Ultrafast review of ambulatory EEGs with deep learning.

OBJECTIVE: Interictal epileptiform discharges (IED) are hallmark biomarkers of epilepsy which are typically detected through visual analysis. Deep learning has shown potential in automating IED detection, which could reduce the burden of visual analysis in clinical practice. This is particularly relevant for ambulatory electroencephalograms (EEGs), as these entail longer review times. METHODS: We applied a previously trained neural network to an independent dataset of 100 ambulatory EEGs (average duration 20.6 h). From these, 42 EEGs contained IEDs, 25 were abnormal without IEDs and 33 were normal. The algorithm flagged 2 second epochs that it considered IEDs. The EEGs were provided to an expert, who used NeuroCenter EEG to review the recordings. The expert concluded if each recording contained IEDs, and was timed during the process. RESULTS: The conclusion of the reviewer was the same as the EEG report in 97% of the recordings. Three EEGs contained IEDs that were not detected based on the flagged epochs. Review time for the 100 EEGs was approximately 4 h, with half of the recordings taking <2 minutes to review. CONCLUSIONS: Our network can be used to reduce time spent on visual analysis in the clinic by 50-75 times with high reliability. SIGNIFICANCE: Given the large time reduction potential and high success rate, this algorithm can be used in the clinic to aid in visual analysis.

Neurophysiology State Dynamics Underlying Acute Neurologic Recovery After Cardiac Arrest.

BACKGROUND AND OBJECTIVES: Epileptiform activity and burst suppression are neurophysiology signatures reflective of severe brain injury after cardiac arrest. We aimed to delineate the evolution of coma neurophysiology feature ensembles associated with recovery from coma after cardiac arrest. METHODS: Adults in acute coma after cardiac arrest were included in a retrospective database involving 7 hospitals. The combination of 3 quantitative EEG features (burst suppression ratio [BSup], spike frequency [SpF], and Shannon entropy [En]) was used to define 5 distinct neurophysiology states: epileptiform high entropy (EHE: SpF ≥4 per minute and En ≥5); epileptiform low entropy (ELE: SpF ≥4 per minute and <5 En); nonepileptiform high entropy (NEHE: SpF <4 per minute and ≥5 En); nonepileptiform low entropy (NELE: SpF <4 per minute and <5 En), and burst suppression (BSup ≥50% and SpF <4 per minute). State transitions were measured at consecutive 6-hour blocks between 6 and 84 hours after return of spontaneous circulation. Good neurologic outcome was defined as best cerebral performance category 1-2 at 3-6 months. RESULTS: One thousand thirty-eight individuals were included (50,224 hours of EEG), and 373 (36%) had good outcome. Individuals with EHE state had a 29% rate of good outcome, while those with ELE had 11%. Transitions out of an EHE or BSup state to an NEHE state were associated with good outcome (45% and 20%, respectively). No individuals with ELE state lasting >15 hours had good recovery. DISCUSSION: Transition to high entropy states is associated with an increased likelihood of good outcome despite preceding epileptiform or burst suppression states. High entropy may reflect mechanisms of resilience to hypoxic-ischemic brain injury.

Preservation of thalamocortical circuitry is essential for good recovery after cardiac arrest.

Continuous electroencephalographam (EEG) monitoring contributes to prediction of neurological outcome in comatose cardiac arrest survivors. While the phenomenology of EEG abnormalities in postanoxic encephalopathy is well known, the pathophysiology, especially the presumed role of selective synaptic failure, is less understood. To further this understanding, we estimate biophysical model parameters from the EEG power spectra from individual patients with a good or poor recovery from a postanoxic encephalopathy. This biophysical model includes intracortical, intrathalamic, and corticothalamic synaptic strengths, as well as synaptic time constants and axonal conduction delays. We used continuous EEG measurements from hundred comatose patients recorded during the first 48 h postcardiac arrest, 50 with a poor neurological outcome [cerebral performance category ( CPC = 5 ) ] and 50 with a good neurological outcome ( CPC = 1 ). We only included patients that developed (dis-)continuous EEG activity within 48 h postcardiac arrest. For patients with a good outcome, we observed an initial relative excitation in the corticothalamic loop and corticothalamic propagation that subsequently evolved towards values observed in healthy controls. For patients with a poor outcome, we observed an initial increase in the cortical excitation-inhibition ratio, increased relative inhibition in the corticothalamic loop, delayed corticothalamic propagation of neuronal activity, and severely prolonged synaptic time constants that did not return to physiological values. We conclude that the abnormal EEG evolution in patients with a poor neurological recovery after cardiac arrest may result from persistent and selective synaptic failure that includes corticothalamic circuitry and also delayed corticothalamic propagation.

Early EEG monitoring predicts clinical outcome in patients with moderate to severe traumatic brain injury.

There is a need for reliable predictors in patients with moderate to severe traumatic brain injury to assist clinical decision making. We assess the ability of early continuous EEG monitoring at the intensive care unit (ICU) in patients with traumatic brain injury (TBI) to predict long term clinical outcome and evaluate its complementary value to current clinical standards. We performed continuous EEG measurements in patients with moderate to severe TBI during the first week of ICU admission. We assessed the Extended Glasgow Outcome Scale (GOSE) at 12 months, dichotomized into poor (GOSE 1-3) and good (GOSE 4-8) outcome. We extracted EEG spectral features, brain symmetry index, coherence, aperiodic exponent of the power spectrum, long range temporal correlations, and broken detailed balance. A random forest classifier using feature selection was trained to predict poor clinical outcome based on EEG features at 12, 24, 48, 72 and 96 h after trauma. We compared our predictor with the IMPACT score, the best available predictor, based on clinical, radiological and laboratory findings. In addition we created a combined model using EEG as well as the clinical, radiological and laboratory findings. We included hundred-seven patients. The best prediction model using EEG parameters was found at 72 h after trauma with an AUC of 0.82 (0.69-0.92), specificity of 0.83 (0.67-0.99) and sensitivity of 0.74 (0.63-0.93). The IMPACT score predicted poor outcome with an AUC of 0.81 (0.62-0.93), sensitivity of 0.86 (0.74-0.96) and specificity of 0.70 (0.43-0.83). A model using EEG and clinical, radiological and laboratory parameters resulted in a better prediction of poor outcome (p < 0.001) with an AUC of 0.89 (0.72-0.99), sensitivity of 0.83 (0.62-0.93) and specificity of 0.85 (0.75-1.00). EEG features have potential use for predicting clinical outcome and decision making in patients with moderate to severe TBI and provide complementary information to current clinical standards.

Comparison of state-of-the-art deep learning architectures for detection of freezing of gait in Parkinson's disease.

Introduction: Freezing of gait (FOG) is one of the most debilitating motor symptoms experienced by patients with Parkinson's disease (PD). FOG detection is possible using acceleration data from wearable sensors, and a convolutional neural network (CNN) is often used to determine the presence of FOG epochs. We compared the performance of a standard CNN for the detection of FOG with two more complex networks, which are well suited for time series data, the MiniRocket and the InceptionTime. Methods: We combined acceleration data of people with PD across four studies. The final data set was split into a training (80%) and hold-out test (20%) set. A fifth study was included as an unseen test set. The data were windowed (2 s) and five-fold cross-validation was applied. The CNN, MiniRocket, and InceptionTime models were evaluated using a receiver operating characteristic (ROC) curve and its area under the curve (AUC). Multiple sensor configurations were evaluated for the best model. The geometric mean was subsequently calculated to select the optimal threshold. The selected model and threshold were evaluated on the hold-out and unseen test set. Results: A total of 70 participants (23.7 h, 9% FOG) were included in this study for training and testing, and in addition, 10 participants provided an unseen test set (2.4 h, 11% FOG). The CNN performed best (AUC = 0.86) in comparison to the InceptionTime (AUC = 0.82) and MiniRocket (AUC = 0.76) models. For the CNN, we found a similar performance for a seven-sensor configuration (lumbar, upper and lower legs and feet; AUC = 0.86), six-sensor configuration (upper and lower legs and feet; AUC = 0.87), and two-sensor configuration (lower legs; AUC = 0.86). The optimal threshold of 0.45 resulted in a sensitivity of 77% and a specificity of 58% for the hold-out set (AUC = 0.72), and a sensitivity of 85% and a specificity of 68% for the unseen test set (AUC = 0.90). Conclusion: We confirmed that deep learning can be used to detect FOG in a large, heterogeneous dataset. The CNN model outperformed more complex networks. This model could be employed in future personalized interventions, with the ultimate goal of using automated FOG detection to trigger real-time cues to alleviate FOG in daily life.

Outcome Prediction of Postanoxic Coma: A Comparison of Automated Electroencephalography Analysis Methods.

BACKGROUND: To compare three computer-assisted quantitative electroencephalography (EEG) prediction models for the outcome prediction of comatose patients after cardiac arrest regarding predictive performance and robustness to artifacts. METHODS: A total of 871 continuous EEGs recorded up to 3 days after cardiac arrest in intensive care units of five teaching hospitals in the Netherlands were retrospectively analyzed. Outcome at 6 months was dichotomized as "good" (Cerebral Performance Category 1-2) or "poor" (Cerebral Performance Category 3-5). Three prediction models were implemented: a logistic regression model using two quantitative features, a random forest model with nine features, and a deep learning model based on a convolutional neural network. Data from two centers were used for training and fivefold cross-validation (n = 663), and data from three other centers were used for external validation (n = 208). Model output was the probability of good outcome. Predictive performances were evaluated by using receiver operating characteristic analysis and the calculation of predictive values. Robustness to artifacts was evaluated by using an artifact rejection algorithm, manually added noise, and randomly flattened channels in the EEG. RESULTS: The deep learning network showed the best overall predictive performance. On the external test set, poor outcome could be predicted by the deep learning network at 24 h with a sensitivity of 54% (95% confidence interval [CI] 44-64%) at a false positive rate (FPR) of 0% (95% CI 0-2%), significantly higher than the logistic regression (sensitivity 33%, FPR 0%) and random forest models (sensitivity 13%, FPR, 0%) (p < 0.05). Good outcome at 12 h could be predicted by the deep learning network with a sensitivity of 78% (95% CI 52-100%) at a FPR of 12% (95% CI 0-24%) and by the logistic regression model with a sensitivity of 83% (95% CI 83-83%) at a FPR of 3% (95% CI 3-3%), both significantly higher than the random forest model (sensitivity 1%, FPR 0%) (p < 0.05). The results of the deep learning network were the least affected by the presence of artifacts, added white noise, and flat EEG channels. CONCLUSIONS: A deep learning model outperformed logistic regression and random forest models for reliable, robust, EEG-based outcome prediction of comatose patients after cardiac arrest.

Effects of targeted temperature management at 33 °C vs. 36 °C on comatose patients after cardiac arrest stratified by the severity of encephalopathy.

OBJECTIVES: To assess neurological outcome after targeted temperature management (TTM) at 33 °C vs. 36 °C, stratified by the severity of encephalopathy based on EEG-patterns at 12 and 24 h. DESIGN: Post hoc analysis of prospective cohort study. SETTING: Five Dutch Intensive Care units. PATIENTS: 479 adult comatose post-cardiac arrest patients. INTERVENTIONS: TTM at 33 °C (n = 270) or 36 °C (n = 209) and continuous EEG monitoring. MEASUREMENTS AND MAIN RESULTS: Outcome according to the cerebral performance category (CPC) score at 6 months post-cardiac arrest was similar after 33 °C and 36 °C. However, when stratified by the severity of encephalopathy based on EEG-patterns at 12 and 24 h after cardiac arrest, the proportion of good outcome (CPC 1-2) in patients with moderate encephalopathy was significantly larger after TTM at 33 °C (66% vs. 45%; Odds Ratios 2.38, 95% CI = 1.32-4.30; p = 0.004). In contrast, with mild encephalopathy, there was no statistically significant difference in the proportion of patients with good outcome between 33 °C and 36 °C (88% vs. 81%; OR 1.68, 95% CI = 0.65-4.38; p = 0.282). Ordinal regression analysis showed a shift towards higher CPC scores when treated with TTM 33 °C as compared with 36 °C in moderate encephalopathy (cOR 2.39; 95% CI = 1.40-4.08; p = 0.001), but not in mild encephalopathy (cOR 0.81 95% CI = 0.41-1.59; p = 0.537). Adjustment for initial cardiac rhythm and cause of arrest did not change this relationship. CONCLUSIONS: Effects of TTM probably depend on the severity of encephalopathy in comatose patients after cardiac arrest. These results support inclusion of predefined subgroup analyses based on EEG measures of the severity of encephalopathy in future clinical trials.

Treating Rhythmic and Periodic EEG Patterns in Comatose Survivors of Cardiac Arrest.

BACKGROUND: Whether the treatment of rhythmic and periodic electroencephalographic (EEG) patterns in comatose survivors of cardiac arrest improves outcomes is uncertain. METHODS: We conducted an open-label trial of suppressing rhythmic and periodic EEG patterns detected on continuous EEG monitoring in comatose survivors of cardiac arrest. Patients were randomly assigned in a 1:1 ratio to a stepwise strategy of antiseizure medications to suppress this activity for at least 48 consecutive hours plus standard care (antiseizure-treatment group) or to standard care alone (control group); standard care included targeted temperature management in both groups. The primary outcome was neurologic outcome according to the score on the Cerebral Performance Category (CPC) scale at 3 months, dichotomized as a good outcome (CPC score indicating no, mild, or moderate disability) or a poor outcome (CPC score indicating severe disability, coma, or death). Secondary outcomes were mortality, length of stay in the intensive care unit (ICU), and duration of mechanical ventilation. RESULTS: We enrolled 172 patients, with 88 assigned to the antiseizure-treatment group and 84 to the control group. Rhythmic or periodic EEG activity was detected a median of 35 hours after cardiac arrest; 98 of 157 patients (62%) with available data had myoclonus. Complete suppression of rhythmic and periodic EEG activity for 48 consecutive hours occurred in 49 of 88 patients (56%) in the antiseizure-treatment group and in 2 of 83 patients (2%) in the control group. At 3 months, 79 of 88 patients (90%) in the antiseizure-treatment group and 77 of 84 patients (92%) in the control group had a poor outcome (difference, 2 percentage points; 95% confidence interval, -7 to 11; P = 0.68). Mortality at 3 months was 80% in the antiseizure-treatment group and 82% in the control group. The mean length of stay in the ICU and mean duration of mechanical ventilation were slightly longer in the antiseizure-treatment group than in the control group. CONCLUSIONS: In comatose survivors of cardiac arrest, the incidence of a poor neurologic outcome at 3 months did not differ significantly between a strategy of suppressing rhythmic and periodic EEG activity with the use of antiseizure medication for at least 48 hours plus standard care and standard care alone. (Funded by the Dutch Epilepsy Foundation; TELSTAR ClinicalTrials.gov number, NCT02056236.).

Predicting Neurological Outcome From Electroencephalogram Dynamics in Comatose Patients After Cardiac Arrest With Deep Learning.

OBJECTIVE: Most cardiac arrest patients who are successfully resuscitated are initially comatose due to hypoxic-ischemic brain injury. Quantitative electroencephalography (EEG) provides valuable prognostic information. However, prior approaches largely rely on snapshots of the EEG, without taking advantage of temporal information. METHODS: We present a recurrent deep neural network with the goal of capturing temporal dynamics from longitudinal EEG data to predict long-term neurological outcomes. We utilized a large international dataset of continuous EEG recordings from 1,038 cardiac arrest patients from seven hospitals in Europe and the US. Poor outcome was defined as a Cerebral Performance Category (CPC) score of 3-5, and good outcome as CPC score 0-2 at 3 to 6-months after cardiac arrest. Model performance is evaluated using 5-fold cross validation. RESULTS: The proposed approach provides predictions which improve over time, beginning from an area under the receiver operating characteristic curve (AUC-ROC) of 0.78 (95% CI: 0.72-0.81) at 12 hours, and reaching 0.88 (95% CI: 0.85-0.91) by 66 h after cardiac arrest. At 66 h, (sensitivity, specificity) points of interest on the ROC curve for predicting poor outcomes were (32,99)%, (55,95)%, and (62,90)%, (99,23)%, (95,47)%, and (90,62)%; whereas for predicting good outcome, the corresponding operating points were (17,99)%, (47,95)%, (62,90)%, (99,19)%, (95,48)%, (70,90)%. Moreover, the model provides predicted probabilities that closely match the observed frequencies of good and poor outcomes (calibration error 0.04). CONCLUSIONS AND SIGNIFICANCE: These findings suggest that accounting for EEG trend information can substantially improve prediction of neurologic outcomes for patients with coma following cardiac arrest.

Predicting neurological outcome in comatose patients after cardiac arrest with multiscale deep neural networks.

OBJECTIVE: Electroencephalography (EEG) is an important tool for neurological outcome prediction after cardiac arrest. However, the complexity of continuous EEG data limits timely and accurate interpretation by clinicians. We develop a deep neural network (DNN) model to leverage complex EEG trends for early and accurate assessment of cardiac arrest coma recovery likelihood. METHODS: We developed a multiscale DNN combining convolutional neural networks (CNN) and recurrent neural networks (long short-term memory [LSTM]) using EEG and demographic information (age, gender, shockable rhythm) from a multicenter cohort of 1,038 cardiac arrest patients. The CNN learns EEG feature representations while the multiscale LSTM captures short-term and long-term EEG dynamics on multiple time scales. Poor outcome is defined as a Cerebral Performance Category (CPC) score of 3-5 and good outcome as CPC score 1-2 at 3-6 months after cardiac arrest. Performance is evaluated using area under the receiver operating characteristic curve (AUC) and calibration error. RESULTS: Model performance increased with EEG duration, with AUC increasing from 0.83 (95% Confidence Interval [CI] 0.79-0.87 at 12h to 0.91 (95%CI 0.88-0.93) at 66h. Sensitivity of good and poor outcome prediction was 77% and 75% at a specificity of 90%, respectively. Sensitivity of poor outcome was 50% at a specificity of 99%. Predicted probability was well matched to the observation frequency of poor outcomes, with a calibration error of 0.11 [0.09-0.14]. CONCLUSIONS: These results demonstrate that incorporating EEG evolution over time improves the accuracy of neurologic outcome prediction for patients with coma after cardiac arrest.

Modulation of the Somatosensory Evoked Potential by Attention and Spinal Cord Stimulation.

Introduction: Spinal Cord Stimulation (SCS) is a last-resort treatment for patients with intractable chronic pain in whom pharmacological and other treatments have failed. Conventional tonic SCS is accompanied by tingling sensations. More recent stimulation protocols like burst SCS are not sensed by the patient while providing similar levels of pain relief. It has been previously reported that conventional tonic SCS can attenuate sensory-discriminative processing in several brain areas, but that burst SCS might have additional effects on the medial, motivational-affective pain system. In this explorative study we assessed the influence of attention on the somatosensory evoked brain responses under conventional tonic SCS as well as burst SCS regime. Methods: Twelve chronic pain patients with an implanted SCS device had 2-weeks evaluation periods with three different SCS settings (conventional tonic SCS, burst SCS, and sham SCS). At the end of each period, an electro-encephalography (EEG) measurement was done, at which patients received transcutaneous electrical pulses at the tibial nerve to induce somatosensory evoked potentials (SEP). SEP data was acquired while patients were attending the applied pulses and while they were mind wandering. The effects of attention as well as SCS regimes on the SEP were analyzed by comparing amplitudes of early and late latencies at the vertex as well as brain activity at full cortical maps. Results: Pain relief obtained by the various SCS settings varied largely among patients. Early SEP responses were not significantly affected by attention nor SCS settings (i.e., burst, tonic, and sham). However, late SEP responses (P300) were reduced with tonic and burst SCS: conventional tonic SCS reduced P300 brain activity in the unattended condition, while burst SCS reduced P300 brain activity in both attended and unattended conditions. Conclusion: Burst spinal cord stimulation for the treatment of chronic pain seems to reduce cortical attention that is or can be directed to somatosensory stimuli to a larger extent than conventional spinal cord stimulation treatment. This is a first step in understanding why in selected chronic pain patients burst SCS is more effective than tonic SCS and how neuroimaging could assist in personalizing SCS treatment.

Machine learning for detection of interictal epileptiform discharges.

The electroencephalogram (EEG) is a fundamental tool in the diagnosis and classification of epilepsy. In particular, Interictal Epileptiform Discharges (IEDs) reflect an increased likelihood of seizures and are routinely assessed by visual analysis of the EEG. Visual assessment is, however, time consuming and prone to subjectivity, leading to a high misdiagnosis rate and motivating the development of automated approaches. Research towards automating IED detection started 45 years ago. Approaches range from mimetic methods to deep learning techniques. We review different approaches to IED detection, discussing their performance and limitations. Traditional machine learning and deep learning methods have yielded the best results so far and their application in the field is still growing. Standardization of datasets and outcome measures is necessary to compare models more objectively and decide which should be implemented in a clinical setting.

Efficient use of clinical EEG data for deep learning in epilepsy.

OBJECTIVE: Automating detection of Interictal Epileptiform Discharges (IEDs) in electroencephalogram (EEG) recordings can reduce the time spent on visual analysis for the diagnosis of epilepsy. Deep learning has shown potential for this purpose, but the scarceness of expert annotated data creates a bottleneck in the process. METHODS: We used EEGs from 50 patients with focal epilepsy, 49 patients with generalized epilepsy (IEDs were visually labeled by experts) and 67 controls. The data was filtered, downsampled and cut into two second epochs. We increased the number of input samples containing IEDs through temporal shifting and using different montages. A VGG C convolutional neural network was trained to detect IEDs. RESULTS: Using the dataset with more samples, we reduced the false positive rate from 2.11 to 0.73 detections per minute at the intersection of sensitivity and specificity. Sensitivity increased from 63% to 96% at 99% specificity. The model became less sensitive to the position of the IED in the epoch and montage. CONCLUSIONS: Temporal shifting and use of different EEG montages improves performance of deep neural networks in IED detection. SIGNIFICANCE: Dataset augmentation can reduce the need for expert annotation, facilitating the training of neural networks, potentially leading to a fundamental shift in EEG analysis.

EEG functional connectivity contributes to outcome prediction of postanoxic coma.

OBJECTIVE: To investigate the additional value of EEG functional connectivity features, in addition to non-coupling EEG features, for outcome prediction of comatose patients after cardiac arrest. METHODS: Prospective, multicenter cohort study. Coherence, phase locking value, and mutual information were calculated in 19-channel EEGs at 12 h, 24 h and 48 h after cardiac arrest. Three sets of machine learning classification models were trained and validated with functional connectivity, EEG non-coupling features, and a combination of these. Neurological outcome was assessed at six months and categorized as "good" (Cerebral Performance Category [CPC] 1-2) or "poor" (CPC 3-5). RESULTS: We included 594 patients (46% good outcome). A sensitivity of 51% (95% CI: 34-56%) at 100% specificity in predicting poor outcome was achieved by the best functional connectivity-based classifier at 12 h after cardiac arrest, while the best non-coupling-based model reached a sensitivity of 32% (0-54%) at 100% specificity using data at 12 h and 48 h. Combination of both sets of features achieved a sensitivity of 73% (50-77%) at 100% specificity. CONCLUSION: Functional connectivity measures improve EEG based prediction models for poor outcome of postanoxic coma. SIGNIFICANCE: Functional connectivity features derived from early EEG hold potential to improve outcome prediction of coma after cardiac arrest.

Relevance of Somatosensory Evoked Potential Amplitude After Cardiac Arrest.

Objective: We present relations of SSEP amplitude with neurological outcome and of SSEP amplitude with EEG amplitude in comatose patients after cardiac arrest. Methods: This is a post hoc analysis of a prospective cohort study in comatose patients after cardiac arrest. Amplitude of SSEP recordings obtained within 48-72 h, and EEG patterns obtained at 12 and 24h after cardiac arrest were related to good (CPC 1-2) or poor (CPC 3-5) outcome at 6 months. In 39% of the study population multiple SSEP measurements were performed. Additionally, SSEP amplitude was related to mean EEG amplitude. Results: We included 138 patients (77% poor outcome). Absent SSEP responses, a N20 amplitude <0.4 µV within 48-72 h, and suppressed or synchronous EEG with suppressed background at 12 or 24 h after cardiac arrest were invariably associated with a poor outcome. Combined, these tests reached a sensitivity for prediction of poor outcome up to 58 at 100% specificity. N20 amplitude increased with a mean of 0.55 µV per day in patients with a poor outcome, and remained stable with a good outcome. There was no statistically significant correlation between SSEP and EEG amplitudes in 182 combined SSEP and EEG measurements (R 2 < 0.01). Conclusions: N20 amplitude <0.4 µV is invariably associated with poor outcome. There is no correlation between SSEP and EEG amplitude. Significance: SSEP amplitude analysis may contribute to outcome prediction after cardiac arrest.

Predicting outcome in patients with moderate to severe traumatic brain injury using electroencephalography.

BACKGROUND: Better outcome prediction could assist in reliable quantification and classification of traumatic brain injury (TBI) severity to support clinical decision-making. We developed a multifactorial model combining quantitative electroencephalography (qEEG) measurements and clinically relevant parameters as proof of concept for outcome prediction of patients with moderate to severe TBI. METHODS: Continuous EEG measurements were performed during the first 7 days of ICU admission. Patient outcome at 12 months was dichotomized based on the Extended Glasgow Outcome Score (GOSE) as poor (GOSE 1-2) or good (GOSE 3-8). Twenty-three qEEG features were extracted. Prediction models were created using a Random Forest classifier based on qEEG features, age, and mean arterial blood pressure (MAP) at 24, 48, 72, and 96 h after TBI and combinations of two time intervals. After optimization of the models, we added parameters from the International Mission for Prognosis And Clinical Trial Design (IMPACT) predictor, existing of clinical, CT, and laboratory parameters at admission. Furthermore, we compared our best models to the online IMPACT predictor. RESULTS: Fifty-seven patients with moderate to severe TBI were included and divided into a training set (n = 38) and a validation set (n = 19). Our best model included eight qEEG parameters and MAP at 72 and 96 h after TBI, age, and nine other IMPACT parameters. This model had high predictive ability for poor outcome on both the training set using leave-one-out (area under the receiver operating characteristic curve (AUC) = 0.94, specificity 100%, sensitivity 75%) and validation set (AUC = 0.81, specificity 75%, sensitivity 100%). The IMPACT predictor independently predicted both groups with an AUC of 0.74 (specificity 81%, sensitivity 65%) and 0.84 (sensitivity 88%, specificity 73%), respectively. CONCLUSIONS: Our study shows the potential of multifactorial Random Forest models using qEEG parameters to predict outcome in patients with moderate to severe TBI.

Association between somatosensory evoked potentials and EEG in comatose patients after cardiac arrest.

OBJECTIVE: To analyze the association between SSEP results and EEG results in comatose patients after cardiac arrest, including the added value of repeated SSEP measurements. METHODS: Continuous EEG was measured in 619 patients during the first 3-5 days after cardiac arrest. SSEPs were recorded daily in the first 55 patients, and on indication in later patients. EEGs were visually classified at 12, 24, 48, and 72 h after cardiac arrest, and at the time of SSEP. Outcome at 6 m was dichotomized as good (Cerebral Performance Category 1-2) or poor (CPC 3-5). SSEP and EEG results were related to outcome. Additionally, SSEP results were related to the EEG patterns at the time of SSEP. RESULTS: Absent SSEP responses and suppressed or synchronous EEG on suppressed background ≥24 h after cardiac arrest were invariably associated with poor outcome. SSEP and EEG identified different patients with poor outcome (joint sensitivity 39% at specificity 100%). N20 responses were always preserved in continuous traces at >8 Hz. Absent SSEPs did not re-emerge during the first five days. CONCLUSIONS: SSEP and EEG results may diverge after cardiac arrest. SIGNIFICANCE: SSEP and EEG together identify more patients without chance of recovery than one of these alone.

Early electroencephalography for outcome prediction of postanoxic coma: A prospective cohort study.

OBJECTIVE: To provide evidence that early electroencephalography (EEG) allows for reliable prediction of poor or good outcome after cardiac arrest. METHODS: In a 5-center prospective cohort study, we included consecutive, comatose survivors of cardiac arrest. Continuous EEG recordings were started as soon as possible and continued up to 5 days. Five-minute EEG epochs were assessed by 2 reviewers, independently, at 8 predefined time points from 6 hours to 5 days after cardiac arrest, blinded for patients' actual condition, treatment, and outcome. EEG patterns were categorized as generalized suppression (<10 µV), synchronous patterns with ≥50% suppression, continuous, or other. Outcome at 6 months was categorized as good (Cerebral Performance Category [CPC] = 1-2) or poor (CPC = 3-5). RESULTS: We included 850 patients, of whom 46% had a good outcome. Generalized suppression and synchronous patterns with ≥50% suppression predicted poor outcome without false positives at ≥6 hours after cardiac arrest. Their summed sensitivity was 0.47 (95% confidence interval [CI] = 0.42-0.51) at 12 hours and 0.30 (95% CI = 0.26-0.33) at 24 hours after cardiac arrest, with specificity of 1.00 (95% CI = 0.99-1.00) at both time points. At 36 hours or later, sensitivity for poor outcome was ≤0.22. Continuous EEG patterns at 12 hours predicted good outcome, with sensitivity of 0.50 (95% CI = 0.46-0.55) and specificity of 0.91 (95% CI = 0.88-0.93); at 24 hours or later, specificity for the prediction of good outcome was <0.90. INTERPRETATION: EEG allows for reliable prediction of poor outcome after cardiac arrest, with maximum sensitivity in the first 24 hours. Continuous EEG patterns at 12 hours after cardiac arrest are associated with good recovery. ANN NEUROL 2019;86:203-214.

Outcome Prediction in Postanoxic Coma With Deep Learning.

OBJECTIVES: Visual assessment of the electroencephalogram by experienced clinical neurophysiologists allows reliable outcome prediction of approximately half of all comatose patients after cardiac arrest. Deep neural networks hold promise to achieve similar or even better performance, being more objective and consistent. DESIGN: Prospective cohort study. SETTING: Medical ICU of five teaching hospitals in the Netherlands. PATIENTS: Eight-hundred ninety-five consecutive comatose patients after cardiac arrest. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Continuous electroencephalogram was recorded during the first 3 days after cardiac arrest. Functional outcome at 6 months was classified as good (Cerebral Performance Category 1-2) or poor (Cerebral Performance Category 3-5). We trained a convolutional neural network, with a VGG architecture (introduced by the Oxford Visual Geometry Group), to predict neurologic outcome at 12 and 24 hours after cardiac arrest using electroencephalogram epochs and outcome labels as inputs. Output of the network was the probability of good outcome. Data from two hospitals were used for training and internal validation (n = 661). Eighty percent of these data was used for training and cross-validation, the remaining 20% for independent internal validation. Data from the other three hospitals were used for external validation (n = 234). Prediction of poor outcome was most accurate at 12 hours, with a sensitivity in the external validation set of 58% (95% CI, 51-65%) at false positive rate of 0% (CI, 0-7%). Good outcome could be predicted at 12 hours with a sensitivity of 48% (CI, 45-51%) at a false positive rate of 5% (CI, 0-15%) in the external validation set. CONCLUSIONS: Deep learning of electroencephalogram signals outperforms any previously reported outcome predictor of coma after cardiac arrest, including visual electroencephalogram assessment by trained electroencephalogram experts. Our approach offers the potential for objective and real time, bedside insight in the neurologic prognosis of comatose patients after cardiac arrest.