN architecture variants have been proposed in order to detect, and treat the abnormal brain activities at an early stage. ...
Precise monitoring and diagnosis of epilepsy by manual analysis of EEG signals are challenging due to the low doctor‐to‐patient ratio, and shortage of medical resources. To automate this diagnosis in real‐time, EEG based Brain–Computer Interface (BCI) system with integration of artificial intelligence techniques will prove to be propitious. This work proposes an end‐to‐end, one‐dimensional atrous conv‐net‐based architecture for automatic epilepsy diagnosis using EEG signals with a conceptual framework of the EEG‐BCI system for routine monitoring and clinical use. The proposed architecture has a robust backbone of six blocks of atrous convolutional layers activated with exponential linear unit functions. The six blocks are followed by the addition of a long short‐term memory layer for automatic feature extraction and sequential EEG data analysis. The efficacy of the proposed architecture has been verified on three publicly available EEG datasets using various evaluation metrics, feature maps, test set evaluation, and ablation studies. An average training and validation accuracy of 96.16% and 90.80% has been achieved upon multiple runs for the three datasets. Ablation experiments indicate that the addition of each block contributed to increasing 17%–25% accuracy scores during the classification of epileptic and non‐epileptic EEG signals. The real‐time EEG‐BCI has been analyzed using weight optimization of the proposed architecture through the NVIDIA Tensor RT framework on a 40 GB DGX A100 NVIDIA workstation. The proposed architecture has generalized well in comparison with the existing techniques for the three EEG datasets and achieved a low training and validation loss with optimum evaluation metrics. This makes the proposed architecture suitable for future EEG‐BCI system deployment in the automatic diagnosis of epilepsy.
... Afshar et al. [13] proposed a new deep learning structure that uses multiple features from 35 patients EEG signals to continuously predict the BIS value, achieving an accuracy of 88.71% and an improvement in area under the curve (AUC) of 15% on average, when compared to traditional DoA estimation methods. On a different approach, Jiang et al. [14] uses EEG signals, pre-analysed through sample entropy as an input to train an ANN model that tries to provide a valuable reference to DoA. ...
Application of artificial intelligence (AI) in medicine is quickly expanding. Despite the amount of evidence and promising results, a thorough overview of the current state of AI in clinical practice of anesthesiology is needed. Therefore, our study aims to systematically review the application of AI in this context.
A systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We searched Medline and Web of Science for articles published up to November 2022 using terms related with AI and clinical practice of anesthesiology. Articles that involved animals, editorials, reviews and sample size lower than 10 patients were excluded. Characteristics and accuracy measures from each study were extracted.
A total of 46 articles were included in this review. We have grouped them into 4 categories with regard to their clinical applicability: (1) Depth of Anesthesia Monitoring; (2) Image-guided techniques related to Anesthesia; (3) Prediction of events/risks related to Anesthesia; (4) Drug administration control. Each group was analyzed, and the main findings were summarized. Across all fields, the majority of AI methods tested showed superior performance results compared to traditional methods.
AI systems are being integrated into anesthesiology clinical practice, enhancing medical professionals’ skills of decision-making, diagnostic accuracy, and therapeutic response.
... Other areas of interest include postoperative complications, such as postoperative delirium [55], perioperative transfusion medicine, and predicting the risk of bleeding [56]. Additionally, researchers are investigating models that can accurately estimate the depth of anesthesia [57] and emergence timing [30,58], and can be useful to predict difficult endotracheal intubation [59] and perioperative hypotension [60]. These research areas demonstrate that the primary objective of studying AI in anesthesia is to improve clinical management by accurately predicting potential complications and suggesting optimal therapeutic strategies in real time [61]. ...
The field of anesthesia has always been at the forefront of innovation and technology, and the integration of Artificial Intelligence (AI) represents the next frontier in anesthesia care. The use of AI and its subtypes, such as machine learning, has the potential to improve efficiency, reduce costs, and ameliorate patient outcomes. AI can assist with decision making, but its primary advantage lies in empowering anesthesiologists to adopt a proactive approach to address clinical issues. The potential uses of AI in anesthesia can be schematically grouped into clinical decision support and pharmacologic and mechanical robotic applications. Tele-anesthesia includes strategies of telemedicine, as well as device networking, for improving logistics in the operating room, and augmented reality approaches for training and assistance. Despite the growing scientific interest, further research and validation are needed to fully understand the benefits and limitations of these applications in clinical practice. Moreover, the ethical implications of AI in anesthesia must also be considered to ensure that patient safety and privacy are not compromised. This paper aims to provide a comprehensive overview of AI in anesthesia, including its current and potential applications, and the ethical considerations that must be considered to ensure the safe and effective use of the technology.
... Other areas of interest include postoperative complications, such as postoperative delirium [49], perioperative transfusion medicine, and predicting the risk of bleeding [50]. Additionally, researchers are investigating models that can accurately estimate the depth of anesthesia [51], predict difficult endotracheal intubation [52], and identify perioperative hypotension [53]. These research areas demonstrate that the primary objective of studying AI in anesthesia is to improve clinical management by accurately predicting potential complications and suggesting optimal therapeutic strategies in real-time [54]. ...
The field of anesthesia has always been at the forefront of innovation and technology, and the integration of Artificial Intelligence (AI) represents the next frontier in anesthesia care. The use of AI and its subtypes such as machine learning has the potential to improve efficiency, reduce costs, and improve patient outcomes. AI can assist with decision-making, but its primary advantage lies in empowering anesthesiologists to adopt a proactive approach to address clinical issues. However, further research and validation are needed to fully understand the benefits and limitations of these applications in clinical practice. Moreover, the ethical implications of AI in anesthesia must also be considered to ensure that patient safety and privacy are not compromised. This paper aims to provide a comprehensive overview of AI in anesthesia, including its current and potential applications, and the ethical considerations that must be considered to ensure the safe and effective use of the technology.
... Methods for automatic anesthetic administration (closed-loop anesthesia) are deeply investigated [18,19]. Postoperative complications such as postoperative delirium [20], perioperative transfusion medicine, and prediction of bleeding risk [21] as well as models to perform accurate estimation of the depth of anesthesia [22], automatic prediction of difficult endotracheal intubation [23], and perioperative hypotension [24] are hot research topics. These aspects underline that the main interest of the study of AI in anesthesia is the improvement of clinical management, accurately predicting possible complications, and suggesting optimal therapeutic strategies in real-time [25]. ...
Background: The scientific literature on Artificial Intelligence (AI) in anesthesia is rapidly growing. Considering that applications of AI strategies can offer paramount support in clinical decision processes, it is crucial to delineate the research features. Bibliometric analyses can provide an overview of research tendencies useful for supplementary investigations in a research field. Methods: The comprehensive literature about AI in anesthesia was checked in the Web of Science (WOS) core collection. Year of publication, journal metrics including impact factor and quartile, title, document type, topic, and article metric (citations) were extracted. The software tool VOSviewer (version 1.6.17) was implemented for the co-occurrence of keywords and the co-citation analyses, and for evaluating research networks (countries and institutions). Results: Altogether, 288 documents were retrieved from the WOS and 154 articles were included in the analysis. The number of articles increased from 4 articles in 2017 to 37 in 2021. Only 34 were observational investigations and 7 RCTs. The most relevant topic is “anesthesia management”. The research network for countries and institutions shows severe gaps. Conclusion: Research on AI in anesthesia is rapidly developing. Further clinical studies are needed. Although different topics are addressed, scientific collaborations must be implemented.
... In order to take use of all that CNN has to offer, we first transform the 1-D ICS flow data into the two-dimensional grayscale pictures that CNN prefers. [4] In order to categorise brain states when under anaesthetics, this article relies on a deep neural network model (called nes-MetaNet). The Anes-MetaNet combines a Convolutional Neural Network (CNN) for extracting power spectral characteristics, a time importance model based on (LSTM) networks. ...
This article provides a comprehensive review of recent developments in deep learning, including both its theoretical underpinnings and its innovative practical applications. This article provides a historical overview of the significant uses of deep learning algorithms. Comparing the deep learning approach with more traditional algorithms, this article highlights the method's advantages and benefits. This includes the deep learning approach's layer-based hierarchy and nonlinear operations. The state-of the-art review also gives a high-level explanation of deep learning's advantages and growth in popularity. With deep learning, meaningful hierarchical linkages within the data can be discovered algorithmically without the need for painstaking hand-crafting of features, which is especially useful in the era of medical big data. We discuss the state-of-the-art in medical picture segmentation, localization, detection, and registration, as well as its most important applications.
Accurate assessment of consciousness during general anesthesia is crucial for optimizing anesthetic dosage and patient safety. Current electroencephalogram-based monitoring devices can be inaccurate or unreliable in specific surgical contexts ( e . g . cephalic procedures). This study investigated the feasibility of using electrocardiogram (ECG) features and machine learning to differentiate between awake and anesthetized states. A cohort of 48 patients undergoing surgery under general anesthesia at the Tours hospital was recruited. ECG-derived features were extracted, including spectral power, heart rate variability and complexity metrics, as well as heart rate fragmentation indices (HRF). These features were augmented by a range of physiological variables. The aim was to evaluate a number of machine learning algorithms in order to identify the most appropriate method for classifying the awake and anesthetized states. The gradient boosting algorithm achieved the highest accuracy (0.84). Notably, HRF metrics exhibited the strongest predictive power across all models tested. The generalizability of this ECG-based approach was further assessed using public datasets (VitalDB, Fantasia, and MIT-BIH Polysomnographic), achieving accuracies above 0.80. This study provides evidence that ECG-based classification methods can effectively distinguish awake from anesthetized states, with HRF indices playing a pivotal role in this classification.
Author summary
General anesthesia monitoring is critical for optimizing patient safety and outcomes. While electroencephalogram (EEG)-based systems are commonly used, they have limitations in accuracy and applicability, particularly in cases where EEG electrodes placement is challenging or impossible, such as during cephalic surgeries or when patients have forehead skin lesions. Here, a novel approach using electrocardiogram (ECG) signals and machine learning techniques was used to differentiate between awake and anesthetized states during surgery. A total of 48 patients undergoing surgical procedures under general anaesthesia at the Tours hospital were selected for inclusion in the study. This investigation focused on heart rate fragmentation indices, metrics designed for assessing biological versus chronological age, derived from ECG recordings. The gradient boosting algorithm demonstrates performance comparable to leading methods reported in the literature for this classification task. Importantly, model generalizability was confirm through successful application to publicly available datasets. This article highlights the potential of ECG signals as an alternative source for deriving depth of anesthesia indices, offering increased versatility in clinical settings where EEG monitoring is challenging or contraindicated.
General anesthesia typically involves three key components: amnesia, analgesia, and immobilization. Monitoring the depth of anesthesia (DOA) during surgery is crucial for personalizing anesthesia regimens and ensuring precise drug delivery. Since general anesthetics act primarily on the brain, this organ becomes the target for monitoring DOA. Electroencephalogram (EEG) can record the electrical activity generated by various brain tissues, enabling anesthesiologists to monitor the DOA from real‐time changes in a patient's brain activity during surgery. This monitoring helps to optimize anesthesia medication, prevent intraoperative awareness, and reduce the incidence of cardiovascular and other adverse events, contributing to anesthesia safety. Different anesthetic drugs exert different effects on the EEG characteristics, which have been extensively studied in commonly used anesthetic drugs. However, due to the limited understanding of the biological basis of consciousness and the mechanisms of anesthetic drugs acting on the brain, combined with the effects of various factors on existing EEG monitors, DOA cannot be accurately expressed via EEG. The lack of patient reactivity during general anesthesia does not necessarily indicate unconsciousness, highlighting the importance of distinguishing the mechanisms of consciousness and conscious connectivity when monitoring perioperative anesthesia depth. Although EEG is an important means of monitoring DOA, continuous optimization is necessary to extract characteristic information from EEG to monitor DOA, and EEG monitoring technology based on artificial intelligence analysis is an emerging research direction.
For accurate Direction of Arrival (DOA) estimation in challenging high reverberation and low signal-to-noise ratio (SNR) scenarios, various deep learning (DL) techniques have been developed to incorporate and enhance existing algorithms. However, addressing the challenges of enhancing network manipulability, constructing efficient learning features, and integrating algorithms in a rational manner remains a set of significant hurdles. In this paper, we use DL to obtain the Speech Presence Probability (SPP) to construct the optimal statistics estimates, which are then combined with traditional algorithms to achieve accurate DOA estimation. Specifically, we explore the application of the
a posteriori
SPP in DOA estimation, design a reverberation separation model for practical scenarios, and derive and validate a new computational equation for SPP under this model. Besides, we propose a frequency bin-wise network structure to improve network fitting efficiency and construct input features accordingly. Moreover, by adopting a combined structure, we avoid full-angle network feature training and instead train on partial angles under deliberate subset classification. We then evaluate the DOA estimation performance for the entire direction range with fine resolution using this approach. Simulation results demonstrate that the proposed method requires smaller data sets compared to end-to-end deep learning algorithms. Furthermore, the results validate that the proposed method outperforms both DL-based end-to-end approaches and traditional full-band approaches in terms of accuracy and error rate across various reverberation and signal-to-noise ratio conditions.
Clinical manifestations and standard psychological tests have been widely used to diagnose autism spectrum disorder (ASD) patients and evaluate their severity level. The gold-standard criterion to diagnose ASD patients is the childhood autism rating scale (CARS), which is a qualitative questionnaire that is filled out through a systematic interview while no physiological test/record is performed to determine this score. To make the diagnosis process quantitative, electroencephalography (EEG) signals have been repeatedly analyzed to differentiate healthy subjects from ASD patients. However, the precise relationship between the abnormal behavior of EEG signals and different ASD severity levels is not well investigated. Here, we use CARS to qualitatively determine the severity level of 14 autistic children, who voluntarily enrolled in our study. We recorded their EEG signals from 19 scalp channels when they were awake in the idle state and elicited three informative features including approximation entropy, multiscale entropy and sample entropy in successive time frames. Among the three measures of entropy, the last one exhibits the highest sensitivity, where its correlation coefficient (CC) exceeds 0.7, on the electrode positions T6 (CC [Formula: see text] 0.74), P4 (CC [Formula: see text] 0.76) and Cz (CC [Formula: see text] 0.75). Results of sample entropy in channels Cz-versus-Pz and P4-versus-FP2 show that a simple K-nearest neighbor classifier can provide 93% classification accuracy among patients with mild, moderate and severe ASD levels. Comparing the proposed method to the conventional ones, we also extracted power spectral density features from the channels but they failed to identify the ASD severity level with an acceptable accuracy.
Artificial intelligence applications have great potential to enhance perioperative care. This article explores promising areas for artificial intelligence in anesthesiology; expertise, stakeholders, and infrastructure for development; and barriers and challenges to implementation.
Objective: Monitoring the depth of anaesthesia (DOA) during surgery is of critical importance. However, during surgery electroencephalography (EEG) is usually subject to various disturbances that affect the accuracy of DOA. Therefore, accurately estimating noise in EEG and reliably assessing DOA remains an important challenge. In this paper, we proposed a signal quality index (SQI) network (SQINet) for assessing the EEG signal quality and a DOA network (DOANet) for analyzing EEG signals to precisely estimate DOA. The two networks are termed SQI-DOANet.
Approach: The SQINet contained a shallow convolutional neural network to quickly determine the quality of the EEG signal. The DOANet comprised a feature extraction module for extracting features, a dual attention module for fusing multi-channel and multi-scale information, and a gated multilayer perceptron module for extracting temporal information. The performance of the SQI-DOANet model was validated by training and testing the model on the large VitalDB database, with the bispectral index (BIS) as the reference standard.
Main results: The proposed DOANet yielded a Pearson correlation coefficient with the BIS score of 0.88 in the 5-fold cross-validation, with a mean absolute error (MAE) of 4.81. The mean Pearson correlation coefficient of SQI-DOANet with the BIS score in the 5-fold cross-validation was 0.82, with an MAE of 5.66.
Significance: The SQI-DOANet model outperformed three compared methods. The proposed SQI-DOANet may be used as a new deep learning method for DOA estimation. The code of the SQI-DOANet will be made available publicly at https://github.com/YuRui8879/SQI-DOANet.
: Anesthesia is the process of inducing and experiencing various conditions, such as painlessness, immobility, and amnesia, to facilitate surgeries and other medical procedures. During the administration of anesthesia, anesthesiologists face critical decision-making moments, considering the significance of the procedure and potential complications resulting from anesthesia-related choices. In recent years, artificial intelligence (AI) has emerged as a supportive tool for anesthesia decisions, given its potential to assist with control and management tasks. This study aims to conduct a comprehensive review of articles on the intersection of AI and anesthesia. A review was conducted by searching PubMed for peer-reviewed articles published between 2020 and early 2022, using keywords related to anesthesia and AI. The articles were categorized into nine distinct groups: "Depth of anesthesia", "Control of anesthesia delivery", "Control of mechanical ventilation and weaning", "Event prediction", "Ultrasound guidance", "Pain management", "Operating room logistic", "Monitoring", and "Neuro-critical care". Four reviewers meticulously examined the selected articles to extract relevant information. The studies within each category were reviewed by considering items such as the purpose and type of anesthesia, AI algorithms, dataset, data accessibility, and evaluation criteria. To enhance clarity, each category was analyzed with a higher resolution than previous review articles, providing readers with key points, limitations, and potential areas for future research to facilitate a better understanding of each concept. The advancements in AI techniques hold promise in significantly enhancing anesthesia practices and improving the overall experience for anesthesiologists.
Accurate monitoring of anesthesia status is very important in surgery, as it can guide anesthesiologists, reduce drug usage, and reduce postoperative adverse effects. However, due to the complex interactions between anesthetic drugs and the central nervous system, there is no perfect monitoring method. In recent years, the development of artificial intelligence technology has offered the possibility of using machine learning algorithms to achieve more accurate monitoring of anesthesia depth. In this paper, four levels of anesthesia states were classified and multifaceted feature values were extracted from Electroencephalogram (EEG) signals, a convolutional neural network-based KRDGB-CNN model was constructed, which was based on K-nearest neighbor (KNN), Random Forest (RF), Decision Tree (DT), Gaussian Naive Baye (GNB), and Back propagation Neural Network (BP), and fused by Convolutional Neural Network (CNN) algorithm for decision layers. By evaluating the model performance on the collected data, the results show that the model outperforms existing algorithms in terms of classification accuracy and specificity, and can effectively improve the robustness and accuracy of the algorithm.
The performance of Adaboost is highly sensitive to noisy and outlier samples. This is therefore the weights of these samples are exponentially increased in successive rounds. In this paper, three novel schemes are proposed to hunt the corrupted samples and eliminate them through the training process. The methods are: I) a hybrid method based on K-means clustering and K-nearest neighbor, II) a two-layer Adaboost, and III) soft margin support vector machines. All of these solutions are compared to the standard Adaboost on thirteen Gunnar Raetsch’s datasets under three levels of class-label noise. To test the proposed method on a real application, electroencephalography (EEG) signals of 20 schizophrenic patients and 20 age-matched control subjects, are recorded via 20 channels in the idle state. Several features including autoregressive coefficients, band power and fractal dimension are extracted from EEG signals of all participants. Sequential feature subset selection technique is adopted to select the discriminative EEG features. Experimental results imply that exploiting the proposed hunting techniques enhance the Adaboost performance as well as alleviating its robustness against unconfident and noisy samples over Raetsch benchmark and EEG features of the two groups.
Many patients suffer from postoperative pain after surgery, which causes discomfort and influences recovery after the operation. During surgery, the anesthetists usually rely on their own experience when anesthetizing, which is not stable for avoiding postoperative pain. Hence, it is essential to predict postoperative pain and give proper doses accordingly. Recently, the relevance of various clinical parameters and nociception has been investigated in many works, and several indices have been proposed for measuring the level of nociception. However, expensive advanced equipment is required when applying advanced medical technologies, which is not accessible to most institutions. In our work, we propose a deep learning model based on a dynamic graph transformer framework named DoseFormer to predict postoperative pain in a short period after an operation utilizing dynamic patient data recorded in existing widely utilized equipment (e.g., anesthesia monitor). DoseFormer consists of two modules: (i) We design a temporal model utilizing a long short-term memory (LSTM) model with an attention mechanism to capture dynamic intraoperative data of the patient and output a hybrid semantic embedding representing the patient information. (ii) We design a graph transformer network (GTN) to infer the postoperative pain level utilizing the relations across the patient embeddings. We evaluate the DoseFormer system with the medical records of over 999 patients undergoing cardiothoracic surgery in the Fourth Affiliated Hospital of Zhejiang University School of Medicine. The experimental results show that our model achieves 92.16% accuracy for postoperative pain prediction and has a better comprehensive performance compared with baselines.
Improper estimation of depth of anaesthesia (DoA) during surgical procedure may lead to intraoperative awareness and is associated with postoperative complications such as hypotension and hypoperfusion of heart and brain. Therefore, monitoring DoA is critical in these surgical operations. In this paper, a machine learning based DoA estimation is proposed using six electroencephalogram (EEG) features including five spectral and one statistical feature. A decision tree classifier is trained with EEG and Bispectral Index (BIS) data from 100 patients using the extracted features. The developed model showed an accuracy of 56.65%. The accuracy improved to 59% after deploying grid search and gradient boost methods.KeywordsBISDepth of anaesthesiaEEGFeature extractionMachine learning
In this study, a novel approach for feature selection has been presented in order to overcome the challenge of classifying positive and negative risk prediction in cryptocurrency market which contains high fluctuation. This approach is based on maximizing information gain with simultaneous minimizing the similarity of selected features to achieve a proper feature set for improving the classification accuracy. The proposed method was compared to other feature selection techniques such as sequential and bidirectional feature selection, univariate feature selection and least absolute shrinkage and selection operator. To evaluate the feature selection techniques, several classifiers were employed such as XGBoost, k-nearest neighbor, support vector machine, random forest, logistic regression, long short-term memory, and deep neural networks. The features were elicited from the time series of Bitcoin, Binance, and Ethereum cryptocurrencies. The results of applying the selected features to different classifiers indicated that XGBoost and random forest provided better results on the time series datasets. Furthermore, the proposed feature selection method achieved the best results on two (out of three) cryptocurrencies. The accuracy in the best state varied between 55% to 68% for different time series. It is worth mentioning that preprocessed features were used in this research, meaning that raw data (candle data) was used to derive efficient features that can explain the problem and help the classifiers in predicting the labels.
Background:
Proper maintenance of hypnosis is crucial for ensuring the safety of patients undergoing surgery. Accordingly, indicators, such as the Bispectral index (BIS), have been developed to monitor hypnotic levels. However, the black-box nature of the algorithm coupled with the hardware makes it challenging to understand the underlying mechanisms of the algorithms and integrate them with other monitoring systems, thereby limiting their use.
Objective:
We propose an interpretable deep learning model that forecasts BIS values 25 s in advance using 30 s electroencephalogram (EEG) data.
Material and methods:
The proposed model utilized EEG data as a predictor, which is then decomposed into amplitude and phase components using fast Fourier Transform. An attention mechanism was applied to interpret the importance of these components in predicting BIS. The predictability of the model was evaluated on both regression and binary classification tasks, where the former involved predicting a continuous BIS value, and the latter involved classifying a dichotomous status at a BIS value of 60. To evaluate the interpretability of the model, we analyzed the attention values expressed in the amplitude and phase components according to five ranges of BIS values. The proposed model was trained and evaluated using datasets collected from two separate medical institutions.
Results and conclusion:
The proposed model achieved excellent performance on both the internal and external validation datasets. The model achieved a root-mean-square error of 6.614 for the regression task, and an area under the receiver operating characteristic curve of 0.937 for the binary classification task. Interpretability analysis provided insight into the relationship between EEG frequency components and BIS values. Specifically, the attention mechanism revealed that higher BIS values were associated with increased amplitude attention values in high-frequency bands and increased phase attention values in various frequency bands. This finding is expected to facilitate a more profound understanding of the BIS prediction mechanism, thereby contributing to the advancement of anesthesia technologies.
Objective:
Cardiac auscultation is the most practiced non-invasive and cost-effective procedure for the early diagnosis of heart diseases. While machine learning based systems can aid in automatically screening patients, the robustness of these systems is affected by numerous factors including the stethoscope/sensor, environment and data collection protocol. This paper studies the adverse effect of domain variability on heart sound abnormality detection and develops strategies to address this problem.
Methods:
We propose a novel Convolutional Neural Network (CNN) layer, consisting of time-convolutional (tConv) units, that emulate Finite Impulse Response (FIR) filters. The filter coefficients can be updated via backpropagation and be stacked in the front-end of the network as a learnable filterbank.
Results:
On publicly available multi-domain datasets, the proposed method surpasses the top-scoring systems found in the literature for heart sound abnormality detection (a binary classification task). We utilized sensitivity, specificity, F-1 score and Macc (average of sensitivity and specificity) as performance metrics. Our systems achieved relative improvements of up to 11.84% in terms of MAcc, compared to state-of-the-art methods.
Conclusion:
The results demonstrate the effectiveness of the proposed learnable filterbank CNN architecture in achieving robustness towards sensor/domain variability in PCG signals.
Significance:
The proposed methods pave the way for deploying automated cardiac screening systems in diversified and underserved communities.
The identification of sleep stages is essential in the diagnostics of sleep disorders, among which obstructive sleep apnea (OSA) is one of the most prevalent. However, manual scoring of sleep stages is time-consuming, subjective, and costly. To overcome this shortcoming, we aimed to develop an accurate deep learning approach for automatic classification of sleep stages and to study the effect of OSA severity on the classification accuracy. Overnight polysomnographic recordings from a public dataset of healthy individuals (Sleep-EDF, n=153) and from a clinical dataset (n=891) of patients with suspected OSA were used to develop a combined convolutional and long short-term memory neural network. On the public dataset, the model achieved sleep staging accuracy of 83.7% (κ=0.77) with a single frontal EEG channel and 83.9% (κ=0.78) when supplemented with EOG. For the clinical dataset, the model achieved accuracies of 82.9% (κ=0.77) and 83.8% (κ=0.78) with a single EEG channel and two channels (EEG+EOG), respectively. The sleep staging accuracy decreased with increasing OSA severity. The single-channel accuracy ranged from 84.5% (κ=0.79) for individuals without OSA diagnosis to 76.5% (κ=0.68) for severe OSA patients. In conclusion, deep learning enables automatic sleep staging for suspected OSA patients with high accuracy and expectedly, the accuracy lowered with increasing OSA severity. Furthermore, the accuracies achieved in the public dataset were superior to previously published state-of-the-art methods. Adding an EOG channel did not significantly increase the accuracy. The automatic, single-channel-based sleep staging could enable easy, accurate, and cost-efficient integration of EEG recording into diagnostic ambulatory recordings.
Bispectral index (BIS), a useful marker of anaesthetic depth, is calculated by a statistical multivariate model using nonlinear functions of electroencephalography-based subparameters. However, only a portion of the proprietary algorithm has been identified. We investigated the BIS algorithm using clinical big data and machine learning techniques. Retrospective data from 5,427 patients who underwent BIS monitoring during general anaesthesia were used, of which 80% and 20% were used as training datasets and test datasets, respectively. A histogram of data points was plotted to define five BIS ranges representing the depth of anaesthesia. Decision tree analysis was performed to determine the electroencephalography subparameters and their thresholds for classifying five BIS ranges. Random sample consensus regression analyses were performed using the subparameters to derive multiple linear regression models of BIS calculation in five BIS ranges. The performance of the decision tree and regression models was externally validated with positive predictive value and median absolute error, respectively. A four-level depth decision tree was built with four subparameters such as burst suppression ratio, power of electromyogram, 95% spectral edge frequency, and relative beta ratio. Positive predictive values were 100%, 80%, 80%, 85% and 89% in the order of increasing BIS in the five BIS ranges. The average of median absolute errors of regression models was 4.1 as BIS value. A data driven BIS calculation algorithm using multiple electroencephalography subparameters with different weights depending on BIS ranges has been proposed. The results may help the anaesthesiologists interpret the erroneous BIS values observed during clinical practice.
Context:
Electroencephalography (EEG) is a complex signal and can require several years of training, as well as advanced signal processing and feature extraction methodologies to be correctly interpreted. Recently, deep learning (DL) has shown great promise in helping make sense of EEG signals due to its capacity to learn good feature representations from raw data. Whether DL truly presents advantages as compared to more traditional EEG processing approaches, however, remains an open question.
Objective:
In this work, we review 154 papers that apply DL to EEG, published between January 2010 and July 2018, and spanning different application domains such as epilepsy, sleep, brain-computer interfacing, and cognitive and affective monitoring. We extract trends and highlight interesting approaches from this large body of literature in order to inform future research and formulate recommendations.
Methods:
Major databases spanning the fields of science and engineering were queried to identify relevant studies published in scientific journals, conferences, and electronic preprint repositories. Various data items were extracted for each study pertaining to (1) the data, (2) the preprocessing methodology, (3) the DL design choices, (4) the results, and (5) the reproducibility of the experiments. These items were then analyzed one by one to uncover trends.
Results:
Our analysis reveals that the amount of EEG data used across studies varies from less than ten minutes to thousands of hours, while the number of samples seen during training by a network varies from a few dozens to several millions, depending on how epochs are extracted. Interestingly, we saw that more than half the studies used publicly available data and that there has also been a clear shift from intra-subject to inter-subject approaches over the last few years. About [Formula: see text] of the studies used convolutional neural networks (CNNs), while [Formula: see text] used recurrent neural networks (RNNs), most often with a total of 3-10 layers. Moreover, almost one-half of the studies trained their models on raw or preprocessed EEG time series. Finally, the median gain in accuracy of DL approaches over traditional baselines was [Formula: see text] across all relevant studies. More importantly, however, we noticed studies often suffer from poor reproducibility: a majority of papers would be hard or impossible to reproduce given the unavailability of their data and code.
Significance:
To help the community progress and share work more effectively, we provide a list of recommendations for future studies and emphasize the need for more reproducible research. We also make our summary table of DL and EEG papers available and invite authors of published work to contribute to it directly. A planned follow-up to this work will be an online public benchmarking portal listing reproducible results.
One of the most challenging predictive data analysis efforts is accurate prediction of depth of anesthesia (DOA) indicators which has attracted a growing attention since it provides patients a safe surgical environment in case of secondary damage caused by intraoperative awareness or brain injury. However, many researchers put heavily handcraft feature extraction or carefully tailored feature engineering to each patient to achieve very high sensitivity and low false prediction rate for a particular dataset. This limits the benefit of the proposed approaches if a different dataset is used. Recently, representations learned using deep convolutional neural network (CNN) for object recognition are becoming widely used model of the processing hierarchy in the human visual system. The correspondence between models and brain signals that holds the acquired activity at high temporal resolution has been explored less exhaustively. In this paper, deep learning CNN with a range of different architectures, is designed for identifying related activities from raw electroencephalography (EEG). Specifically, an improved short-time Fourier transform (STFT) is used to stand for the time-frequency information after extracting the spectral images of the original EEG as input to CNN. Then CNN models are designed and trained to predict the DOA levels from EEG spectrum without handcrafted features, which presents an intuitive mapping process with high efficiency and reliability. As a result, the best trained CNN model achieved an accuracy of 93.50%, interpreted as CNN’s deep learning to approximate the DOA by senior anesthesiologists, which highlights the potential of deep CNN combined with advanced visualization techniques for EEG-based brain mapping.
In this work, we conducted a literature review about deep learning (DNN, RNN, CNN, and so on) for analyzing EEG data for decoding the activity of human's brain and diagnosing disease and explained details about various architectures for understanding the details of CNN and RNN. It has analyzed a word, which presented a model based on CNN and LSTM methods, and how these methods can be used to both optimize and set up the hyper parameters of deep learning architecture. Later, it is studied how semi‐supervised learning on EEG data analytics can be applied. We review some studies about different methods of semi‐supervised learning on EEG data analytics and discussing the importance of semi‐supervised learning for analyzing EEG data. In this paper, we also discuss the most common applications for human EEG research and review some papers about the application of EEG data analytics such as Neuromarketing, human factors, social interaction, and BCI. Finally, some future trends of development and research in this area, according to the theoretical background on deep learning, are given.
Guidelines are presented for safe practice in the use of intravenous drug infusions for general anaesthesia. When maintenance of general anaesthesia is by intravenous infusion, this is referred to as total intravenous anaesthesia. Although total intravenous anaesthesia has advantages for some patients, the commonest technique used for maintenance of anaesthesia in the UK and Ireland remains the administration of an inhaled volatile anaesthetic. However, the use of an inhalational technique is sometimes not possible, and in some situations, inhalational anaesthesia is contraindicated. Therefore, all anaesthetists should be able to deliver total intravenous anaesthesia competently and safely. For the purposes of simplicity, these guidelines will use the term total intravenous anaesthesia but also encompass techniques involving a combination of intravenous infusion and inhalational anaesthesia. This document is intended as a guideline for safe practice when total intravenous anaesthesia is being used, and not as a review of the pros and cons of total intravenous anaesthesia vs. inhalational anaesthesia in situations where both techniques are possible.
Estimating the depth of anaesthesia (DoA) in operations has always been a challenging issue due to the underlying complexity of the brain mechanisms. Electroencephalogram (EEG) signals are undoubtedly the most widely used signals for measuring DoA. In this paper, a novel EEG-based index is proposed to evaluate DoA for 24 patients receiving general anaesthesia with different levels of unconsciousness. Sample Entropy (SampEn) algorithm was utilised in order to acquire the chaotic features of the signals. After calculating the SampEn from the EEG signals, Random Forest was utilised for developing learning regression models with Bispectral index (BIS) as the target. Correlation coefficient, mean absolute error, and area under the curve (AUC) were used to verify the perioperative performance of the proposed method. Validation comparisons with typical nonstationary signal analysis methods (i.e., recurrence analysis and permutation entropy) and regression methods (i.e., neural network and support vector machine) were conducted. To further verify the accuracy and validity of the proposed methodology, the data is divided into four unconsciousness-level groups on the basis of BIS levels. Subsequently, analysis of variance (ANOVA) was applied to the corresponding index (i.e., regression output). Results indicate that the correlation coefficient improved to 0.72 ± 0.09 after filtering and to 0.90 ± 0.05 after regression from the initial values of 0.51 ± 0.17. Similarly, the final mean absolute error dramatically declined to 5.22 ± 2.12. In addition, the ultimate AUC increased to 0.98 ± 0.02, and the ANOVA analysis indicates that each of the four groups of different anaesthetic levels demonstrated significant difference from the nearest levels. Furthermore, the Random Forest output was extensively linear in relation to BIS, thus with better DoA prediction accuracy. In conclusion, the proposed method provides a concrete basis for monitoring patients’ anaesthetic level during surgeries.
The current anaesthesia information management system (AIMS) has limited capability for the acquisition of high-quality vital signs data. We have developed a Vital Recorder program to overcome the disadvantages of AIMS and to support research. Physiological data of surgical patients were collected from 10 operating rooms using the Vital Recorder. The basic equipment used were a patient monitor, the anaesthesia machine, and the bispectral index (BIS) monitor. Infusion pumps, cardiac output monitors, regional oximeter, and rapid infusion device were added as required. The automatic recording option was used exclusively and the status of recording was frequently checked through web monitoring. Automatic recording was successful in 98.5% (4,272/4,335) cases during eight months of operation. The total recorded time was 13,489 h (3.2 ± 1.9 h/case). The Vital Recorder's automatic recording and remote monitoring capabilities enabled us to record physiological big data with minimal effort. The Vital Recorder also provided time-synchronised data captured from a variety of devices to facilitate an integrated analysis of vital signs data. The free distribution of the Vital Recorder is expected to improve data access for researchers attempting physiological data studies and to eliminate inequalities in research opportunities due to differences in data collection capabilities.
Deep learning with convolutional neural networks (deep ConvNets) has revolutionized computer vision through end-to-end learning, that is, learning from the raw data. There is increasing interest in using deep ConvNets for end-to-end EEG analysis, but a better understanding of how to design and train ConvNets for end-to-end EEG decoding and how to visualize the informative EEG features the ConvNets learn is still needed. Here, we studied deep ConvNets with a range of different architectures, designed for decoding imagined or executed tasks from raw EEG. Our results show that recent advances from the machine learning field, including batch normalization and exponential linear units, together with a cropped training strategy, boosted the deep ConvNets decoding performance, reaching at least as good performance as the widely used filter bank common spatial patterns (FBCSP) algorithm (mean decoding accuracies 82.1% FBCSP, 84.0% deep ConvNets). While FBCSP is designed to use spectral power modulations, the features used by ConvNets are not fixed a priori. Our novel methods for visualizing the learned features demonstrated that ConvNets indeed learned to use spectral power modulations in the alpha, beta, and high gamma frequencies, and proved useful for spatially mapping the learned features by revealing the topography of the causal contributions of features in different frequency bands to the decoding decision. Our study thus shows how to design and train ConvNets to decode task-related information from the raw EEG without handcrafted features and highlights the potential of deep ConvNets combined with advanced visualization techniques for EEG-based brain mapping. Hum Brain Mapp, 2017.
With a massive influx of multimodality data, the role of data analytics in health informatics has grown rapidly in the last decade. This has also prompted increasing interests in the generation of analytical, data driven models based on machine learning in health informatics. Deep learning, a technique with its foundation in artificial neural networks, is emerging in recent years as a powerful tool for machine learning, promising to reshape the future of artificial intelligence. Rapid improvements in computational power, fast data storage and parallelization have also contributed to the rapid uptake of the technology in addition to its predictive power and ability to generate automatically optimized high-level features and semantic interpretation from the input data. This article presents a comprehensive up-to-date review of research employing deep learning in health informatics, providing a critical analysis of the relative merit and potential pitfalls of the technique as well as its future outlook. The paper mainly focuses on key applications of deep learning in the fields of translational bioinformatics, medical imaging, pervasive sensing, medical informatics and public health.
Recurrent neural networks (RNNs) have proven to be powerful models in
problems involving sequential data. Recently, RNNs have been augmented with
"attention" mechanisms which allow the network to focus on different parts of
an input sequence when computing their output. We propose a simplified model of
attention which is applicable to feed-forward neural networks and demonstrate
that it can solve some long-term memory problems (specifically, those where
temporal order doesn't matter). In fact, we show empirically that our model can
solve these problems for sequence lengths which are both longer and more widely
varying than the best results attained with RNNs.
Determining depth of anesthesia is a challenging problem in the context of biomedical signal processing. Various methods have been suggested to determine a quantitative index as depth of anesthesia, but most of these methods suffer from high sensitivity during the surgery. A novel method based on energy scattering of samples in the wavelet domain is suggested to represent the basic content of electroencephalogram (EEG) signal. In this method, first EEG signal is decomposed into different sub-bands, then samples are squared and energy of samples sequence is constructed through each scale and time, which is normalized and finally entropy of the resulted sequences is suggested as a reliable index. Empirical Results showed that applying the proposed method to the EEG signals can classify the awake, moderate and deep anesthesia states similar to BIS.
We propose a deep convolutional neural network architecture codenamed
"Inception", which was responsible for setting the new state of the art for
classification and detection in the ImageNet Large-Scale Visual Recognition
Challenge 2014 (ILSVRC 2014). The main hallmark of this architecture is the
improved utilization of the computing resources inside the network. This was
achieved by a carefully crafted design that allows for increasing the depth and
width of the network while keeping the computational budget constant. To
optimize quality, the architectural decisions were based on the Hebbian
principle and the intuition of multi-scale processing. One particular
incarnation used in our submission for ILSVRC 2014 is called GoogLeNet, a 22
layers deep network, the quality of which is assessed in the context of
classification and detection.
Neural machine translation is a recently proposed approach to machine
translation. Unlike the traditional statistical machine translation, the neural
machine translation aims at building a single neural network that can be
jointly tuned to maximize the translation performance. The models proposed
recently for neural machine translation often belong to a family of
encoder-decoders and consists of an encoder that encodes a source sentence into
a fixed-length vector from which a decoder generates a translation. In this
paper, we conjecture that the use of a fixed-length vector is a bottleneck in
improving the performance of this basic encoder-decoder architecture, and
propose to extend this by allowing a model to automatically (soft-)search for
parts of a source sentence that are relevant to predicting a target word,
without having to form these parts as a hard segment explicitly. With this new
approach, we achieve a translation performance comparable to the existing
state-of-the-art phrase-based system on the task of English-to-French
translation. Furthermore, qualitative analysis reveals that the
(soft-)alignments found by the model agree well with our intuition.
Recently, several attempts have been made to find the depth of anesthesia (DOA) by analyzing the ongoing electroencephalogram (EEG) signals during surgical operations. Nevertheless, specialists still do not rely on these indexes because they cannot accurately track the transitions of anesthetic depth. This paper presents an effective EEG-based index that is fast to compute and acts very accurate in practice. To determine the proposed index, first EEG signals are denoised with an adaptive thresholding method. The wavelet transform is then applied to the clean EEG signals in order to decompose the signal into brain-match subspaces and the proposed feature extracted from each subspace to monitor the DOA. EEG signals of 8 subjects were recorded during the surgical operation. Experimental results exhibit the proposed features highly correlated with the BIS index (the most popular EEG-based index)through different anesthetic levels. Moreover, in some cases the introduced index outperformed the BIS and the clinical observation confirmed this superiority.
Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O. 1. Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.
We studied the spectral characteristics of the EEG burst suppression patterns (BSP) of two intravenous anesthetics, propofol and thiopental. Based on the obtained results, we developed a method for automatic segmentation, classification and compact presentation of burst suppression patterns.
The spectral analysis was performed with the short time Fourier transform and with autoregressive modeling to provide information of frequency contents of bursts. This information was used when designing appropriate filters for segmentation algorithms. The adaptive segmentation was carried out using two different nonparametric methods. The first one was based on the absolute values of amplitudes and is referred to as the ADIF method. The second method used the absolute values of the Nonlinear Energy Operator (NLEO) and is referred to as the NLEO method. Both methods have been described earlier but they were modified for the purposes of BSP detection. The signal was classified to bursts, suppressions and artifacts. Automatic classification was compared with manual classification. Results. The NLEO method was more accurate, especially in the detection of artifacts. NLEO method classified correctly 94.0% of the propofol data and 92.8% of the thiopental data. With the ADIF method, the results were 90.5% and 88.1% respectively.
Our results show that burst suppression caused by the different anesthetics can be reliably detected with our segmentation and classification methods. The analysis of normal and pathological EEG, however, should include information of the anesthetic used. Knowledge of the normal variation of the EEG is necessary in order to detect the abnormal BSP of, for instance, seizure patients.
Recurrent neural networks can be used to map input sequences to output sequences, such as for recognition, production or prediction problems. However, practical difficulties have been reported in training recurrent neural networks to perform tasks in which the temporal contingencies present in the input/output sequences span long intervals. We show why gradient based learning algorithms face an increasingly difficult problem as the duration of the dependencies to be captured increases. These results expose a trade-off between efficient learning by gradient descent and latching on information for long periods. Based on an understanding of this problem, alternatives to standard gradient descent are considered.
In the first part of this paper, a regular recurrent neural
network (RNN) is extended to a bidirectional recurrent neural network
(BRNN). The BRNN can be trained without the limitation of using input
information just up to a preset future frame. This is accomplished by
training it simultaneously in positive and negative time direction.
Structure and training procedure of the proposed network are explained.
In regression and classification experiments on artificial data, the
proposed structure gives better results than other approaches. For real
data, classification experiments for phonemes from the TIMIT database
show the same tendency. In the second part of this paper, it is shown
how the proposed bidirectional structure can be easily modified to allow
efficient estimation of the conditional posterior probability of
complete symbol sequences without making any explicit assumption about
the shape of the distribution. For this part, experiments on real data
are reported
Background
The electrocardiogram (ECG) is one of the most commonly used diagnostic tools in medicine and healthcare. Deep learning methods have achieved promising results on predictive healthcare tasks using ECG signals.
Objective
This paper presents a systematic review of deep learning methods for ECG data from both modeling and application perspectives.
Methods
We extracted papers that applied deep learning (deep neural network) models to ECG data that were published between January 1st of 2010 and February 29th of 2020 from Google Scholar, PubMed, and the Digital Bibliography & Library Project. We then analyzed each article according to three factors: tasks, models, and data. Finally, we discuss open challenges and unsolved problems in this area.
Results
The total number of papers extracted was 191. Among these papers, 108 were published after 2019. Different deep learning architectures have been used in various ECG analytics tasks, such as disease detection/classification, annotation/localization, sleep staging, biometric human identification, and denoising.
Conclusion
The number of works on deep learning for ECG data has grown explosively in recent years. Such works have achieved accuracy comparable to that of traditional feature-based approaches and ensembles of multiple approaches can achieve even better results. Specifically, we found that a hybrid architecture of a convolutional neural network and recurrent neural network ensemble using expert features yields the best results. However, there are some new challenges and problems related to interpretability, scalability, and efficiency that must be addressed. Furthermore, it is also worth investigating new applications from the perspectives of datasets and methods.
Significance
This paper summarizes existing deep learning research using ECG data from multiple perspectives and highlights existing challenges and problems to identify potential future research directions.
This comprehensive, evidence-based book is intended to serve as a reference for medical practitioners involved in the perioperative care of neurosurgical patients. Fundamental aspects of neuroanesthesiology and neurocritical care are thoroughly examined across 101 chapters, outlining key elements that are crucial to a care provider’s knowledge of the practice. These elements include specific diagnostic procedures and treatment options, concepts and applicable details of the available neurosurgical interventions and techniques, and mechanisms necessary to provide top of the line care for patients. Each chapter features definitive and distinct areas of this multi-specialty discipline, and is designed to guide the reader from problem to solution in situations that can arise in the clinical setting.
Essentials of Neurosurgical Anesthesia & Critical Care, 2nd edition is a problem-oriented approach textbook that will aid a wide variety of readers in handling day-to-day issues and developments relevant to the perioperative care of neurosurgical patients.
Data is often plagued by noise which encumbers machine learning of clinically useful biomarkers and electroencephalogram (EEG) data is no exemption. Intracranial EEG (iEEG) data enhances the training of deep learning models of the human brain, yet is often prohibitive due to the invasive recording process. A more convenient alternative is to record brain activity using scalp electrodes. However, the inherent noise associated with scalp EEG data often impedes the learning process of neural models, achieving substandard performance. Here, an ensemble deep learning architecture for nonlinearly mapping scalp to iEEG data is proposed. The proposed architecture exploits the information from a limited number of joint scalp-intracranial recording to establish a novel methodology for detecting the epileptic discharges from the sEEG of a general population of subjects. Statistical tests and qualitative analysis have revealed that the generated pseudo-intracranial data are highly correlated with the true intracranial data. This facilitated the detection of IEDs from the scalp recordings where such waveforms are not often visible. As a real-world clinical application, these pseudo-iEEGs are then used by a convolutional neural network for the automated classification of intracranial epileptic discharges (IEDs) and non-IED of trials in the context of epilepsy analysis. Although the aim of this work was to circumvent the unavailability of iEEG and the limitations of sEEG, we have achieved a classification accuracy of 68% an increase of 6% over the previously proposed linear regression mapping.
Background:
The discrepancy between predicted effect-site concentration and measured bispectral index is problematic during intravenous anesthesia with target-controlled infusion of propofol and remifentanil. We hypothesized that bispectral index during total intravenous anesthesia would be more accurately predicted by a deep learning approach.
Methods:
Long short-term memory and the feed-forward neural network were sequenced to simulate the pharmacokinetic and pharmacodynamic parts of an empirical model, respectively, to predict intraoperative bispectral index during combined use of propofol and remifentanil. Inputs of long short-term memory were infusion histories of propofol and remifentanil, which were retrieved from target-controlled infusion pumps for 1,800 s at 10-s intervals. Inputs of the feed-forward network were the outputs of long short-term memory and demographic data such as age, sex, weight, and height. The final output of the feed-forward network was the bispectral index. The performance of bispectral index prediction was compared between the deep learning model and previously reported response surface model.
Results:
The model hyperparameters comprised 8 memory cells in the long short-term memory layer and 16 nodes in the hidden layer of the feed-forward network. The model training and testing were performed with separate data sets of 131 and 100 cases. The concordance correlation coefficient (95% CI) were 0.561 (0.560 to 0.562) in the deep learning model, which was significantly larger than that in the response surface model (0.265 [0.263 to 0.266], P < 0.001).
Conclusions:
The deep learning model-predicted bispectral index during target-controlled infusion of propofol and remifentanil more accurately compared to the traditional model. The deep learning approach in anesthetic pharmacology seems promising because of its excellent performance and extensibility.
Deeper neural networks are more difficult to train. We present a residual
learning framework to ease the training of networks that are substantially
deeper than those used previously. We explicitly reformulate the layers as
learning residual functions with reference to the layer inputs, instead of
learning unreferenced functions. We provide comprehensive empirical evidence
showing that these residual networks are easier to optimize, and can gain
accuracy from considerably increased depth. On the ImageNet dataset we evaluate
residual nets with a depth of up to 152 layers---8x deeper than VGG nets but
still having lower complexity. An ensemble of these residual nets achieves
3.57% error on the ImageNet test set. This result won the 1st place on the
ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100
and 1000 layers.
The depth of representations is of central importance for many visual
recognition tasks. Solely due to our extremely deep representations, we obtain
a 28% relative improvement on the COCO object detection dataset. Deep residual
nets are foundations of our submissions to ILSVRC & COCO 2015 competitions,
where we also won the 1st places on the tasks of ImageNet detection, ImageNet
localization, COCO detection, and COCO segmentation.
Training Deep Neural Networks is complicated by the fact that the
distribution of each layer's inputs changes during training, as the parameters
of the previous layers change. This slows down the training by requiring lower
learning rates and careful parameter initialization, and makes it notoriously
hard to train models with saturating nonlinearities. We refer to this
phenomenon as internal covariate shift, and address the problem by normalizing
layer inputs. Our method draws its strength from making normalization a part of
the model architecture and performing the normalization for each training
mini-batch}. Batch Normalization allows us to use much higher learning rates
and be less careful about initialization. It also acts as a regularizer, in
some cases eliminating the need for Dropout. Applied to a state-of-the-art
image classification model, Batch Normalization achieves the same accuracy with
14 times fewer training steps, and beats the original model by a significant
margin. Using an ensemble of batch-normalized networks, we improve upon the
best published result on ImageNet classification: reaching 4.9% top-5
validation error (and 4.8% test error), exceeding the accuracy of human raters.
We introduce Adam, an algorithm for first-order gradient-based optimization
of stochastic objective functions. The method is straightforward to implement
and is based an adaptive estimates of lower-order moments of the gradients. The
method is computationally efficient, has little memory requirements and is well
suited for problems that are large in terms of data and/or parameters. The
method is also ap- propriate for non-stationary objectives and problems with
very noisy and/or sparse gradients. The method exhibits invariance to diagonal
rescaling of the gradients by adapting to the geometry of the objective
function. The hyper-parameters have intuitive interpretations and typically
require little tuning. Some connections to related algorithms, on which Adam
was inspired, are discussed. We also analyze the theoretical convergence
properties of the algorithm and provide a regret bound on the convergence rate
that is comparable to the best known results under the online convex
optimization framework. We demonstrate that Adam works well in practice when
experimentally compared to other stochastic optimization methods.
Monitoring depth of anesthesia using the Electroencephalogram (EEG) is a major ongoing challenge for anesthetists. The EEG is the recording of brain electrical activity and it contains valuable information related to the different physiological states of the brain. This paper proposes a novel automated method for assessing the anesthesia depth level which consists of two steps. Initially, the sample entropy and permutation entropy features are extracted from the EEG signal. As EEG derived parameters represent different aspects of EEG features, it would be reasonable to use multiple parameters to assess the effect of anesthetic. These features quantify the amount of complexity or irregularity in EEG data and are conceptually simple, computationally efficient and artifact resistant. Then, the extracted features are used as input to an artificial neural network that is a data processing system based on the structure of a biological nervous system. The experimental results indicate that an overall accuracy of 88% can be obtained during sevoflurane in 17 patients to classify the EEG into awake, light, general and deep anesthetised states. Also this method yields a classification accuracy of 92.4% to distinguish between awake and general anesthesia in an independent database of propofol and desflurane in 129 patients. Considering the high accuracy of this method, a new EEG monitoring system could be developed to assist the anesthesiologist to estimate depth of anesthesia quickly and accurately.
This book provides the first comprehensive treatment of feed-forward neural networks from the perspective of statistical pattern recognition. After introducing the basic concepts of pattern recognition, the book describes techniques for modelling probability density functions, and discusses the properties and relative merits of the multi-layer perceptron and radial basis function network models. It also motivates the use of various forms of error functions, and reviews the principal algorithms for error function minimization. As well as providing a detailed discussion of learning and generalization in neural networks, the book also covers the important topics of data processing, feature extraction, and prior knowledge. The book concludes with an extensive treatment of Bayesian techniques and their applications to neural networks.
Localization of seizure sources prior to neurosurgery is crucial. In this paper, a new method is proposed to localize the seizure sources from multi-channel electroencephalogram (EEG) signals. Blind source separation based on second order blind identification (SOBI) is primarily applied to estimate the brain source signals in each window of the EEG signals. A new clustering method based on rival penalized competitive learning (RPCL) is then developed to cluster the rows of the estimated unmixing matrices in all the windows. The algorithm also includes pre and post-processing stages. By multiplying each cluster center to the EEG signals, the brain signal sources are approximated. According to a complexity value measure, the main seizure source signal is separated from the others. This signal is projected back to the electrodes’ space and is subjected to the dipole source localization using a single dipole model. The simulation results verify the accuracy of the system. In addition, correct localization of the seizure source is consistent with the clinical tests derived using the simultaneous intracranial recordings.