Epilepsy diagnosis using a clinical decision tool and artificially intelligent electroencephalography.

OBJECTIVE: To construct a tool for non-experts to calculate the probability of epilepsy based on easily obtained clinical information combined with an artificial intelligence readout of the electroencephalogram (AI-EEG). MATERIALS AND METHODS: We performed a chart review of 205 consecutive patients aged 18 years or older who underwent routine EEG. We created a point system to calculate the pre-EEG probability of epilepsy in a pilot study cohort. We also computed a post-test probability based on AI-EEG results. RESULTS: One hundred and four (50.7%) patients were female, the mean age was 46 years, and 110 (53.7%) were diagnosed with epilepsy. Findings favoring epilepsy included developmental delay (12.6% vs 1.1%), prior neurological injury (51.4% vs 30.9%), childhood febrile seizures (4.6% vs 0.0%), postictal confusion (43.6% vs 20.0%), and witnessed convulsions (63.6% vs 21.1%); findings favoring alternative diagnoses were lightheadedness (3.6% vs 15.8%) or onset after prolonged sitting or standing (0.9% vs 7.4%). The final point system included 6 predictors: Presyncope (-3 points), cardiac history (-1), convulsion or forced head turn (+3), neurological disease history (+2), multiple prior spells (+1), postictal confusion (+2). Total scores of ≤1 point predicted <5% probability of epilepsy, while cumulative scores ≥7 predicted >95%. The model showed excellent discrimination (AUROC: 0.86). A positive AI-EEG substantially increases the probability of epilepsy. The impact is greatest when the pre-EEG probability is near 30%. SIGNIFICANCE: A decision tool using a small number of historical clinical features accurately predicts the probability of epilepsy. In indeterminate cases, AI-assisted EEG helps resolve uncertainty. This tool holds promise for use by healthcare workers without specialty epilepsy training if validated in an independent cohort.

Predicting neurosurgical outcomes in focal epilepsy patients using computational modelling.

SEE EISSA AND SCHEVON DOI101093/AWW332 FOR A SCIENTIFIC COMMENTARY ON THIS ARTICLE: Surgery can be a last resort for patients with intractable, medically refractory epilepsy. For many of these patients, however, there is substantial risk that the surgery will be ineffective. The prediction of who is likely to benefit from a surgical approach is crucial for being able to inform patients better, conduct principled prospective clinical trials, and ultimately tailor therapeutic approaches to these patients more effectively. Dynamical computational models, informed with patient data, can be used to make predictions and give mechanistic insight. In this study, we develop patient-specific dynamical network models of epileptogenic cortex. We infer the network connectivity matrix from non-seizure electrographic recordings of patients and use these connectivity matrices as the network structure in our model. The model simulates the dynamics of a bi-stable switch at every node in this network, meaning that every node starts in a background state, but has the ability to transit to a co-existing seizure state. Whether a transition happens in a node is partly determined by the stochastic nature of the input to the node, but also by the input the node receives from other connected nodes in the network. By conducting simulations with such a model, we can detect the average transition time for nodes in a given network, and therefore define nodes with a short transition time as highly epileptogenic. In a retrospective study, we found that in some patients the regions with high epileptogenicity in the model overlap with those identified clinically as the seizure onset zone. Moreover, it was found that the resection of these regions in the model reduces the overall likelihood of a seizure. Following removal of these regions in the model, we predicted surgical outcomes and compared these to actual patient outcomes. Our predictions were found to be 81.3% accurate on a dataset of 16 patients with intractable epilepsy. Intriguingly, in patients with unsuccessful outcomes, the proposed computational approach is able to suggest alternative resection sites. The model presented here gives mechanistic insight as to why surgery may be unsuccessful in some patients. This may aid clinicians in presurgical evaluation by providing a tool to explore various surgical options, offering complementary information to existing clinical techniques.

Modulation of EEG functional connectivity networks in subjects undergoing repetitive transcranial magnetic stimulation.

Transcranial magnetic stimulation (TMS) is a noninvasive brain stimulation technique that utilizes magnetic fluxes to alter cortical activity. Continuous theta-burst repetitive TMS (cTBS) results in long-lasting decreases in indices of cortical excitability, and alterations in performance of behavioral tasks. We investigated the effects of cTBS on cortical function via functional connectivity and graph theoretical analysis of EEG data. Thirty-one channel resting-state EEG recordings were obtained before and after 40 s of cTBS stimulation to the left primary motor cortex. Functional connectivity between nodes was assessed in multiple frequency bands using lagged max-covariance, and subsequently thresholded to construct undirected graphs. After cTBS, we find widespread decreases in functional connectivity in the alpha band. There are also simultaneous increases in functional connectivity in the high-beta bands, especially amongst anterior and interhemispheric connections. The analysis of the undirected graphs reveals that interhemispheric and interregional connections are more likely to be modulated after cTBS than local connections. There is also a shift in the topology of network connectivity, with an increase in the clustering coefficient after cTBS in the beta bands, and a decrease in clustering and increase in path length in the alpha band, with the alpha-band connectivity primarily decreased near the site of stimulation. cTBS produces widespread alterations in cortical functional connectivity, with resulting shifts in cortical network topology.

Increased task-relevant fMRI responsiveness in comatose cardiac arrest patients is associated with improved neurologic outcomes.

Early prediction of the recovery of consciousness in comatose cardiac arrest patients remains challenging. We prospectively studied task-relevant fMRI responses in 19 comatose cardiac arrest patients and five healthy controls to assess the fMRI's utility for neuroprognostication. Tasks involved instrumental music listening, forward and backward language listening, and motor imagery. Task-specific reference images were created from group-level fMRI responses from the healthy controls. Dice scores measured the overlap of individual subject-level fMRI responses with the reference images. Task-relevant responsiveness index (Rindex) was calculated as the maximum Dice score across the four tasks. Correlation analyses showed that increased Dice scores were significantly associated with arousal recovery (P < 0.05) and emergence from the minimally conscious state (EMCS) by one year (P < 0.001) for all tasks except motor imagery. Greater Rindex was significantly correlated with improved arousal recovery (P = 0.002) and consciousness (P = 0.001). For patients who survived to discharge (n = 6), the Rindex's sensitivity was 75% for predicting EMCS (n = 4). Task-based fMRI holds promise for detecting covert consciousness in comatose cardiac arrest patients, but further studies are needed to confirm these findings. Caution is necessary when interpreting the absence of task-relevant fMRI responses as a surrogate for inevitable poor neurological prognosis.

A quantitative approach to evaluating interictal epileptiform discharges based on interpretable quantitative criteria.

OBJECTIVE: To provide quantitative measures of the six International Federation of Clinical Neurophysiology (IFCN) criteria for interictal epileptiform discharge (IED) identification and estimate the likelihood of a candidate IED being epileptiform. METHODS: We designed an algorithm to identify five fiducial landmarks (onset, peak, trough, slow-wave peak, offset) of a candidate IED, and from these to quantify the six IFCN features of IEDs. Another model was trained with these features to quantify the probability that the waveform is epileptiform and incorporated into a user-friendly interface. RESULTS: The model's performance is excellent (area under the receiver operating characteristic curve (AUROC) = 0.88; calibration error 0.03) but lower than human experts (receiver operating characteristic (ROC) curve is below experts' operating points) or a deep neural-network model (SpikeNet; AUCROC = 0.97; calibration error 0.04). The six features were all significant (p<0.001), but not equally important when determining potential epileptiform nature of candidate IEDs: waveform asymmetry was the most (coefficient 0.64) and duration the least discriminative (coefficient 0.09). CONCLUSIONS: Our approach quantifies the six IFCN criteria for IED identification and combines them in an easily interpretable, accessible fashion that accurately captures the likelihood that a candidate waveform is epileptiform. SIGNIFICANCE: This model may assist clinical electroencephalographers decide whether candidate waveforms are epileptiform and may assist trainees learn to identify IEDs.

COVID-19-induced headache in Boston and the vicinity.

Headache is a common neurological symptom of Coronavirus disease 2019 (COVID-19) patients. However, the prevalence, comorbidities, and ethnic susceptibilities of COVID-19-induced headaches are not well-defined. We performed a retrospective chart review of patients who tested positive for SARS-CoV2 by reverse transcriptase-polymerase chain reaction (RT-PCR) in March and April 2020 at Massachusetts General Hospital, Boston, Massachusetts, USA. In the study, we identified 450 patients, 202 (44.9%) male, and 248 (55.1%) female, who tested positive for COVID-19. Headache is a significant painful symptom affecting 26% of patients. Female predominance is determined in sore throat, nasal congestion, hypogeusia, headache, and ear pain. In contrast, pneumonia and inpatient hospitalization were more prevalent in males. Younger patients (< 50) were more likely to develop sore throat, fatigue, anosmia, hypogeusia, ear pain, myalgia /arthralgia, and headache. In contrast, older (> 50) patients were prone to develop pneumonia and required hospitalization. Ethnic subgroup analysis suggests Hispanic patients were prone to headaches, nausea/vomiting, nasal congestion, fever, fatigue, anosmia, and myalgia/arthralgia compared to non-Hispanics. Headache risk factors include nausea/vomiting, sore throat, nasal congestion, fever, cough, fatigue, anosmia, hypogeusia, dizziness, ear pain, eye pain, and myalgia/arthralgia. Our study demonstrates regional gender, age, and ethnic variabilities in COVID symptomatology in Boston and the vicinity. It identifies mild viral, painful, and neurological symptoms are positive predictors of headache development in COVID-19.

Brain health scores to predict neurological outcomes from electronic health records.

BACKGROUND: Preserving brain health is a critical priority in primary care, yet screening for these risk factors in face-to-face primary care visits is challenging to scale to large populations. We aimed to develop automated brain health risk scores calculated from data in the electronic health record (EHR) enabling population-wide brain health screening in advance of patient care visits. METHODS: This retrospective cohort study included patients with visits to an outpatient neurology clinic at Massachusetts General Hospital, between January 2010 and March 2021. Survival analysis with an 11-year follow-up period was performed to predict the risk of intracranial hemorrhage, ischemic stroke, depression, death and composite outcome of dementia, Alzheimer's disease, and mild cognitive impairment. Variables included age, sex, vital signs, laboratory values, employment status and social covariates pertaining to marital, tobacco and alcohol status. Random sampling was performed to create a training (70%) set for hyperparameter tuning in internal 5-fold cross validation and an external hold-out testing (30%) set of patients, both stratified by age. Risk ratios for high and low risk groups were evaluated in the hold-out test set, using 1000 bootstrapping iterations to calculate 95% confidence intervals (CI). RESULTS: The cohort comprised 17,040 patients with an average age of 49 ± 15.6 years; majority were males (57 %), White (78 %) and non-Hispanic (80 %). The low and high groups average risk ratios [95 % CI] were: intracranial hemorrhage 0.46 [0.45-0.48] and 2.07 [1.95-2.20], ischemic stroke 0.57 [0.57-0.59] and 1.64 [1.52-1.69], depression 0.68 [0.39-0.74] and 1.29 [0.78-1.38], composite of dementia 0.27 [0.26-0.28] and 3.52 [3.18-3.81] and death 0.24 [0.24-0.24] and 3.96 [3.91-4.00]. CONCLUSIONS: Simple risk scores derived from routinely collected EHR accurately quantify the risk of developing common neurologic and psychiatric diseases. These scores can be computed automatically, prior to medical care visits, and may thus be useful for large-scale brain health screening.

On-demand EEG education through competition - A novel, app-based approach to learning to identify interictal epileptiform discharges.

Objective: Misinterpretation of EEGs harms patients, yet few resources exist to help trainees practice interpreting EEGs. We therefore sought to evaluate a novel educational tool to teach trainees how to identify interictal epileptiform discharges (IEDs) on EEG. Methods: We created a public EEG test within the iOS app DiagnosUs using a pool of 13,262 candidate IEDs. Users were shown a candidate IED on EEG and asked to rate it as epileptiform (IED) or not (non-IED). They were given immediate feedback based on a gold standard. Learning was analyzed using a parametric model. We additionally analyzed IED features that best correlated with expert ratings. Results: Our analysis included 901 participants. Users achieved a mean improvement of 13% over 1,000 questions and an ending accuracy of 81%. Users and experts appeared to rely on a similar set of IED morphologic features when analyzing candidate IEDs. We additionally identified particular types of candidate EEGs that remained challenging for most users even after substantial practice. Conclusions: Users improved in their ability to properly classify candidate IEDs through repeated exposure and immediate feedback. Significance: This app-based learning activity has great potential to be an effective supplemental tool to teach neurology trainees how to accurately identify IEDs on EEG.

SCRIB: Set-Classifier with Class-Specific Risk Bounds for Blackbox Models.

Despite deep learning (DL) success in classification problems, DL classifiers do not provide a sound mechanism to decide when to refrain from predicting. Recent works tried to control the overall prediction risk with classification with rejection options. However, existing works overlook the different significance of different classes. We introduce Set-classifier with Class-specific RIsk Bounds (SCRIB) to tackle this problem, assigning multiple labels to each example. Given the output of a black-box model on the validation set, SCRIB constructs a set-classifier that controls the class-specific prediction risks. The key idea is to reject when the set classifier returns more than one label. We validated SCRIB on several medical applications, including sleep staging on electroencephalogram (EEG) data, X-ray COVID image classification, and atrial fibrillation detection based on electrocardiogram (ECG) data. SCRIB obtained desirable class-specific risks, which are 35%-88% closer to the target risks than baseline methods.

Automated EEG-based prediction of delayed cerebral ischemia after subarachnoid hemorrhage.

OBJECTIVE: Delayed cerebral ischemia (DCI) is a leading complication of aneurysmal subarachnoid hemorrhage (SAH) and electroencephalography (EEG) is increasingly used to evaluate DCI risk. Our goal is to develop an automated DCI prediction algorithm integrating multiple EEG features over time. METHODS: We assess 113 moderate to severe grade SAH patients to develop a machine learning model that predicts DCI risk using multiple EEG features. RESULTS: Multiple EEG features discriminate between DCI and non-DCI patients when aligned either to SAH time or to DCI onset. DCI and non-DCI patients have significant differences in alpha-delta ratio (0.08 vs 0.05, p < 0.05) and percent alpha variability (0.06 vs 0.04, p < 0.05), Shannon entropy (p < 0.05) and epileptiform discharge burden (205 vs 91 discharges per hour, p < 0.05) based on whole brain and vascular territory averaging. Our model improves predictions by emphasizing the most informative features at a given time with an area under the receiver-operator curve of 0.73, by day 5 after SAH and good calibration between 48-72 hours (calibration error 0.13). CONCLUSIONS: Our proposed model obtains good performance in DCI prediction. SIGNIFICANCE: We leverage machine learning to enable rapid, automated, multi-featured EEG assessment and has the potential to increase the utility of EEG for DCI prediction.

Rapid annotation of seizures and interictal-ictal-injury continuum EEG patterns.

BACKGROUND: Manual annotation of seizures and interictal-ictal-injury continuum (IIIC) patterns in continuous EEG (cEEG) recorded from critically ill patients is a time-intensive process for clinicians and researchers. In this study, we evaluated the accuracy and efficiency of an automated clustering method to accelerate expert annotation of cEEG. NEW METHOD: We learned a local dictionary from 97 ICU patients by applying k-medoids clustering to 592 features in the time and frequency domains. We utilized changepoint detection (CPD) to segment the cEEG recordings. We then computed a bag-of-words (BoW) representation for each segment. We further clustered the segments by affinity propagation. EEG experts scored the resulting clusters for each patient by labeling only the cluster medoids. We trained a random forest classifier to assess validity of the clusters. RESULTS: Mean pairwise agreement of 62.6% using this automated method was not significantly different from interrater agreements using manual labeling (63.8%), demonstrating the validity of the method. We also found that it takes experts using our method 5.31 ±â€¯4.44 min to label the 30.19 ±â€¯3.84 h of cEEG data, more than 45 times faster than unaided manual review, demonstrating efficiency. COMPARISON WITH EXISTING METHODS: Previous studies of EEG data labeling have generally yielded similar human expert interrater agreements, and lower agreements with automated methods. CONCLUSIONS: Our results suggest that long EEG recordings can be rapidly annotated by experts many times faster than unaided manual review through the use of an advanced clustering method.

Deep active learning for Interictal Ictal Injury Continuum EEG patterns.

OBJECTIVES: Seizures and seizure-like electroencephalography (EEG) patterns, collectively referred to as "ictal interictal injury continuum" (IIIC) patterns, are commonly encountered in critically ill patients. Automated detection is important for patient care and to enable research. However, training accurate detectors requires a large labeled dataset. Active Learning (AL) may help select informative examples to label, but the optimal AL approach remains unclear. METHODS: We assembled >200,000 h of EEG from 1,454 hospitalized patients. From these, we collected 9,808 labeled and 120,000 unlabeled 10-second EEG segments. Labels included 6 IIIC patterns. In each AL iteration, a Dense-Net Convolutional Neural Network (CNN) learned vector representations for EEG segments using available labels, which were used to create a 2D embedding map. Nearest-neighbor label spreading within the embedding map was used to create additional pseudo-labeled data. A second Dense-Net was trained using real- and pseudo-labels. We evaluated several strategies for selecting candidate points for experts to label next. Finally, we compared two methods for class balancing within queries: standard balanced-based querying (SBBQ), and high confidence spread-based balanced querying (HCSBBQ). RESULTS: Our results show: 1) Label spreading increased convergence speed for AL. 2) All query criteria produced similar results to random sampling. 3) HCSBBQ query balancing performed best. Using label spreading and HCSBBQ query balancing, we were able to train models approaching expert-level performance across all pattern categories after obtaining ∼7000 expert labels. CONCLUSION: Our results provide guidance regarding the use of AL to efficiently label large EEG datasets in critically ill patients.

A standardized nomenclature for spectrogram EEG patterns: Inter-rater agreement and correspondence with common intensive care unit EEG patterns.

OBJECTIVE: To determine the inter-rater agreement (IRA) of a standardized nomenclature for EEG spectrogram patterns, and to estimate the probability distribution of ictal-interictal continuum (IIC) patterns vs. other EEG patterns within each category in this nomenclature. METHODS: We defined seven spectrogram categories: "Solid Flames", "Irregular Flames", "Broadband-monotonous", "Narrowband-monotonous", "Stripes", "Low power", and "Artifact". Ten electroencephalographers scored 115 spectrograms and the corresponding raw EEG samples. Gwet's agreement coefficient was used to calculate IRA. RESULTS: Solid Flames represented seizures or IIC patterns 69.4% of the time. Irregular Flames represented seizures or IIC patterns 38.7% of the time. Broadband-monotonous primarily corresponded with seizures or IIC (54.3%) and Narrowband-monotonous with focal or generalized slowing (43.8%). Stripes were associated with burst-suppression (37.2%) and generalized suppression (34.4%). Low Power category was associated with generalized suppression (94%). There was "near perfect" agreement for Solid Flames (κ = 94.36), Low power (κ = 92.61), and Artifact (κ = 93.72). There was "substantial agreement" for all other categories (κ = 74.65-79.49). CONCLUSIONS: This EEG spectrogram nomenclature has high IRA among electroencephalographers. SIGNIFICANCE: The nomenclature can be a useful tool for EEG screening. Future studies are needed to determine if using this nomenclature shortens time to IIC identification, and how best to use it in practice to reduce time to intervention.

Resident training and interrater agreements using the ACNS critical care EEG terminology.

PURPOSE: Electroencephalography (EEG) remains the gold standard for identifying rhythmic and periodic patterns in critically ill patients. Residents have frequent exposures to EEG and critically ill patients during their training. Our study aimed to assess resident competency in the use of the American Clinical Neurophysiology Society (ACNS) critical care EEG terminology. METHODS: After self-guided reading and a 2-hour session reviewing the ACNS critical care EEG Terminology training slides, 16 adult neurology residents (PGY 2-4) completed the ACNS certification test. Performance scores were reported as average percent agreement (PA%) with a previously established 5-member expert panel. Interrater agreement was calculated to gauge consensus among peers within the resident cohort. Self-reported comfort levels using the terminology were also obtained. RESULTS: The overall pass rate for our cohort was 50% and the median score was 74%. The terms with the highest PA% were: seizures (86.4%), main term 1 (78%), main term 2 (74%). Interrater agreement scores (kappa values) were almost perfect for seizure, and substantial for main terms 1 and 2. CONCLUSIONS: Our data suggests that with minimal investment, adult neurology residents at various stages of training can effectively learn the ACNS critical care EEG Terminology.

Real-time segmentation of burst suppression patterns in critical care EEG monitoring.

OBJECTIVE: Develop a real-time algorithm to automatically discriminate suppressions from non-suppressions (bursts) in electroencephalograms of critically ill adult patients. METHODS: A real-time method for segmenting adult ICU EEG data into bursts and suppressions is presented based on thresholding local voltage variance. Results are validated against manual segmentations by two experienced human electroencephalographers. We compare inter-rater agreement between manual EEG segmentations by experts with inter-rater agreement between human vs automatic segmentations, and investigate the robustness of segmentation quality to variations in algorithm parameter settings. We further compare the results of using these segmentations as input for calculating the burst suppression probability (BSP), a continuous measure of depth-of-suppression. RESULTS: Automated segmentation was comparable to manual segmentation, i.e. algorithm-vs-human agreement was comparable to human-vs-human agreement, as judged by comparing raw EEG segmentations or the derived BSP signals. Results were robust to modest variations in algorithm parameter settings. CONCLUSIONS: Our automated method satisfactorily segments burst suppression data across a wide range adult ICU EEG patterns. Performance is comparable to or exceeds that of manual segmentation by human electroencephalographers. SIGNIFICANCE: Automated segmentation of burst suppression EEG patterns is an essential component of quantitative brain activity monitoring in critically ill and anesthetized adults. The segmentations produced by our algorithm provide a basis for accurate tracking of suppression depth.