Objective Classification of mTBI Using Machine Learning on a Combination of Frontopolar Electroencephalography Measurements and Self-reported Symptoms.
McNerney M Windy,Hobday Thomas,Cole Betsy,Ganong Rick,Winans Nina,Matthews Dennis,Hood Jim,Lane Stephen
Sports medicine - open
BACKGROUND:The reliable diagnosis of a mild traumatic brain injury (mTBI) is a pervasive problem in sports and in the military. The frequency and severity of each occurrence, while difficult to quantify, may impact long term cognitive function and quality of life. Despite the new revelations concerning brain disfunction from head injuries, individuals still feel pressure to remain on the field despite a debilitating injury. In this study, we evaluated the accuracy of a system that could be employed on the sidelines or in the locker room to provide an immediate objective mTBI assessment. METHODS:Participants consisted of 38 individuals with a recent mTBI and 47 controls with no history of mTBI within the last 5 years. Participants were administered a simple symptom questionnaire, behavioral tests, and resting state EEG was measured using three frontopolar electrodes. An advanced machine learning algorithm called boosting was utilized to classify subjects into either injured or controls using power spectral densities on 1-min of resting EEG and the symptom questionnaire. RESULTS:Results based on leave-one-out cross-validation revealed that the addition of EEG measurements boosted the accuracy to approximately 91 ± 2% compared to 82 ± 4% from the symptom questionnaire alone. CONCLUSION:This study demonstrated the potential benefit of including EEG measurements to diagnose suspected brain injury patients. This is a step toward accurate and objective classification measurements that can be implemented on the field as a future injury assessment tool.
Classifying major depression patients and healthy controls using EEG, eye tracking and galvanic skin response data.
Ding Xinfang,Yue Xinxin,Zheng Rui,Bi Cheng,Li Dai,Yao Guizhong
Journal of affective disorders
OBJECTIVE:Major depression disorder (MDD) is one of the most prevalent mental disorders worldwide. Diagnosing depression in the early stage is crucial to treatment process. However, due to depression's comorbid nature and the subjectivity in diagnosis, an early diagnosis could be challenging. Recently, machine learning approaches have been used to process Electroencephalography (EEG) and neuroimaging data to facilitate the diagnosis. In the present study, we used a multimodal machine learning approach involving EEG, eye tracking and galvanic skin response data as input to classify depression patients and healthy controls. METHODS:One hundred and forty-four MDD depression patients and 204 matched healthy controls were recruited. They were required to watch a series of affective and neutral stimuli while EEG, eye tracking information and galvanic skin response were recorded via a set of low-cost, portable devices. Three machine learning algorithms including Random Forests, Logistic Regression and Support Vector Machine (SVM) were trained to build dichotomous classification model. RESULTS:The results showed that the highest classification f1 score was obtained by Logistic Regression algorithms, with accuracy = 79.63%, precision = 76.67%, recall = 85.19% and f1 score = 80.70% LIMITATIONS: No hospitalized patients were available; only outpatients were included in the present study. The sample consisted mostly of young adult, and no elder patients were included. CONCLUSIONS:The machine learning approach can be a useful tool for classifying MDD patients and healthy controls and may help for diagnostic processes.
Prediction of antiepileptic drug treatment outcomes using machine learning.
Colic Sinisa,Wither Robert G,Lang Min,Zhang Liang,Eubanks James H,Bardakjian Berj L
Journal of neural engineering
OBJECTIVE:Antiepileptic drug (AED) treatments produce inconsistent outcomes, often necessitating patients to go through several drug trials until a successful treatment can be found. This study proposes the use of machine learning techniques to predict epilepsy treatment outcomes of commonly used AEDs. APPROACH:Machine learning algorithms were trained and evaluated using features obtained from intracranial electroencephalogram (iEEG) recordings of the epileptiform discharges observed in Mecp2-deficient mouse model of the Rett Syndrome. Previous work have linked the presence of cross-frequency coupling (I ) of the delta (2-5 Hz) rhythm with the fast ripple (400-600 Hz) rhythm in epileptiform discharges. Using the I to label post-treatment outcomes we compared support vector machines (SVMs) and random forest (RF) machine learning classifiers for providing likelihood scores of successful treatment outcomes. MAIN RESULTS:(a) There was heterogeneity in AED treatment outcomes, (b) machine learning techniques could be used to rank the efficacy of AEDs by estimating likelihood scores for successful treatment outcome, (c) I features yielded the most effective a priori identification of appropriate AED treatment, and (d) both classifiers performed comparably. SIGNIFICANCE:Machine learning approaches yielded predictions of successful drug treatment outcomes which in turn could reduce the burdens of drug trials and lead to substantial improvements in patient quality of life.
Machine-learning-based classification between post-traumatic stress disorder and major depressive disorder using P300 features.
Shim Miseon,Jin Min Jin,Im Chang-Hwan,Lee Seung-Hwan
BACKGROUND:The development of optimal classification criteria for specific mental disorders which share similar symptoms is an important issue for precise diagnosis. We investigated whether P300 features in both sensor-level and source-level could be effectively used to classify post-traumatic stress disorder (PTSD) and major depressive disorder (MDD). METHOD:EEG signals were recorded from fifty-one PTSD patients, 67 MDD patients, and 39 healthy controls (HCs) while performing an auditory oddball task. Amplitude and latency of P300 were evaluated, and the current source analysis of P300 components was conducted using sLORETA. Finally, we classified two groups using machine-learning methods with both sensor- and source-level features. Moreover, we checked the comorbidity effects using the same approaches (PTSD-mono diagnosis (PTSDm, n = 28) and PTSD-comorbid diagnosis (PTSDc, n = 23)). RESULTS:PTSD showed significantly reduced P300 amplitudes and prolonged latency compared to HCs and MDD. Moreover, PTSD showed significantly reduced source activities, and the source activities were significantly correlated with symptoms of depression and anxiety. Also, the best classification accuracy at each pair was as follows: 80.00% (PTSD-HCs), 67.92% (MDD-HCs), 70.34% (PTSD-MDD), 82.09% (PTSDm-HCs), 71.58% (PTSDm-MDD), 82.56% (PTSDc-HCs), and 76.67% (PTSDc- MDD). CONCLUSION:Since abnormal P300 reflects pathophysiological characteristics of PTSD, PTSD patients were well-discriminated from MDD and HCs when using P300 features. Thus, altered P300 characteristics in both sensor- and source-level may be useful biomarkers to diagnosis PTSD.
Classification of Alzheimer's Disease with Respect to Physiological Aging with Innovative EEG Biomarkers in a Machine Learning Implementation.
Vecchio Fabrizio,Miraglia Francesca,Alù Francesca,Menna Matteo,Judica Elda,Cotelli Maria,Rossini Paolo Maria
Journal of Alzheimer's disease : JAD
BACKGROUND:Several studies investigated clinical and instrumental differences to make diagnosis of dementia in general and in Alzheimer's disease (AD) in particular with the aim to classify, at the individual level, AD patients and healthy controls cooperating with neuropsychological tests for an early diagnosis. Advanced network analysis of electroencephalographic (EEG) rhythms provides information on dynamic brain connectivity and could be used in classification processes. If successfully reached, this goal would add a low-cost, easily accessible, and non-invasive technique with neuropsychological tests. OBJECTIVE:To investigate the possibility to automatically classify physiological versus pathological aging from cortical sources' connectivity based on a support vector machine (SVM) applied to EEG small-world parameter. METHODS:A total of 295 subjects were recruited: 120 healthy volunteers and 175 AD. Graph theory functions were applied to undirected and weighted networks obtained by lagged linear coherence evaluated by eLORETA. A machine-learning classifier (SVM) was applied. EEG frequency bands were: delta (2-4 Hz), theta (4-8 Hz), alpha1 (8-10.5 Hz), alpha2 (10.5-13 Hz), beta1 (13-20 Hz), beta2 (20-30 Hz), and gamma (30-40 Hz). RESULTS:The receiver operating characteristic curve showed AUC of 0.97±0.03 (indicating very high classification accuracy). The classifier showed 95% ±5% sensitivity, 96% ±3% specificity, and 95% ±3% accuracy for the classification. CONCLUSION:EEG connectivity analysis via a combination of source/connectivity biomarkers, highly correlating with neuropsychological AD diagnosis, could represent a promising tool in identification of AD patients. This approach represents a low-cost and non-invasive method, one that utilizes widely available techniques which, when combined, reach high sensitivity/specificity and optimal classification accuracy on an individual basis (0.97 of AUC).
EEG Source Network for the Diagnosis of Schizophrenia and the Identification of Subtypes Based on Symptom Severity-A Machine Learning Approach.
Kim Jeong-Youn,Lee Hyun Seo,Lee Seung-Hwan
Journal of clinical medicine
A precise diagnosis and a comprehensive assessment of symptom severity are important clinical issues in patients with schizophrenia (SZ). We investigated whether electroencephalography (EEG) features obtained from EEG source network analyses could be effectively applied to classify the SZ subtypes based on symptom severity. Sixty-four electrode EEG signals were recorded from 119 patients with SZ (53 males and 66 females) and 119 normal controls (NC, 51 males and 68 females) during resting-state with closed eyes. Brain network features (global and local clustering coefficient and global path length) were calculated from EEG source activities. According to positive, negative, and cognitive/disorganization symptoms, the SZ patients were divided into two groups (high and low) by positive and negative syndrome scale (PANSS). To select features for classification, we used the sequential forward selection (SFS) method. The classification accuracy was evaluated using 10 by 10-fold cross-validation with the linear discriminant analysis (LDA) classifier. The best classification accuracy was 80.66% for estimating SZ patients from the NC group. The best classification accuracy between low and high groups in positive, negative, and cognitive/disorganization symptoms were 88.10%, 75.25%, and 77.78%, respectively. The selected features well-represented the pathological brain regions of SZ. Our study suggested that resting-state EEG network features could successfully classify between SZ patients and the NC, and between low and high SZ groups in positive, negative, and cognitive/disorganization symptoms.
Comparison of machine learning models for seizure prediction in hospitalized patients.
Struck Aaron F,Rodriguez-Ruiz Andres A,Osman Gamaledin,Gilmore Emily J,Haider Hiba A,Dhakar Monica B,Schrettner Matthew,Lee Jong W,Gaspard Nicolas,Hirsch Lawrence J,Westover M Brandon,
Annals of clinical and translational neurology
OBJECTIVE:To compare machine learning methods for predicting inpatient seizures risk and determine the feasibility of 1-h screening EEG to identify low-risk patients (<5% seizures risk in 48 h). METHODS:The Critical Care EEG Monitoring Research Consortium (CCEMRC) multicenter database contains 7716 continuous EEGs (cEEG). Neural networks (NN), elastic net logistic regression (EN), and sparse linear integer model (RiskSLIM) were trained to predict seizures. RiskSLIM was used previously to generate 2HELPS2B model of seizure predictions. Data were divided into training (60% for model fitting) and evaluation (40% for model evaluation) cohorts. Performance was measured using area under the receiver operating curve (AUC), mean risk calibration (CAL), and negative predictive value (NPV). A secondary analysis was performed using Monte Carlo simulation (MCS) to normalize all EEG recordings to 48 h and use only the first hour of EEG as a "screening EEG" to generate predictions. RESULTS:RiskSLIM recreated the 2HELPS2B model. All models had comparable AUC: evaluation cohort (NN: 0.85, EN: 0.84, 2HELPS2B: 0.83) and MCS (NN: 0.82, EN; 0.82, 2HELPS2B: 0.81) and NPV (absence of seizures in the group that the models predicted to be low risk): evaluation cohort (NN: 97%, EN: 97%, 2HELPS2B: 97%) and MCS (NN: 97%, EN: 99%, 2HELPS2B: 97%). 2HELPS2B model was able to identify the largest proportion of low-risk patients. INTERPRETATION:For seizure risk stratification of hospitalized patients, the RiskSLIM generated 2HELPS2B model compares favorably to the complex NN and EN generated models. 2HELPS2B is able to accurately and quickly identify low-risk patients with only a 1-h screening EEG.
Electroencephalography-Derived Prognosis of Functional Recovery in Acute Stroke Through Machine Learning Approaches.
Chiarelli Antonio Maria,Croce Pierpaolo,Assenza Giovanni,Merla Arcangelo,Granata Giuseppe,Giannantoni Nadia Mariagrazia,Pizzella Vittorio,Tecchio Franca,Zappasodi Filippo
International journal of neural systems
Stroke, if not lethal, is a primary cause of disability. Early assessment of markers of recovery can allow personalized interventions; however, it is difficult to deliver indexes in the acute phase able to predict recovery. In this perspective, evaluation of electrical brain activity may provide useful information. A machine learning approach was explored here to predict post-stroke recovery relying on multi-channel electroencephalographic (EEG) recordings of few minutes performed at rest. A data-driven model, based on partial least square (PLS) regression, was trained on 19-channel EEG recordings performed within 10 days after mono-hemispheric stroke in 101 patients. The band-wise (delta: 1-4[Formula: see text]Hz, theta: 4-7[Formula: see text]Hz, alpha: 8-14[Formula: see text]Hz and beta: 15-30[Formula: see text]Hz) EEG effective powers were used as features to predict the recovery at 6 months (based on clinical status evaluated through the NIH Stroke Scale, NIHSS) in an optimized and cross-validated framework. In order to exploit the multimodal contribution to prognosis, the EEG-based prediction of recovery was combined with NIHSS scores in the acute phase and both were fed to a nonlinear support vector regressor (SVR). The prediction performance of EEG was at least as good as that of the acute clinical status scores. evaluation of the features exploited by the analysis highlighted a lower delta and higher alpha activity in patients showing a positive outcome, independently of the affected hemisphere. The multimodal approach showed better prediction capabilities compared to the acute NIHSS scores alone ([Formula: see text] versus [Formula: see text], AUC = 0.80 versus AUC = 0.70, [Formula: see text]). The multimodal and multivariate model can be used in acute phase to infer recovery relying on standard EEG recordings of few minutes performed at rest together with clinical assessment, to be exploited for early and personalized therapies. The easiness of performing EEG may allow such an approach to become a standard-of-care and, thanks to the increasing number of labeled samples, further improving the model predictive power.
A Biomarker for Discriminating Between Migraine With and Without Aura: Machine Learning on Functional Connectivity on Resting-State EEGs.
Frid Alex,Shor Meirav,Shifrin Alla,Yarnitsky David,Granovsky Yelena
Annals of biomedical engineering
Advanced analyses of electroencephalography (EEG) are rapidly becoming an important tool in understanding the brain's processing of pain. To date, it appears that none have been explored as a way of distinguishing between migraine patients with aura (MWA) vs. those without aura (MWoA). In this work, we apply a mixture of predictive, e.g., classification methods and attribute-selection techniques, and traditional explanatory, e.g., statistical, analyses on functional connectivity measures extracted from EEG signal acquired from at-rest participants (N = 52) during their interictal period and tested them against the distinction between MWA and MWoA. We show that a functional connectivity metric of EEG data obtained during resting state can serve as a sole biomarker to differentiate between MWA and MWoA. Using the proposed analysis, we not only have been able to present high classification results (average classification of 84.62%) but also to discuss the underlying neurophysiological mechanisms upon which our technique is based. Additionally, a more traditional statistical analysis on the selected features reveals that MWoA patients show higher than average connectivity in the Theta band (p = 0.03) at rest than MWAs. We propose that our data-driven analysis pipeline can be used for resting-EEG analysis in any clinical context.
EEG may serve as a biomarker in Huntington's disease using machine learning automatic classification.
Odish Omar F F,Johnsen Kristinn,van Someren Paul,Roos Raymund A C,van Dijk J Gert
Reliable markers measuring disease progression in Huntington's disease (HD), before and after disease manifestation, may guide a therapy aimed at slowing or halting disease progression. Quantitative electroencephalography (qEEG) may provide a quantification method for possible (sub)cortical dysfunction occurring prior to or concomitant with motor or cognitive disturbances observed in HD. In this pilot study we construct an automatic classifier distinguishing healthy controls from HD gene carriers using qEEG and derive qEEG features that correlate with clinical markers known to change with disease progression in HD, with the aim of exploring biomarker potential. We included twenty-six HD gene carriers (49.7 ± 8.5 years) and 25 healthy controls (52.7 ± 8.7 years). EEG was recorded for three minutes with subjects at rest. An EEG index was created by applying statistical pattern recognition to a large set of EEG features, which was subsequently tested using 10-fold cross-validation. The index resulted in a continuous variable ranging from 0 to 1: a low value indicating a state close to normal and a high value pointing to HD. qEEG features that correlate specifically with commonly used clinical markers in HD research were derived. The classification index had a specificity of 83%, a sensitivity of 83% and an accuracy of 83%. The area under the curve of the receiver operator characteristic curve was 0.9. qEEG analysis on subsets of electrophysiological features resulted in two highly significant correlations with clinical scores. The results of this pilot study suggest that qEEG may serve as a biomarker in HD. The indices correlating with modalities changing with the progression of the disease may lead to tools based on qEEG that help monitor efficacy in intervention studies.
Depression recognition using machine learning methods with different feature generation strategies.
Li Xiaowei,Zhang Xin,Zhu Jing,Mao Wandeng,Sun Shuting,Wang Zihan,Xia Chen,Hu Bin
Artificial intelligence in medicine
The diagnosis of depression almost exclusively depends on doctor-patient communication and scale analysis, which have the obvious disadvantages such as patient denial, poor sensitivity, subjective biases and inaccuracy. An objective, automated method that predicts clinical outcomes in depression is essential for increasing the accuracy of depression recognition and treatments. This paper aims at better recognizing depression using the transformation of EEG features and machine learning methods. An experiment based on emotional face stimuli task was conducted, and twenty-eight subjects' EEG data were recorded from 128-channel HydroCel Geodesic Sensor Net (HCGSN) by Net Station software. The Mini International Neuropsychiatric Interview (MINI) was used by psychiatrists as the criterion for diagnosis of depression patients. The power spectral density and activity were respectively extracted as original features using Auto-regress model and Hjorth algorithm with different time windows. Two separate approaches processed the features: ensemble learning and deep learning. For the ensemble learning, a deep forest transformed the original features to new features that potentially improve feature engineering and a support vector machine (SVM) that was applied as classifier. For deep learning method, we added spatial information of EEG caps to both features by image conversion and adopted convolutional neural network (CNN) to recognize them. The performance of both methods was evaluated for separated and total frequency bands. As a result, the best accuracy obtained was 89.02% when we used the ensemble model and power spectral density. The best accuracy of deep learning method was 84.75% using the activity. These experimental results prove the efficiency of the proposed methods and show that EEG could be used as a reliable indicator for depression recognition, which makes it possible for EEG-based portable system design and application in auxiliary depression recognition in the future.
Leveraging Machine Learning Approaches for Predicting Antidepressant Treatment Response Using Electroencephalography (EEG) and Clinical Data.
Jaworska Natalia,de la Salle Sara,Ibrahim Mohamed-Hamza,Blier Pierre,Knott Verner
Frontiers in psychiatry
Individuals with major depressive disorder (MDD) vary in their response to antidepressants. However, identifying objective biomarkers, prior to or early in the course of treatment that can predict antidepressant efficacy, remains a challenge. Individuals with MDD participated in a 12-week antidepressant pharmacotherapy trial. Electroencephalographic (EEG) data was collected before and 1 week post-treatment initiation in 51 patients. Response status at week 12 was established with the Montgomery-Asberg Depression Scale (MADRS), with a ≥50% decrease characterizing responders ( = 27/24 responders/non-responders). We used a machine learning (ML)-approach for predicting response status. We focused on Random Forests, though other ML methods were compared. First, we used a tree-based estimator to select a relatively small number of significant features from: (a) demographic/clinical data (age, sex, individual item/total MADRS scores at baseline, week 1, change scores); (b) scalp-level EEG power; (c) source-localized current density (via exact low-resolution electromagnetic tomography [eLORETA] software). Second, we applied kernel principal component analysis to reduce and map important features. Third, a set of ML models were constructed to classify response outcome based on mapped features. For each dataset, predictive features were extracted, followed by a model of all predictive features, and finally by a model of the predictive features. Fifty eLORETA features were predictive of response (across bands, both time-points); alpha/theta eLORETA features showed the highest predictive value. Eighty-eight scalp EEG features were predictive of response (across bands, both time-points), with theta/alpha being most predictive. Clinical/demographic data consisted of 31 features, with the most important being week 1 "concentration difficulty" scores. When all features were included into one model, its predictive utility was high (88% accuracy). When the important features were extracted in the final model, 12 predictive features emerged (78% accuracy), including baseline scalp-EEG frontopolar theta, parietal alpha and frontopolar alpha. These findings suggest that ML models of pre- and early treatment-emergent EEG profiles and clinical features can serve as tools for predicting antidepressant response. While this must be replicated using large independent samples, it lays the groundwork for research on personalized, "biomarker"-based treatment approaches.
A machine learning framework involving EEG-based functional connectivity to diagnose major depressive disorder (MDD).
Mumtaz Wajid,Ali Syed Saad Azhar,Yasin Mohd Azhar Mohd,Malik Aamir Saeed
Medical & biological engineering & computing
Major depressive disorder (MDD), a debilitating mental illness, could cause functional disabilities and could become a social problem. An accurate and early diagnosis for depression could become challenging. This paper proposed a machine learning framework involving EEG-derived synchronization likelihood (SL) features as input data for automatic diagnosis of MDD. It was hypothesized that EEG-based SL features could discriminate MDD patients and healthy controls with an acceptable accuracy better than measures such as interhemispheric coherence and mutual information. In this work, classification models such as support vector machine (SVM), logistic regression (LR) and Naïve Bayesian (NB) were employed to model relationship between the EEG features and the study groups (MDD patient and healthy controls) and ultimately achieved discrimination of study participants. The results indicated that the classification rates were better than chance. More specifically, the study resulted into SVM classification accuracy = 98%, sensitivity = 99.9%, specificity = 95% and f-measure = 0.97; LR classification accuracy = 91.7%, sensitivity = 86.66%, specificity = 96.6% and f-measure = 0.90; NB classification accuracy = 93.6%, sensitivity = 100%, specificity = 87.9% and f-measure = 0.95. In conclusion, SL could be a promising method for diagnosing depression. The findings could be generalized to develop a robust CAD-based tool that may help for clinical purposes.
Detection of Transient Bursts in the EEG of Preterm Infants using Time-Frequency Distributions and Machine Learning.
Murphy Brian M,Goulding Robert M,O'Toole John M
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
Short-duration bursts of spontaneous activity are important markers of maturation in the electroencephalogram (EEG) of premature infants. This paper examines the application of a feature-less machine learning approach for detecting these bursts. EEGs were recorded over the first 3 days of life for infants with a gestational age below 30 weeks. Bursts were annotated on the EEG from 36 infants. In place of feature extraction, the time-series EEG is transformed into a time-frequency distribution (TFD). A gradient boosting machine is then trained directly on the whole TFD using a leave-one-out procedure. TFD kernel parameters, length of the Doppler and lag windows, are selected within a nested cross-validation procedure during training. Results indicate that detection performance is sensitive to Doppler-window length but not lag-window length. Median area under the receiver operator characteristic for detection is 0.881 (inter-quartile range 0.850 to 0.913). Examination of feature importance highlights a critical wideband region <15 Hz in the TFD. Burst detection methods form an important component in any fully-automated brain-health index for the vulnerable preterm infant.
Machine learning without a feature set for detecting bursts in the EEG of preterm infants.
O' Toole John M,Boylan Geraldine B
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
Deep neural networks enable learning directly on the data without the domain knowledge needed to construct a feature set. This approach has been extremely successful in almost all machine learning applications. We propose a new framework that also learns directly from the data, without extracting a feature set. We apply this framework to detecting bursts in the EEG of premature infants. The EEG is recorded within days of birth in a cohort of infants without significant brain injury and born <; 30 weeks of gestation. The method first transforms the time-domain signal to the time-frequency domain and then trains a machine learning method, a gradient boosting machine, on each time-slice of the time-frequency distribution. We control for oversampling the time-frequency distribution with a significant reduction (<; 1%) in memory and computational complexity. The proposed method achieves similar accuracy to an existing multi-feature approach: area under the characteristic curve of 0.98 (with 95% confidence interval of 0.96 to 0.99), with a median sensitivity of 95% and median specificity of 94%. The proposed framework presents an accurate, simple, and computational efficient implementation as an alternative to both the deep learning approach and to the manual generation of a feature set.
Potential EEG biomarkers of sedation doses in intensive care patients unveiled by using a machine learning approach.
Sanz-García Ancor,Pérez-Romero Miriam,Pastor Jesús,Sola Rafael G,Vega-Zelaya Lorena,Vega Gema,Monasterio Fernando,Torrecilla Carmen,Pulido Paloma,Ortega Guillermo J
Journal of neural engineering
OBJECTIVE:Sedation of neurocritically ill patients is one of the most challenging situation in ICUs. Quantitative knowledge on the sedation effect on brain activity in that complex scenario could help to uncover new markers for sedation assessment. Hence, we aim to evaluate the existence of changes of diverse EEG-derived measures in deeply-sedated (RASS-Richmond agitation-sedation scale -4 and -5) neurocritically ill patients, and also whether sedation doses are related with those eventual changes. APPROACH:We performed an observational prospective cohort study in the intensive care unit of the Hospital de la Princesa. Twenty-six adult patients suffered from traumatic brain injury and subarachnoid hemorrhage were included in the present study. Long-term continuous electroencephalographic (EEG) recordings (2141 h) and hourly annotated information were used to determine the relationship between intravenous sedation infusion doses and network and spectral EEG measures. To do that, two different strategies were followed: assessment of the statistical dependence between both variables using the Spearman correlation rank and by performing an automatic classification method based on a machine learning algorithm. MAIN RESULTS:More than 60% of patients presented a correlation greater than 0.5 in at least one of the calculated EEG measures with the sedation dose. The automatic classification method presented an accuracy of 84.3% in discriminating between different sedation doses. In both cases the nodes' degree was the most relevant measurement. SIGNIFICANCE:The results presented here provide evidences of brain activity changes during deep sedation linked to sedation doses. Particularly, the capability of network EEG-derived measures in discriminating between different sedation doses could be the framework for the development of accurate methods for sedation levels assessment.
Use of Machine Learning for Predicting Escitalopram Treatment Outcome From Electroencephalography Recordings in Adult Patients With Depression.
Zhdanov Andrey,Atluri Sravya,Wong Willy,Vaghei Yasaman,Daskalakis Zafiris J,Blumberger Daniel M,Frey Benicio N,Giacobbe Peter,Lam Raymond W,Milev Roumen,Mueller Daniel J,Turecki Gustavo,Parikh Sagar V,Rotzinger Susan,Soares Claudio N,Brenner Colleen A,Vila-Rodriguez Fidel,McAndrews Mary Pat,Kleffner Killian,Alonso-Prieto Esther,Arnott Stephen R,Foster Jane A,Strother Stephen C,Uher Rudolf,Kennedy Sidney H,Farzan Faranak
JAMA network open
Importance:Social and economic costs of depression are exacerbated by prolonged periods spent identifying treatments that would be effective for a particular patient. Thus, a tool that reliably predicts an individual patient's response to treatment could significantly reduce the burden of depression. Objective:To estimate how accurately an outcome of escitalopram treatment can be predicted from electroencephalographic (EEG) data on patients with depression. Design, Setting, and Participants:This prognostic study used a support vector machine classifier to predict treatment outcome using data from the first Canadian Biomarker Integration Network in Depression (CAN-BIND-1) study. The CAN-BIND-1 study comprised 180 patients (aged 18-60 years) diagnosed with major depressive disorder who had completed 8 weeks of treatment. Of this group, 122 patients had EEG data recorded before the treatment; 115 also had EEG data recorded after the first 2 weeks of treatment. Interventions:All participants completed 8 weeks of open-label escitalopram (10-20 mg) treatment. Main Outcomes and Measures:The ability of EEG data to predict treatment outcome, measured as accuracy, specificity, and sensitivity of the classifier at baseline and after the first 2 weeks of treatment. The treatment outcome was defined in terms of change in symptom severity, measured by the Montgomery-Åsberg Depression Rating Scale, before and after 8 weeks of treatment. A patient was designated as a responder if the Montgomery-Åsberg Depression Rating Scale score decreased by at least 50% during the 8 weeks and as a nonresponder if the score decrease was less than 50%. Results:Of the 122 participants who completed a baseline EEG recording (mean [SD] age, 36.3 [12.7] years; 76 [62.3%] female), the classifier was able to identify responders with an estimated accuracy of 79.2% (sensitivity, 67.3%; specificity, 91.0%) when using only the baseline EEG data. For a subset of 115 participants who had additional EEG data recorded after the first 2 weeks of treatment, use of these data increased the accuracy to 82.4% (sensitivity, 79.2%; specificity, 85.5%). Conclusions and Relevance:These findings demonstrate the potential utility of EEG as a treatment planning tool for escitalopram therapy. Further development of the classification tools presented in this study holds the promise of expediting the search for optimal treatment for each patient.
Percept-related EEG classification using machine learning approach and features of functional brain connectivity.
Hramov Alexander E,Maksimenko Vladimir,Koronovskii Alexey,Runnova Anastasiya E,Zhuravlev Maxim,Pisarchik Alexander N,Kurths Jürgen
Chaos (Woodbury, N.Y.)
Machine learning is a promising approach for electroencephalographic (EEG) trials classification. Its efficiency is largely determined by the feature extraction and selection techniques reducing the dimensionality of input data. Dimensionality reduction is usually implemented via the mathematical approaches (e.g., principal component analysis, linear discriminant analysis, etc.) regardless of the origin of analyzed data. We hypothesize that since EEG features are determined by certain neurophysiological processes, they should have distinctive characteristics in spatiotemporal domain. If so, it is possible to specify the set of EEG principal features based on the prior knowledge about underlying neurophysiological processes. To test this hypothesis, we consider the classification of EEG trials related to the perception of ambiguous visual stimuli. We observe that EEG features, underlying the different ambiguous stimuli interpretations, are defined by the network properties of neuronal activity. Having analyzed functional neural interactions, we specify the brain area in which neural network architecture exhibits differences for different classes of EEG trials. We optimize the feedforward multilayer perceptron and develop a strategy for the training set selection to maximize the classification accuracy, being 85% when all channels are used. The revealed localization of the percept-related features allows about 95% accuracy, when the number of channels is reduced up to 90%. Obtained results can be used for classification of EEG trials associated with more complex cognitive tasks. Taking into account that cognitive activity is subserved by a distributed functional cortical network, its topological properties have to be considered when selecting optimal features for EEG trial classification.
Electroencephalography-based machine learning for cognitive profiling in Parkinson's disease: Preliminary results.
Betrouni Nacim,Delval Arnaud,Chaton Laurence,Defebvre Luc,Duits Annelien,Moonen Anja,Leentjens Albert F G,Dujardin Kathy
Movement disorders : official journal of the Movement Disorder Society
BACKGROUND:Cognitive symptoms are common in patients with Parkinson's disease. Characterization of a patient's cognitive profile is an essential step toward the identification of predictors of cognitive worsening. OBJECTIVE:The aim of this study was to investigate the use of the combination of resting-state EEG and data-mining techniques to build characterization models. METHODS:Dense EEG data from 118 patients with Parkinson's disease, classified into 5 different groups according to the severity of their cognitive impairments, were considered. Spectral power analysis within 7 frequency bands was performed on the EEG signals. The obtained quantitative EEG features of 100 patients were mined using 2 machine-learning algorithms to build and train characterization models, namely, support vector machines and k-nearest neighbors models. The models were then blindly tested on data from 18 patients. RESULTS:The overall classification accuracies were 84% and 88% for the support vector machines and k-nearest algorithms, respectively. The worst classifications were observed for patients from groups with small sample sizes, corresponding to patients with the severe cognitive deficits. Whereas for the remaining groups for whom an accurate diagnosis was required to plan the future healthcare, the classification was very accurate. CONCLUSION:These results suggest that EEG features computed from a daily clinical practice exploration modality in-that it is nonexpensive, available anywhere, and requires minimal cooperation from the patient-can be used as a screening method to identify the severity of cognitive impairment in patients with Parkinson's disease. © 2018 International Parkinson and Movement Disorder Society.
A novel multi-modal machine learning based approach for automatic classification of EEG recordings in dementia.
Ieracitano Cosimo,Mammone Nadia,Hussain Amir,Morabito Francesco C
Neural networks : the official journal of the International Neural Network Society
Electroencephalographic (EEG) recordings generate an electrical map of the human brain that are useful for clinical inspection of patients and in biomedical smart Internet-of-Things (IoT) and Brain-Computer Interface (BCI) applications. From a signal processing perspective, EEGs yield a nonlinear and nonstationary, multivariate representation of the underlying neural circuitry interactions. In this paper, a novel multi-modal Machine Learning (ML) based approach is proposed to integrate EEG engineered features for automatic classification of brain states. EEGs are acquired from neurological patients with Mild Cognitive Impairment (MCI) or Alzheimer's disease (AD) and the aim is to discriminate Healthy Control (HC) subjects from patients. Specifically, in order to effectively cope with nonstationarities, 19-channels EEG signals are projected into the time-frequency (TF) domain by means of the Continuous Wavelet Transform (CWT) and a set of appropriate features (denoted as CWT features) are extracted from δ, θ, α, α, β EEG sub-bands. Furthermore, to exploit nonlinear phase-coupling information of EEG signals, higher order statistics (HOS) are extracted from the bispectrum (BiS) representation. BiS generates a second set of features (denoted as BiS features) which are also evaluated in the five EEG sub-bands. The CWT and BiS features are fed into a number of ML classifiers to perform both 2-way (AD vs. HC, AD vs. MCI, MCI vs. HC) and 3-way (AD vs. MCI vs. HC) classifications. As an experimental benchmark, a balanced EEG dataset that includes 63 AD, 63 MCI and 63 HC is analyzed. Comparative results show that when the concatenation of CWT and BiS features (denoted as multi-modal (CWT+BiS) features) is used as input, the Multi-Layer Perceptron (MLP) classifier outperforms all other models, specifically, the Autoencoder (AE), Logistic Regression (LR) and Support Vector Machine (SVM). Consequently, our proposed multi-modal ML scheme can be considered a viable alternative to state-of-the-art computationally intensive deep learning approaches.
Quantitative EEG reactivity and machine learning for prognostication in hypoxic-ischemic brain injury.
Amorim Edilberto,van der Stoel Michelle,Nagaraj Sunil B,Ghassemi Mohammad M,Jing Jin,O'Reilly Una-May,Scirica Benjamin M,Lee Jong Woo,Cash Sydney S,Westover M Brandon
Clinical neurophysiology : official journal of the International Federation of Clinical Neurophysiology
OBJECTIVE:Electroencephalogram (EEG) reactivity is a robust predictor of neurological recovery after cardiac arrest, however interrater-agreement among electroencephalographers is limited. We sought to evaluate the performance of machine learning methods using EEG reactivity data to predict good long-term outcomes in hypoxic-ischemic brain injury. METHODS:We retrospectively reviewed clinical and EEG data of comatose cardiac arrest subjects. Electroencephalogram reactivity was tested within 72 h from cardiac arrest using sound and pain stimuli. A Quantitative EEG (QEEG) reactivity method evaluated changes in QEEG features (EEG spectra, entropy, and frequency features) during the 10 s before and after each stimulation. Good outcome was defined as Cerebral Performance Category of 1-2 at six months. Performance of a random forest classifier was compared against a penalized general linear model (GLM) and expert electroencephalographer review. RESULTS:Fifty subjects were included and sixteen (32%) had good outcome. Both QEEG reactivity methods had comparable performance to expert EEG reactivity assessment for good outcome prediction (mean AUC 0.8 for random forest vs. 0.69 for GLM vs. 0.69 for expert review, respectively; p non-significant). CONCLUSIONS:Machine-learning models utilizing quantitative EEG reactivity data can predict long-term outcome after cardiac arrest. SIGNIFICANCE:A quantitative approach to EEG reactivity assessment may support prognostication in cardiac arrest.
Use of machine learning in predicting clinical response to transcranial magnetic stimulation in comorbid posttraumatic stress disorder and major depression: A resting state electroencephalography study.
Zandvakili Amin,Philip Noah S,Jones Stephanie R,Tyrka Audrey R,Greenberg Benjamin D,Carpenter Linda L
Journal of affective disorders
BACKGROUND:Repetitive transcranial magnetic stimulation (TMS) is clinically effective for major depressive disorder (MDD) and investigational for other conditions including posttraumatic stress disorder (PTSD). Understanding the mechanisms of TMS action and developing biomarkers predicting response remain important goals. We applied a combination of machine learning and electroencephalography (EEG), testing whether machine learning analysis of EEG coherence would (1) predict clinical outcomes in individuals with comorbid MDD and PTSD, and (2) determine whether an individual had received a TMS course. METHODS:We collected resting-state 8-channel EEG before and after TMS (5 Hz to the left dorsolateral prefrontal cortex). We used Lasso regression and Support Vector Machine (SVM) to test the hypothesis that baseline EEG coherence predicted the outcome and to assess if EEG coherence changed after TMS. RESULTS:In our sample, clinical response to TMS were predictable based on pretreatment EEG coherence (n = 29). After treatment, 13/29 had more than 50% reduction in MDD self-report score 12/29 had more than 50% reduction in PTSD self-report score. For MDD, area under roc curve was for MDD was 0.83 (95% confidence interval 0.69-0.94) and for PTSD was 0.71 (95% confidence interval 0.54-0.87). SVM classifier was able to accurately assign EEG recordings to pre- and post-TMS treatment. The accuracy for Alpha, Beta, Theta and Delta bands was 75.4 ± 1.5%, 77.4 ± 1.4%, 73.8 ± 1.5%, and 78.6 ± 1.4%, respectively, all significantly better than chance (50%, p < 0.001). LIMITATION:Limitations of this work include lack of sham condition, modest sample size, and a sparse electrode array. Despite these methodological limitations, we found validated and clinically meaningful results. CONCLUSIONS:Machine learning successfully predicted non-response to TMS with high specificity, and identified pre- and post-TMS status using EEG coherence. This approach may provide mechanistic insights and may also become a clinically useful screening tool for TMS candidates.
Acute brain failure in severe sepsis: a prospective study in the medical intensive care unit utilizing continuous EEG monitoring.
Gilmore Emily J,Gaspard Nicolas,Choi Huimahn A,Cohen Emily,Burkart Kristin M,Chong David H,Claassen Jan,Hirsch Lawrence J
Intensive care medicine
PURPOSE:Investigate the prevalence, risk factors and impact of continuous EEG (cEEG) abnormalities on mortality through the 1-year follow-up period in patients with severe sepsis. METHODS:Prospective, single-center, observational study of consecutive patients admitted with severe sepsis to the Medical ICU at an academic medical center. RESULTS:A total of 98 patients with 100 episodes of severe sepsis were included; 49 patients (50%) were female, median age was 60 (IQR 52-74), the median non-neuro APACHE II score was 23.5 (IQR 18-28) and median non-neuro SOFA score was 8 (IQR 6-11). Twenty-five episodes had periodic discharges (PD), of which 11 had nonconvulsive seizures (NCS). No patient had NCS without PD. Prior neurological history was associated with a higher risk of PD or NCS (45 vs. 17%; CI 1.53-10.43), while the non-neuro APACHE II, non-neuro SOFA, severity of cardiovascular shock and presence of sedation during cEEG were associated with a lower risk of PD or NCS. Clinical seizures before cEEG were associated with a higher risk of nonconvulsive status epilepticus (24 vs. 6%; CI 1.42-19.94) while the non-neuro APACHE II and non-neuro SOFA scores were associated with a lower risk. Lack of EEG reactivity was present in 28% of episodes. In the survival analysis, a lack of EEG reactivity was associated with higher 1-year mortality [mean survival time 3.3 (95% CI 1.8-4.9) vs. 7.5 (6.4-8.7) months; p = 0.002] but the presence of PD or NCS was not [mean survival time 3.3 (95% CI 1.8-4.9) vs. 7.5 (6.4-8.7) months; p = 0.592]. Lack of reactivity was more frequent in patients on continuous sedation during cEEG. In patients with available 1-year data (34% of the episodes), 82% had good functional outcome (mRS ≤ 3, n = 27). There were no significant predictors of functional outcome, late cognition, and no patient with complete follow-up data developed late seizure or new epilepsy. CONCLUSIONS:NCS and PD are common in patients with severe sepsis and altered mental status. They were less frequent among the most severely sick patients and were not associated with outcome in this study. Lack of EEG reactivity was more frequent in patients on continuous sedation and was associated with mortality up to 1 year after discharge. Larger studies are needed to confirm these findings in a broader population and to further evaluate long-term cognitive outcome, risk of late seizure and epilepsy.
Early EEG for outcome prediction of postanoxic coma: prospective cohort study with cost-minimization analysis.
Sondag Lotte,Ruijter Barry J,Tjepkema-Cloostermans Marleen C,Beishuizen Albertus,Bosch Frank H,van Til Janine A,van Putten Michel J A M,Hofmeijer Jeannette
Critical care (London, England)
BACKGROUND:We recently showed that electroencephalography (EEG) patterns within the first 24 hours robustly contribute to multimodal prediction of poor or good neurological outcome of comatose patients after cardiac arrest. Here, we confirm these results and present a cost-minimization analysis. Early prognosis contributes to communication between doctors and family, and may prevent inappropriate treatment. METHODS:A prospective cohort study including 430 subsequent comatose patients after cardiac arrest was conducted at intensive care units of two teaching hospitals. Continuous EEG was started within 12 hours after cardiac arrest and continued up to 3 days. EEG patterns were visually classified as unfavorable (isoelectric, low-voltage, or burst suppression with identical bursts) or favorable (continuous patterns) at 12 and 24 hours after cardiac arrest. Outcome at 6 months was classified as good (cerebral performance category (CPC) 1 or 2) or poor (CPC 3, 4, or 5). Predictive values of EEG measures and cost-consequences from a hospital perspective were investigated, assuming EEG-based decision- making about withdrawal of life-sustaining treatment in the case of a poor predicted outcome. RESULTS:Poor outcome occurred in 197 patients (51% of those included in the analyses). Unfavorable EEG patterns at 24 hours predicted a poor outcome with specificity of 100% (95% CI 98-100%) and sensitivity of 29% (95% CI 22-36%). Favorable patterns at 12 hours predicted good outcome with specificity of 88% (95% CI 81-93%) and sensitivity of 51% (95% CI 42-60%). Treatment withdrawal based on an unfavorable EEG pattern at 24 hours resulted in a reduced mean ICU length of stay without increased mortality in the long term. This gave small cost reductions, depending on the timing of withdrawal. CONCLUSIONS:Early EEG contributes to reliable prediction of good or poor outcome of postanoxic coma and may lead to reduced length of ICU stay. In turn, this may bring small cost reductions.
Early Standard Electroencephalogram Abnormalities Predict Mortality in Septic Intensive Care Unit Patients.
Azabou Eric,Magalhaes Eric,Braconnier Antoine,Yahiaoui Lyria,Moneger Guy,Heming Nicholas,Annane Djillali,Mantz Jean,Chrétien Fabrice,Durand Marie-Christine,Lofaso Frédéric,Porcher Raphael,Sharshar Tarek,
INTRODUCTION:Sepsis is associated with increased mortality, delirium and long-term cognitive impairment in intensive care unit (ICU) patients. Electroencephalogram (EEG) abnormalities occurring at the acute stage of sepsis may correlate with severity of brain dysfunction. Predictive value of early standard EEG abnormalities for mortality in ICU septic patients remains to be assessed. METHODS:In this prospective, single center, observational study, standard EEG was performed, analyzed and classified according to both Synek and Young EEG scales, in consecutive patients acutely admitted in ICU for sepsis. Delirium, coma and the level of sedation were assessed at the time of EEG recording; and duration of sedation, occurrence of in-ICU delirium or death were assessed during follow-up. Adjusted analyses were carried out using multiple logistic regression. RESULTS:One hundred ten patients were included, mean age 63.8 (±18.1) years, median SAPS-II score 38 (29-55). At the time of EEG recording, 46 patients (42%) were sedated and 22 (20%) suffered from delirium. Overall, 54 patients (49%) developed delirium, of which 32 (29%) in the days after EEG recording. 23 (21%) patients died in the ICU. Absence of EEG reactivity was observed in 27 patients (25%), periodic discharges (PDs) in 21 (19%) and electrographic seizures (ESZ) in 17 (15%). ICU mortality was independently associated with a delta-predominant background (OR: 3.36; 95% CI [1.08 to 10.4]), absence of EEG reactivity (OR: 4.44; 95% CI [1.37-14.3], PDs (OR: 3.24; 95% CI [1.03 to 10.2]), Synek grade ≥ 3 (OR: 5.35; 95% CI [1.66-17.2]) and Young grade > 1 (OR: 3.44; 95% CI [1.09-10.8]) after adjustment to Simplified Acute Physiology Score (SAPS-II) at admission and level of sedation. Delirium at the time of EEG was associated with ESZ in non-sedated patients (32% vs 10%, p = 0.037); with Synek grade ≥ 3 (36% vs 7%, p< 0.05) and Young grade > 1 (36% vs 17%, p< 0.001). Occurrence of delirium in the days after EEG was associated with a delta-predominant background (48% vs 15%, p = 0.001); absence of reactivity (39% vs 10%, p = 0.003), Synek grade ≥ 3 (42% vs 17%, p = 0.001) and Young grade >1 (58% vs 17%, p = 0.0001). CONCLUSIONS:In this prospective cohort of 110 septic ICU patients, early standard EEG was significantly disturbed. Absence of EEG reactivity, a delta-predominant background, PDs, Synek grade ≥ 3 and Young grade > 1 at day 1 to 3 following admission were independent predictors of ICU mortality and were associated with occurence of delirium. ESZ and PDs, found in about 20% of our patients. Their prevalence could have been higher, with a still higher predictive value, if they had been diagnosed more thoroughly using continuous EEG.
High-Frequency Oscillations in the Scalp EEG of Intensive Care Unit Patients With Altered Level of Consciousness.
Ferrari-Marinho Taissa,Perucca Piero,Amiri Mina,Dubeau Francois,Gotman Jean,Caboclo Luis Otavio
Journal of clinical neurophysiology : official publication of the American Electroencephalographic Society
PURPOSE:In comatose patients, distinguishing between nonconvulsive status epilepticus and diffuse structural or metabolic encephalopathies is often challenging. Both conditions can generate periodic discharges on EEG with similar morphology and periodicity. We investigated the occurrence of high-frequency oscillations-potential biomarkers of epileptogenesis-on scalp EEG of comatose patients with periodic discharges in the EEG. METHODS:Fifteen patients were included. Patients were divided into three groups, according to underlying etiology: Group 1, seizure related; group 2, structural; group 3, nonstructural. EEG recordings were compared with respect to the presence and rates of gamma (30-80 Hz) and ripples (80-250 Hz). RESULTS:Patients were 23 to 106 years old (median, 68 years); 60% were female. 206 channels were eligible for analysis (median, 15 channels/patient). Overall, 43% of channels showed gamma, and 24% had ripples. Group 2 showed the highest proportion of channels with gamma (47%), followed by group 1 (38%) and group 3 (36%). Mean gamma rates were higher in group 2 (4.65 gamma/min/channel) than in group 1 (1.52) and group 3 (1.44) (P < 0.001). Group 2 showed the highest proportion of channels with ripples (29.2%), followed by group 1 (15%) and group 3 (24.2%). Mean ripple rates were higher in group 2 (5.09 ripple/min/channel) than in group 1 (0.96) and group 3 (0.83) (P < 0.001). CONCLUSIONS:Fast oscillations, including high-frequency oscillations, can be detected in scalp EEG of patients with altered consciousness. High rates of fast activity may suggest an underlying structural brain lesion. Future studies are needed to determine whether fast oscillations in the setting of acute/subacute brain lesions are a biomarker of subsequent development of human epilepsy.