EEG-based outcome prediction after cardiac arrest with convolutional neural networks: Performance and visualization of discriminative features

Computer-assisted analysis of routine EEG to identify hidden biomarkers of epilepsy: A systematic review

Background

Computational analysis of routine electroencephalogram (rEEG) could improve the accuracy of epilepsy diagnosis. We aim to systematically assess the diagnostic performances of computed biomarkers for epilepsy in individuals undergoing rEEG.

Methods

We searched MEDLINE, EMBASE, EBM reviews, IEEE Explore and the grey literature for studies published between January 1961 and December 2022. We included studies reporting a computational method to diagnose epilepsy based on rEEG without relying on the identification of interictal epileptiform discharges or seizures. Diagnosis of epilepsy as per a treating physician was the reference standard. We assessed the risk of bias using an adapted QUADAS-2 tool.

Results

We screened 10 166 studies, and 37 were included. The sample size ranged from 8 to 192 (mean=54). The computed biomarkers were based on linear (43%), non-linear (27%), connectivity (38%), and convolutional neural networks (10%) models. The risk of bias was high or unclear in all studies, more commonly from spectrum effect and data leakage. Diagnostic accuracy ranged between 64% and 100%. We observed high methodological heterogeneity, preventing pooling of accuracy measures.

Conclusion

The current literature provides insufficient evidence to reliably assess the diagnostic yield of computational analysis of rEEG.

Significance

We provide guidelines regarding patient selection, reference standard, algorithms, and performance validation.

Decoding micro-electrocorticographic signals by using explainable 3D convolutional neural network to predict finger movements

BACKGROUND

Electroencephalography (EEG) and electrocorticography (ECoG) recordings have been used to decode finger movements by analyzing brain activity. Traditional methods focused on single bandpass power changes for movement decoding, utilizing machine learning models requiring manual feature extraction.

NEW METHOD

This study introduces a 3D convolutional neural network (3D-CNN) model to decode finger movements using ECoG data. The model employs adaptive, explainable AI (xAI) techniques to interpret the physiological relevance of brain signals. ECoG signals from epilepsy patients during awake craniotomy were processed to extract power spectral density across multiple frequency bands. These data formed a 3D matrix used to train the 3D-CNN to predict finger trajectories.

RESULTS

The 3D-CNN model showed significant accuracy in predicting finger movements, with root-mean-square error (RMSE) values of 0.26-0.38 for single finger movements and 0.20-0.24 for combined movements. Explainable AI techniques, Grad-CAM and SHAP, identified the high gamma (HG) band as crucial for movement prediction, showing specific cortical regions involved in different finger movements. These findings highlighted the physiological significance of the HG band in motor control.

COMPARISON WITH EXISTING METHODS

The 3D-CNN model outperformed traditional machine learning approaches by effectively capturing spatial and temporal patterns in ECoG data. The use of xAI techniques provided clearer insights into the model's decision-making process, unlike the "black box" nature of standard deep learning models.

CONCLUSIONS

The proposed 3D-CNN model, combined with xAI methods, enhances the decoding accuracy of finger movements from ECoG data. This approach offers a more efficient and interpretable solution for brain-computer interface (BCI) applications, emphasizing the HG band's role in motor control.

Artificial Intelligence and Machine Learning Applications in Critically Ill Brain Injured Patients

The utilization of Artificial Intelligence (AI) and Machine Learning (ML) is paving the way for significant strides in patient diagnosis, treatment, and prognostication in neurocritical care. These technologies offer the potential to unravel complex patterns within vast datasets ranging from vast clinical data and EEG (electroencephalogram) readings to advanced cerebral imaging facilitating a more nuanced understanding of patient conditions. Despite their promise, the implementation of AI and ML faces substantial hurdles. Historical biases within training data, the challenge of interpreting multifaceted data streams, and the "black box" nature of ML algorithms present barriers to widespread clinical adoption. Moreover, ethical considerations around data privacy and the need for transparent, explainable models remain paramount to ensure trust and efficacy in clinical decision-making.This article reflects on the emergence of AI and ML as integral tools in neurocritical care, discussing their roles from the perspective of both their scientific promise and the associated challenges. We underscore the importance of extensive validation in diverse clinical settings to ensure the generalizability of ML models, particularly considering their potential to inform critical medical decisions such as withdrawal of life-sustaining therapies. Advancement in computational capabilities is essential for implementing ML in clinical settings, allowing for real-time analysis and decision support at the point of care. As AI and ML are poised to become commonplace in clinical practice, it is incumbent upon health care professionals to understand and oversee these technologies, ensuring they adhere to the highest safety standards and contribute to the realization of personalized medicine. This engagement will be pivotal in integrating AI and ML into patient care, optimizing outcomes in neurocritical care through informed and data-driven decision-making.

Electroencephalogram-based machine learning models to predict neurologic outcome after cardiac arrest: A systematic review

AIM OF THE REVIEW

The primary aim of this systematic review was to investigate the most common electroencephalogram (EEG)-based machine learning (ML) model with the highest Area Under Receiver Operating Characteristic Curve (AUC) in two ML categories, conventional ML and Deep Neural Network (DNN), to predict the neurologic outcomes after cardiac arrest; the secondary aim was to investigate common EEG features applied to ML models.

METHODS

Systematic search of medical literature from PubMed and engineering literature from Compendex up to June 2, 2023. One reviewer screened studies that used EEG-based ML models to predict the neurologic outcomes after cardiac arrest. Four reviewers validated that the studies met selection criteria. Nine variables were manually extracted. The top-five common EEG features were calculated. We evaluated each study's risk of bias using the Quality in Prognosis Studies guideline.

RESULTS

Out of 351 identified studies, 17 studies met the inclusion criteria. Random Forest (RF) (n = 7) was the most common ML model in the conventional ML category (n = 11), followed by Convolutional Neural Network (CNN) (n = 4) in the DNN category (n = 6). The AUCs for RF ranged between 0.8 and 0.97, while CNN had AUCs between 0.7 and 0.92. The top-three commonly used EEG features were band power (n = 12), Shannon's Entropy (n = 11), burst-suppression ratio (n = 9).

CONCLUSIONS

RF and CNN were the two most common ML models with the highest AUCs for predicting the neurologic outcomes after cardiac arrest. Using a multimodal model that combines EEG features and electronic health record data may further improve prognostic performance.

An explainable deep-learning model to stage sleep states in children and propose novel EEG-related patterns in sleep apnea

Automatic deep-learning models used for sleep scoring in children with obstructive sleep apnea (OSA) are perceived as black boxes, limiting their implementation in clinical settings. Accordingly, we aimed to develop an accurate and interpretable deep-learning model for sleep staging in children using single-channel electroencephalogram (EEG) recordings. We used EEG signals from the Childhood Adenotonsillectomy Trial (CHAT) dataset (n = 1637) and a clinical sleep database (n = 980). Three distinct deep-learning architectures were explored to automatically classify sleep stages from a single-channel EEG data. Gradient-weighted Class Activation Mapping (Grad-CAM), an explainable artificial intelligence (XAI) algorithm, was then applied to provide an interpretation of the singular EEG patterns contributing to each predicted sleep stage. Among the tested architectures, a standard convolutional neural network (CNN) demonstrated the highest performance for automated sleep stage detection in the CHAT test set (accuracy = 86.9% and five-class kappa = 0.827). Furthermore, the CNN-based estimation of total sleep time exhibited strong agreement in the clinical dataset (intra-class correlation coefficient = 0.772). Our XAI approach using Grad-CAM effectively highlighted the EEG features associated with each sleep stage, emphasizing their influence on the CNN's decision-making process in both datasets. Grad-CAM heatmaps also allowed to identify and analyze epochs within a recording with a highly likelihood to be misclassified, revealing mixed features from different sleep stages within these epochs. Finally, Grad-CAM heatmaps unveiled novel features contributing to sleep scoring using a single EEG channel. Consequently, integrating an explainable CNN-based deep-learning model in the clinical environment could enable automatic sleep staging in pediatric sleep apnea tests.

Sex-related patterns in the electroencephalogram and their relevance in machine learning classifiers

Deep learning is increasingly being proposed for detecting neurological and psychiatric diseases from electroencephalogram (EEG) data but the method is prone to inadvertently incorporate biases from training data and exploit illegitimate patterns. The recent demonstration that deep learning can detect the sex from EEG implies potential sex-related biases in deep learning-based disease detectors for the many diseases with unequal prevalence between males and females. In this work, we present the male- and female-typical patterns used by a convolutional neural network that detects the sex from clinical EEG (81% accuracy in a separate test set with 142 patients). We considered neural sources, anatomical differences, and non-neural artifacts as sources of differences in the EEG curves. Using EEGs from 1140 patients, we found electrocardiac artifacts to be leaking into the supposedly brain activity-based classifiers. Nevertheless, the sex remained detectable after rejecting heart-related and other artifacts. In the cleaned data, EEG topographies were critical to detect the sex, but waveforms and frequencies were not. None of the traditional frequency bands was particularly important for sex detection. We were able to determine the sex even from EEGs with shuffled time points and therewith completely destroyed waveforms. Researchers should consider neural and non-neural sources as potential origins of sex differences in their data, they should maintain best practices of artifact rejection, even when datasets are large, and they should test their classifiers for sex biases.

AI and machine learning in resuscitation: Ongoing research, new concepts, and key challenges

Aim

Artificial intelligence (AI) and machine learning (ML) are important areas of computer science that have recently attracted attention for their application to medicine. However, as techniques continue to advance and become more complex, it is increasingly challenging for clinicians to stay abreast of the latest research. This overview aims to translate research concepts and potential concerns to healthcare professionals interested in applying AI and ML to resuscitation research but who are not experts in the field.

Main text

We present various research including prediction models using structured and unstructured data, exploring treatment heterogeneity, reinforcement learning, language processing, and large-scale language models. These studies potentially offer valuable insights for optimizing treatment strategies and clinical workflows. However, implementing AI and ML in clinical settings presents its own set of challenges. The availability of high-quality and reliable data is crucial for developing accurate ML models. A rigorous validation process and the integration of ML into clinical practice is essential for practical implementation. We furthermore highlight the potential risks associated with self-fulfilling prophecies and feedback loops, emphasizing the importance of transparency, interpretability, and trustworthiness in AI and ML models. These issues need to be addressed in order to establish reliable and trustworthy AI and ML models.

Conclusion

In this article, we overview concepts and examples of AI and ML research in the resuscitation field. Moving forward, appropriate understanding of ML and collaboration with relevant experts will be essential for researchers and clinicians to overcome the challenges and harness the full potential of AI and ML in resuscitation.

Deep learning for EEG-based prognostication after cardiac arrest: from current research to future clinical applications

Outcome prognostication in comatose patients after cardiac arrest (CA) remains to date a challenge. The major determinant of clinical outcome is the post-hypoxic/ischemic encephalopathy. Electroencephalography (EEG) is routinely used to assess neural functions in comatose patients. Currently, EEG-based outcome prognosis relies on visual evaluation by medical experts, which is time consuming, prone to subjectivity, and oblivious to complex patterns. The field of deep learning has given rise to powerful algorithms for detecting patterns in large amounts of data. Analyzing EEG signals of coma patients with deep neural networks with the goal of assisting in outcome prognosis is therefore a natural application of these algorithms. Here, we provide the first narrative literature review on the use of deep learning for prognostication after CA. Existing studies show overall high performance in predicting outcome, relying either on spontaneous or on auditory evoked EEG signals. Moreover, the literature is concerned with algorithmic interpretability, and has shown that largely, deep neural networks base their decisions on clinically or neurophysiologically meaningful features. We conclude this review by discussing considerations that the fields of artificial intelligence and neurology will need to jointly address in the future, in order for deep learning algorithms to break the publication barrier, and to be integrated in clinical practice.

Time to Awakening and Self-Fulfilling Prophecies After Cardiac Arrest

OBJECTIVES

Withdrawal of life-sustaining therapies for perceived poor neurologic prognosis (WLST-N) is common after resuscitation from cardiac arrest and may bias outcome estimates from models trained using observational data. We compared several approaches to outcome prediction with the goal of identifying strategies to quantify and reduce this bias.

DESIGN

Retrospective observational cohort study.

SETTING

Two academic medical centers ("UPMC" and "University of Alabama Birmingham" [UAB]).

PATIENTS

Comatose adults resuscitated from cardiac arrest.

INTERVENTION

None.

MEASUREMENTS AND MAIN RESULTS

As potential predictors, we considered clinical, laboratory, imaging, and quantitative electroencephalography data available early after hospital arrival. We followed patients until death, discharge, or awakening from coma. We used penalized Cox regression with a least absolute shrinkage and selection operator penalty and five-fold cross-validation to predict time to awakening in UPMC patients and then externally validated the model in UAB patients. This model censored patients after WLST-N, considering subsequent potential for awakening to be unknown. Next, we developed a penalized logistic model predicting awakening, which treated failure to awaken after WLST-N as a true observed outcome, and a separate logistic model predicting WLST-N. We scaled and centered individual patients' Cox and logistic predictions for awakening to allow direct comparison and then explored the difference in predictions across probabilities of WLST-N. Overall, 1,254 patients were included, and 29% awakened. Cox models performed well (mean area under the curve was 0.93 in the UPMC test sets and 0.83 in external validation). Logistic predictions of awakening were systematically more pessimistic than Cox-based predictions for patients at higher risk of WLST-N, suggesting potential for self-fulfilling prophecies to arise when failure to awaken after WLST-N is considered as the ground truth outcome.

CONCLUSIONS

Compared with traditional binary outcome prediction, censoring outcomes after WLST-N may reduce potential for bias and self-fulfilling prophecies.

Artificial Intelligence in Resuscitation: A Scoping Review

Introduction: Cardiac arrest is a significant cause of premature mortality and severe disability. Despite the death rate steadily decreasing over the previous decade, only 22% of survivors achieve good clinical status and only 25% of patients survive until their discharge from the hospital. The objective of this scoping review was to review relevant AI modalities and the main potential applications of AI in resuscitation. Methods: We conducted the literature search for related studies in PubMed, EMBASE, and Google Scholar. We included peer-reviewed publications and articles in the press, pooling and characterizing the data by their model types, goals, and benefits. Results: After identifying 268 original studies, we chose 59 original studies (reporting 1,817,419 patients) to include in the qualitative synthesis. AI-based methods appear to be superior to traditional methods in achieving high-level performance. Conclusion: AI might be useful in predicting cardiac arrest, heart rhythm disorders, and post-cardiac arrest outcomes, as well as in the delivery of drone-delivered defibrillators and notification of dispatchers. AI-powered technologies could be valuable assistants to continuously track patient conditions. Healthcare professionals should assist in the research and development of AI-powered technologies as well as their implementation into clinical practice.

Auditory stimulation and deep learning predict awakening from coma after cardiac arrest

Assessing the integrity of neural functions in coma after cardiac arrest remains an open challenge. Prognostication of coma outcome relies mainly on visual expert scoring of physiological signals, which is prone to subjectivity and leaves a considerable number of patients in a 'grey zone', with uncertain prognosis. Quantitative analysis of EEG responses to auditory stimuli can provide a window into neural functions in coma and information about patients' chances of awakening. However, responses to standardized auditory stimulation are far from being used in a clinical routine due to heterogeneous and cumbersome protocols. Here, we hypothesize that convolutional neural networks can assist in extracting interpretable patterns of EEG responses to auditory stimuli during the first day of coma that are predictive of patients' chances of awakening and survival at 3 months. We used convolutional neural networks (CNNs) to model single-trial EEG responses to auditory stimuli in the first day of coma, under standardized sedation and targeted temperature management, in a multicentre and multiprotocol patient cohort and predict outcome at 3 months. The use of CNNs resulted in a positive predictive power for predicting awakening of 0.83 ± 0.04 and 0.81 ± 0.06 and an area under the curve in predicting outcome of 0.69 ± 0.05 and 0.70 ± 0.05, for patients undergoing therapeutic hypothermia and normothermia, respectively. These results also persisted in a subset of patients that were in a clinical 'grey zone'. The network's confidence in predicting outcome was based on interpretable features: it strongly correlated to the neural synchrony and complexity of EEG responses and was modulated by independent clinical evaluations, such as the EEG reactivity, background burst-suppression or motor responses. Our results highlight the strong potential of interpretable deep learning algorithms in combination with auditory stimulation to improve prognostication of coma outcome.

Unraveling Spatial-Spectral Dynamics of Speech Categorization Speed Using Convolutional Neural Networks

The process of categorizing sounds into distinct phonetic categories is known as categorical perception (CP). Response times (RTs) provide a measure of perceptual difficulty during labeling decisions (i.e., categorization). The RT is quasi-stochastic in nature due to individuality and variations in perceptual tasks. To identify the source of RT variation in CP, we have built models to decode the brain regions and frequency bands driving fast, medium and slow response decision speeds. In particular, we implemented a parameter optimized convolutional neural network (CNN) to classify listeners' behavioral RTs from their neural EEG data. We adopted visual interpretation of model response using Guided-GradCAM to identify spatial-spectral correlates of RT. Our framework includes (but is not limited to): (i) a data augmentation technique designed to reduce noise and control the overall variance of EEG dataset; (ii) bandpower topomaps to learn the spatial-spectral representation using CNN; (iii) large-scale Bayesian hyper-parameter optimization to find best performing CNN model; (iv) ANOVA and posthoc analysis on Guided-GradCAM activation values to measure the effect of neural regions and frequency bands on behavioral responses. Using this framework, we observe that α-β (10-20 Hz) activity over left frontal, right prefrontal/frontal, and right cerebellar regions are correlated with RT variation. Our results indicate that attention, template matching, temporal prediction of acoustics, motor control, and decision uncertainty are the most probable factors in RT variation.

Deep learning in neuroimaging data analysis: Applications, challenges, and solutions

Methods for the analysis of neuroimaging data have advanced significantly since the beginning of neuroscience as a scientific discipline. Today, sophisticated statistical procedures allow us to examine complex multivariate patterns, however most of them are still constrained by assuming inherent linearity of neural processes. Here, we discuss a group of machine learning methods, called deep learning, which have drawn much attention in and outside the field of neuroscience in recent years and hold the potential to surpass the mentioned limitations. Firstly, we describe and explain the essential concepts in deep learning: the structure and the computational operations that allow deep models to learn. After that, we move to the most common applications of deep learning in neuroimaging data analysis: prediction of outcome, interpretation of internal representations, generation of synthetic data and segmentation. In the next section we present issues that deep learning poses, which concerns multidimensionality and multimodality of data, overfitting and computational cost, and propose possible solutions. Lastly, we discuss the current reach of DL usage in all the common applications in neuroimaging data analysis, where we consider the promise of multimodality, capability of processing raw data, and advanced visualization strategies. We identify research gaps, such as focusing on a limited number of criterion variables and the lack of a well-defined strategy for choosing architecture and hyperparameters. Furthermore, we talk about the possibility of conducting research with constructs that have been ignored so far or/and moving toward frameworks, such as RDoC, the potential of transfer learning and generation of synthetic data.

Diagnostic and prognostic EEG analysis of critically ill patients: A deep learning study

Visual interpretation of electroencephalography (EEG) is time consuming, may lack objectivity, and is restricted to features detectable by a human. Computer-based approaches, especially deep learning, could potentially overcome these limitations. However, most deep learning studies focus on a specific question or a single pathology. Here we explore the potential of deep learning for EEG-based diagnostic and prognostic assessment of patients with acute consciousness impairment (ACI) of various etiologies. EEGs from 358 adults from a randomized controlled trial (CERTA, NCT03129438) were retrospectively analyzed. A convolutional neural network was used to predict the clinical outcome (based either on survival or on best cerebral performance category) and to determine the etiology (four diagnostic categories). The largest probability output served as marker for the confidence of the network in its prediction ("certainty factor"); we also systematically compared the predictions with raw EEG data, and used a visualization algorithm (Grad-CAM) to highlight discriminative patterns. When all patients were considered, the area under the receiver operating characteristic curve (AUC) was 0.721 for predicting survival and 0.703 for predicting the outcome based on best CPC; for patients with certainty factor ≥ 60 % the AUCs increased to 0.776 and 0.755 respectively; and for certainty factor ≥ 75 % to 0.852 and 0.879. The accuracy for predicting the etiology was 54.5 %; the accuracy increased to 67.7 %, 70.3 % and 84.1 % for patients with certainty factor of 50 %, 60 % and 75 % respectively. Visual analysis showed that the network learnt EEG patterns typically recognized by human experts, and suggested new criteria. This work demonstrates for the first time the potential of deep learning-based EEG analysis in critically ill patients with various etiologies of ACI. Certainty factor and post-hoc correlation of input data with prediction help to better characterize the method and pave the route for future implementations in clinical routine.

Objective speech intelligibility prediction using a deep learning model with continuous speech-evoked cortical auditory responses

Auditory prostheses provide an opportunity for rehabilitation of hearing-impaired patients. Speech intelligibility can be used to estimate the extent to which the auditory prosthesis improves the user’s speech comprehension. Although behavior-based speech intelligibility is the gold standard, precise evaluation is limited due to its subjectiveness. Here, we used a convolutional neural network to predict speech intelligibility from electroencephalography (EEG). Sixty-four–channel EEGs were recorded from 87 adult participants with normal hearing. Sentences spectrally degraded by a 2-, 3-, 4-, 5-, and 8-channel vocoder were used to set relatively low speech intelligibility conditions. A Korean sentence recognition test was used. The speech intelligibility scores were divided into 41 discrete levels ranging from 0 to 100%, with a step of 2.5%. Three scores, namely 30.0, 37.5, and 40.0%, were not collected. The speech features, i.e., the speech temporal envelope (ENV) and phoneme (PH) onset, were used to extract continuous-speech EEGs for speech intelligibility prediction. The deep learning model was trained by a dataset of event-related potentials (ERP), correlation coefficients between the ERPs and ENVs, between the ERPs and PH onset, or between ERPs and the product of the multiplication of PH and ENV (PHENV). The speech intelligibility prediction accuracies were 97.33% (ERP), 99.42% (ENV), 99.55% (PH), and 99.91% (PHENV). The models were interpreted using the occlusion sensitivity approach. While the ENV models’ informative electrodes were located in the occipital area, the informative electrodes of the phoneme models, i.e., PH and PHENV, were based on the occlusion sensitivity map located in the language processing area. Of the models tested, the PHENV model obtained the best speech intelligibility prediction accuracy. This model may promote clinical prediction of speech intelligibility with a comfort speech intelligibility test.

Predicting the Risk of Dental Implant Loss Using Deep Learning

AIM

To investigate the feasibility of predicting dental implant loss risk with deep learning (DL) based on preoperative cone-beam computed tomography.

MATERIALS AND METHODS

Six hundred and three patients who underwent implant surgery (279 high-risk patients who did and 324 low-risk patients who did not experience implant loss within 5 years) from January 2012 to January 2020 were enrolled. Three models, a logistic regression clinical model (CM) based on clinical features, a DL model based on radiography features, and an integrated model (IM) developed by combining CM with DL, were developed to predict the 5-year implant loss risk. The area under the receiver operating characteristic curve (AUC) was used to evaluate the model performance. Time to implant loss was considered for both groups, and Kaplan-Meier curves were created and compared by the log-rank test.

RESULTS

The IM exhibited the best performance in predicting implant loss risk [AUC = 0.90, 95% confidence interval (CI) 0.84-0.95], followed by the DL model (AUC = 0.87, 95% CI 0.80-0.92) and the CM (AUC = 0.72, 95% CI 0.63-0.79).

CONCLUSION

Our study offers preliminary evidence that both the DL model and IM performed well in predicting implant fate within 5 years and thus may greatly facilitate implant practitioners in assessing preoperative risks.

Revisiting EEG as part of the multidisciplinary approach to post-cardiac arrest care and prognostication: A review

Since the 1960s, EEG has been used to assess the neurologic function of patients in the hours and days after cardiac arrest. Accurate and reliable prognostication after cardiac arrest is vital for tailoring aggressive patient care for those with a high likelihood of recovery and setting appropriate goals of care for those who have a high likelihood of a poor outcome. Attempts to define EEG's role in this process has evolved over the years. In this review, we provide important historical context about EEG's use, it's perceived unreliability in the post-cardiac arrest patient in the past and provide a detailed analysis of how this role has changed recently. A review of the 71 most recent and highest quality studies demonstrates that the introduction of a uniform classification and a timed approach to EEG analysis have positioned EEG as a complementary tool in the multimodal approach for prognostication. The review was created and intended for medical staff in the intensive care units and emphasizes EEG patterns and timing which portend both favorable and poor prognoses. Also, the review addresses the overall quality of the existing studies and discusses future directions to address the knowledge gaps in this field.

High-Throughput Recognition of Tumor Cells Using Label-Free Elemental Characteristics Based on Interpretable Deep Learning

With cancer seriously hampering the increasing life expectancy of people, developing an instant diagnostic method has become an urgent objective. In this work, we developed a label-free laser-induced breakdown spectroscopy (LIBS) method for high-throughput recognition of tumor cells. LIBS spectra were straightly collected from cells dropped on a silicon substrate and built into a deep learning model for simultaneous classification of various cancers. To interpret the result of the deep learning algorithm, gradient-weighted class activation mapping was utilized to a one-dimensional convolution neural network (1D-CNN), and the saliency maps thus obtained amplified the differences between the spectra of cell lines. Overall results showed that the 1D-CNN algorithms achieved a mean sensitivity of 94.00%, a mean specificity of 98.47%, and a mean accuracy of 97.56%. Thus, the proposed method performed satisfactorily and is seen as an interpretable classification process for cancer cell lines.

Convolutional neural networks for decoding electroencephalography responses and visualizing trial by trial changes in discriminant features

BACKGROUND

Deep learning has revolutionized the field of computer vision, where convolutional neural networks (CNNs) extract complex patterns of information from large datasets. The use of deep networks in neuroscience is mainly focused to neuroimaging or brain computer interface -BCI- applications. In electroencephalography (EEG) research, multivariate pattern analysis (MVPA) mainly relies on linear algorithms, which require a homogeneous dataset and assume that discriminant features appear at consistent latencies and electrodes across trials. However, neural responses may shift in time or space during an experiment, resulting in under-estimation of discriminant features. Here, we aimed at using CNNs to classify EEG responses to external stimuli, by taking advantage of time- and space- unlocked neural activity, and at examining how discriminant features change over the course of an experiment, on a trial by trial basis.

NEW METHOD

We present a novel pipeline, consisting of data augmentation, CNN training, and feature visualization techniques, fine-tuned for MVPA on EEG data.

RESULTS

Our pipeline provides high classification performance and generalizes to new datasets. Additionally, we show that the features identified by the CNN for classification are electrophysiologically interpretable and can be reconstructed at the single-trial level to study trial-by-trial evolution of class-specific discriminant activity.

COMPARISON WITH EXISTING TECHNIQUES

The developed pipeline was compared to commonly used MVPA algorithms like logistic regression and support vector machines, as well as to shallow and deep convolutional neural networks. Our approach yielded significantly higher classification performance than existing MVPA techniques (p = 0.006) and comparable results to other CNNs for EEG data.

CONCLUSION

In summary, we present a novel deep learning pipeline for MVPA of EEG data, that can extract trial-by-trial discriminative activity in a data-driven way.

Non-human primate epidural ECoG analysis using explainable deep learning technology

Objective.With the development in the field of neural networks,explainable AI(XAI), is being studied to ensure that artificial intelligence models can be explained. There are some attempts to apply neural networks to neuroscientific studies to explain neurophysiological information with high machine learning performances. However, most of those studies have simply visualized features extracted from XAI and seem to lack an active neuroscientific interpretation of those features. In this study, we have tried to actively explain the high-dimensional learning features contained in the neurophysiological information extracted from XAI, compared with the previously reported neuroscientific results.Approach. We designed a deep neural network classifier using 3D information (3D DNN) and a 3D class activation map (3D CAM) to visualize high-dimensional classification features. We used those tools to classify monkey electrocorticogram (ECoG) data obtained from the unimanual and bimanual movement experiment.Main results. The 3D DNN showed better classification accuracy than other machine learning techniques, such as 2D DNN. Unexpectedly, the activation weight in the 3D CAM analysis was high in the ipsilateral motor and somatosensory cortex regions, whereas the gamma-band power was activated in the contralateral areas during unimanual movement, which suggests that the brain signal acquired from the motor cortex contains information about both contralateral movement and ipsilateral movement. Moreover, the hand-movement classification system used critical temporal information at movement onset and offset when classifying bimanual movements.Significance.As far as we know, this is the first study to use high-dimensional neurophysiological information (spatial, spectral, and temporal) with the deep learning method, reconstruct those features, and explain how the neural network works. We expect that our methods can be widely applied and used in neuroscience and electrophysiology research from the point of view of the explainability of XAI as well as its performance.

Prediction of patient survival following postanoxic coma using EEG data and clinical features

Electroencephalography (EEG) is an effective and non-invasive technique commonly used to monitor brain activity and assist in outcome prediction for comatose patients post cardiac arrest. EEG data may demonstrate patterns associated with poor neurological outcome for patients with hypoxic injury. Thus, both quantitative EEG (qEEG) and clinical data contain prognostic information for patient outcome. In this study we use machine learning (ML) techniques, random forest (RF) and support vector machine (SVM) to classify patient outcome post cardiac arrest using qEEG and clinical feature sets, individually and combined. Our ML experiments show RF and SVM perform better using the joint feature set. In addition, we extend our work by implementing a convolutional neural network (CNN) based on time-frequency images derived from EEG to compare with our qEEG ML models. The results demonstrate significant performance improvement in outcome prediction using non-feature based CNN compared to our feature based ML models. Implementation of ML and DL methods in clinical practice have the potential to improve reliability of traditional qualitative assessments for postanoxic coma patients.

Novel approaches to prediction in severe brain injury

PURPOSE OF REVIEW

Recovery after severe brain injury is variable and challenging to accurately predict at the individual patient level. This review highlights new developments in clinical prognostication with a special focus on the prediction of consciousness and increasing reliance on methods from data science.

RECENT FINDINGS

Recent research has leveraged serum biomarkers, quantitative electroencephalography, MRI, and physiological time-series to build models for recovery prediction. The analysis of high-resolution data and the integration of features from different modalities can be approached with efficient computational techniques.

SUMMARY

Advances in neurophysiology and neuroimaging, in combination with computational methods, represent a novel paradigm for prediction of consciousness and functional recovery after severe brain injury. Research is needed to produce reliable, patient-level predictions that could meaningfully impact clinical decision making.

Targeted temperature management and early neuro-prognostication after cardiac arrest

Targeted temperature management (TTM) is a recommended neuroprotective intervention for coma after out-of-hospital cardiac arrest (OHCA). However, controversies exist concerning the proper implementation and overall efficacy of post-CA TTM, particularly related to optimal timing and depth of TTM and cooling methods. A review of the literature finds that optimizing and individualizing TTM remains an open question requiring further clinical investigation. This paper will summarize the preclinical and clinical trial data to-date, current recommendations, and future directions of this therapy, including new cooling methods under investigation. For now, early induction, maintenance for at least 24 hours, and slow rewarming utilizing endovascular methods may be preferred. Moreover, timely and accurate neuro-prognostication is valuable for guiding ethical and cost-effective management of post-CA coma. Current evidence for early neuro-prognostication after TTM suggests that a combination of initial prediction models, biomarkers, neuroimaging, and electrophysiological methods is the optimal strategy in predicting neurological functional outcomes.

Accurate age classification using manual method and deep convolutional neural network based on orthopantomogram images

Age estimation is an important challenge in many fields, including immigrant identification, legal requirements, and clinical treatments. Deep learning techniques have been applied for age estimation recently but lacking performance comparison between manual and machine learning methods based on a large sample of dental orthopantomograms (OPGs). In total, we collected 10,257 orthopantomograms for the study. We derived logistic regression linear models for each legal age threshold (14, 16, and 18 years old) for manual method and developed the end-to-end convolutional neural network (CNN) which classified the dental age directly to compare with the manual method. Both methods are based on left mandibular eight permanent teeth or the third molar separately. Our results show that compared with the manual methods (92.5%, 91.3%, and 91.8% for age thresholds of 14, 16, and 18, respectively), the end-to-end CNN models perform better (95.9%, 95.4%, and 92.3% for age thresholds of 14, 16, and 18, respectively). This work proves that CNN models can surpass humans in age classification, and the features extracted by machines may be different from that defined by human.

Grad-CAM helps interpret the deep learning models trained to classify multiple sclerosis types using clinical brain magnetic resonance imaging

BACKGROUND

Deep learning using convolutional neural networks (CNNs) has shown great promise in advancing neuroscience research. However, the ability to interpret the CNNs lags far behind, confounding their clinical translation.

NEW METHOD

We interrogated 3 heatmap-generating techniques that have increasing generalizability for CNN interpretation: class activation mapping (CAM), gradient (Grad)-CAM, and Grad-CAM++. To investigate the impact of CNNs on heatmap generation, we also examined 6 different models trained to classify brain magnetic resonance imaging into 3 types: relapsing-remitting multiple sclerosis (RRMS), secondary progressive MS (SPMS), and control. Further, we designed novel methods to visualize and quantify the heatmaps to improve interpretability.

RESULTS

Grad-CAM showed the best heatmap localizing ability, and CNNs with a global average pooling layer and pretrained weights had the best classification performance. Based on the best-performing CNN model, called VGG19, the 95th percentile values of Grad-CAM in SPMS were significantly higher than RRMS, indicating greater heterogeneity. Further, voxel-wise analysis of the thresholded Grad-CAM confirmed the difference identified visually between RRMS and SPMS in discriminative brain regions: occipital versus frontal and occipital, or temporal/parietal.

COMPARISON WITH EXISTING METHODS

No study has examined the CAM methods together using clinical images. There is also lack of study on the impact of CNN architecture on heatmap outcomes, and of technologies to quantify heatmap patterns in clinical settings.

CONCLUSIONS

Grad-CAM outperforms CAM and Grad-CAM++. Integrating Grad-CAM, novel heatmap quantification approaches, and robust CNN models may be an effective strategy in identifying the most crucial brain areas underlying disease development in MS.

Standardized visual EEG features predict outcome in patients with acute consciousness impairment of various etiologies

BACKGROUND

Early prognostication in patients with acute consciousness impairment is a challenging but essential task. Current prognostic guidelines vary with the underlying etiology. In particular, electroencephalography (EEG) is the most important paraclinical examination tool in patients with hypoxic ischemic encephalopathy (HIE), whereas it is not routinely used for outcome prediction in patients with traumatic brain injury (TBI).

METHOD

Data from 364 critically ill patients with acute consciousness impairment (GCS ≤ 11 or FOUR ≤ 12) of various etiologies and without recent signs of seizures from a prospective randomized trial were retrospectively analyzed. Random forest classifiers were trained using 8 visual EEG features-first alone, then in combination with clinical features-to predict survival at 6 months or favorable functional outcome (defined as cerebral performance category 1-2).

RESULTS

The area under the ROC curve was 0.812 for predicting survival and 0.790 for predicting favorable outcome using EEG features. Adding clinical features did not improve the overall performance of the classifier (for survival: AUC = 0.806, p = 0.926; for favorable outcome: AUC = 0.777, p = 0.844). Survival could be predicted in all etiology groups: the AUC was 0.958 for patients with HIE, 0.955 for patients with TBI and other neurosurgical diagnoses, 0.697 for patients with metabolic, inflammatory or infectious causes for consciousness impairment and 0.695 for patients with stroke. Training the classifier separately on subgroups of patients with a given etiology (and thus using less training data) leads to poorer classification performance.

CONCLUSIONS

While prognostication was best for patients with HIE and TBI, our study demonstrates that similar EEG criteria can be used in patients with various causes of consciousness impairment, and that the size of the training set is more important than homogeneity of ACI etiology.

Brain injury after cardiac arrest: from prognostication of comatose patients to rehabilitation

More patients are surviving cardiac arrest than ever before; however, the burden now lies with estimating neurological prognoses in a large number of patients who were initially comatose, in whom the ultimate outcome is unclear. Neurologists, neurointensivists, and clinical neurophysiologists must accurately balance the concern that overly conservative prognostication could leave patients in a severely disabled state, with the possibility that inaccurately pessimistic prognostication could lead to the withdrawal of life-sustaining treatment in patients who might otherwise have a good functional outcome. Prognostic tools have improved greatly, including electrophysiological tests, neuroimaging, and chemical biomarkers. Conclusions about the prognosis should be delayed at least 72 h after arrest to allow for the clearance of sedative drugs. Cognitive impairments, emotional problems, and fatigue are common among patients who have survived cardiac arrest, and often go unrecognised despite being related to caregiver burden and a decreased participation in society. Through simple screening, these problems can be identified, and patients can be provided with adequate information and rehabilitation.

Interpretable and lightweight convolutional neural network for EEG decoding: Application to movement execution and imagination

Convolutional neural networks (CNNs) are emerging as powerful tools for EEG decoding: these techniques, by automatically learning relevant features for class discrimination, improve EEG decoding performances without relying on handcrafted features. Nevertheless, the learned features are difficult to interpret and most of the existing CNNs introduce many trainable parameters. Here, we propose a lightweight and interpretable shallow CNN (Sinc-ShallowNet), by stacking a temporal sinc-convolutional layer (designed to learn band-pass filters, each having only the two cut-off frequencies as trainable parameters), a spatial depthwise convolutional layer (reducing channel connectivity and learning spatial filters tied to each band-pass filter), and a fully-connected layer finalizing the classification. This convolutional module limits the number of trainable parameters and allows direct interpretation of the learned spectral-spatial​ features via simple kernel visualizations. Furthermore, we designed a post-hoc gradient-based technique to enhance interpretation by identifying the more relevant and more class-specific features. Sinc-ShallowNet was evaluated on benchmark motor-execution and motor-imagery datasets and against different design choices and training strategies. Results show that (i) Sinc-ShallowNet outperformed a traditional machine learning algorithm and other CNNs for EEG decoding; (ii) The learned spectral-spatial features matched well-known EEG motor-related activity; (iii) The proposed architecture performed better with a larger number of temporal kernels still maintaining a good compromise between accuracy and parsimony, and with a trialwise rather than a cropped training strategy. In perspective, the proposed approach, with its interpretative capacity, can be exploited to investigate cognitive/motor aspects whose EEG correlates are yet scarcely known, potentially characterizing their relevant features.

Electrographic predictors of successful weaning from anaesthetics in refractory status epilepticus

Intravenous third-line anaesthetic agents are typically titrated in refractory status epilepticus to achieve either seizure suppression or burst suppression on continuous EEG. However, the optimum treatment paradigm is unknown and little data exist to guide the withdrawal of anaesthetics in refractory status epilepticus. Premature withdrawal of anaesthetics risks the recurrence of seizures, whereas the prolonged use of anaesthetics increases the risk of treatment-associated adverse effects. This study sought to measure the accuracy of features of EEG activity during anaesthetic weaning in refractory status epilepticus as predictors of successful weaning from intravenous anaesthetics. We prespecified a successful anaesthetic wean as the discontinuation of intravenous anaesthesia without developing recurrent status epilepticus, and a wean failure as either recurrent status epilepticus or the resumption of anaesthesia for the purpose of treating an EEG pattern concerning for incipient status epilepticus. We evaluated two types of features as predictors of successful weaning: spectral components of the EEG signal, and spatial-correlation-based measures of functional connectivity. The results of these analyses were used to train a classifier to predict wean outcome. Forty-seven consecutive anaesthetic weans (23 successes, 24 failures) were identified from a single-centre cohort of patients admitted with refractory status epilepticus from 2016 to 2019. Spectral components of the EEG revealed no significant differences between successful and unsuccessful weans. Analysis of functional connectivity measures revealed that successful anaesthetic weans were characterized by the emergence of larger, more densely connected, and more highly clustered spatial functional networks, yielding 75.5% (95% confidence interval: 73.1-77.8%) testing accuracy in a bootstrap analysis using a hold-out sample of 20% of data for testing and 74.6% (95% confidence interval 73.2-75.9%) testing accuracy in a secondary external validation cohort, with an area under the curve of 83.3%. Distinct signatures in the spatial networks of functional connectivity emerge during successful anaesthetic liberation in status epilepticus; these findings are absent in patients with anaesthetic wean failure. Identifying features that emerge during successful anaesthetic weaning may allow faster and more successful anaesthetic liberation after refractory status epilepticus.

Deep Learning in Physiological Signal Data: A Survey

Deep Learning (DL), a successful promising approach for discriminative and generative tasks, has recently proved its high potential in 2D medical imaging analysis; however, physiological data in the form of 1D signals have yet to be beneficially exploited from this novel approach to fulfil the desired medical tasks. Therefore, in this paper we survey the latest scientific research on deep learning in physiological signal data such as electromyogram (EMG), electrocardiogram (ECG), electroencephalogram (EEG), and electrooculogram (EOG). We found 147 papers published between January 2018 and October 2019 inclusive from various journals and publishers. The objective of this paper is to conduct a detailed study to comprehend, categorize, and compare the key parameters of the deep-learning approaches that have been used in physiological signal analysis for various medical applications. The key parameters of deep-learning approach that we review are the input data type, deep-learning task, deep-learning model, training architecture, and dataset sources. Those are the main key parameters that affect system performance. We taxonomize the research works using deep-learning method in physiological signal analysis based on: (1) physiological signal data perspective, such as data modality and medical application; and (2) deep-learning concept perspective such as training architecture and dataset sources.

EEG-based outcome prediction after cardiac arrest with convolutional neural networks: Performance and visualization of discriminative features

Prognostication for comatose patients after cardiac arrest is a difficult but essential task. Currently, visual interpretation of electroencephalogram (EEG) is one of the main modality used in outcome prediction. There is a growing interest in computer-assisted EEG interpretation, either to overcome the possible subjectivity of visual interpretation, or to identify complex features of the EEG signal. We used a one-dimensional convolutional neural network (CNN) to predict functional outcome based on 19-channel-EEG recorded from 267 adult comatose patients during targeted temperature management after CA. The area under the receiver operating characteristic curve (AUC) on the test set was 0.885. Interestingly, model architecture and fine-tuning only played a marginal role in classification performance. We then used gradient-weighted class activation mapping (Grad-CAM) as visualization technique to identify which EEG features were used by the network to classify an EEG epoch as favorable or unfavorable outcome, and also to understand failures of the network. Grad-CAM showed that the network relied on similar features than classical visual analysis for predicting unfavorable outcome (suppressed background, epileptiform transients). This study confirms that CNNs are promising models for EEG-based prognostication in comatose patients, and that Grad-CAM can provide explanation for the models' decision-making, which is of utmost importance for future use of deep learning models in a clinical setting.