Automated diagnosis of epilepsy using EEG power spectrum

Diagnosing Epilepsy with Normal Interictal EEG Using Dynamic Network Models

OBJECTIVE

Whereas a scalp electroencephalogram (EEG) is important for diagnosing epilepsy, a single routine EEG is limited in its diagnostic value. Only a small percentage of routine EEGs show interictal epileptiform discharges (IEDs) and overall misdiagnosis rates of epilepsy are 20% to 30%. We aim to demonstrate how network properties in EEG recordings can be used to improve the speed and accuracy differentiating epilepsy from mimics, such as functional seizures - even in the absence of IEDs.

METHODS

In this multicenter study, we analyzed routine scalp EEGs from 218 patients with suspected epilepsy and normal initial EEGs. The patients' diagnoses were later confirmed based on an epilepsy monitoring unit (EMU) admission. About 46% ultimately being diagnosed with epilepsy and 54% with non-epileptic conditions. A logistic regression model was trained using spectral and network-derived EEG features to differentiate between epilepsy and non-epilepsy. Of the 218 patients, 90% were used for training and 10% were held out for testing. Within the training set, 10-fold cross validation was performed. The resulting tool was named "EpiScalp."

RESULTS

EpiScalp achieved an area under the curve (AUC) of 0.940, an accuracy of 0.904, a sensitivity of 0.835, and a specificity of 0.963 in classifying patients as having epilepsy or not.

INTERPRETATION

EpiScalp provides an accurate diagnostic aid from a single initial EEG recording, even in more challenging epilepsy cases with normal initial EEGs. This may represent a paradigm shift in epilepsy diagnosis by deriving an objective measure of epilepsy likelihood from previously uninformative EEGs. ANN NEUROL 2025;97:907-918.

The present and future of seizure detection, prediction, and forecasting with machine learning, including the future impact on clinical trials

Seizures have a profound impact on quality of life and mortality, in part because they can be challenging both to detect and forecast. Seizure detection relies upon accurately differentiating transient neurological symptoms caused by abnormal epileptiform activity from similar symptoms with different causes. Seizure forecasting aims to identify when a person has a high or low likelihood of seizure, which is related to seizure prediction. Machine learning and artificial intelligence are data-driven techniques integrated with neurodiagnostic monitoring technologies that attempt to accomplish both of those tasks. In this narrative review, we describe both the existing software and hardware approaches for seizure detection and forecasting, as well as the concepts for how to evaluate the performance of new technologies for future application in clinical practice. These technologies include long-term monitoring both with and without electroencephalography (EEG) that report very high sensitivity as well as reduced false positive detections. In addition, we describe the implications of seizure detection and forecasting upon the evaluation of novel treatments for seizures within clinical trials. Based on these existing data, long-term seizure detection and forecasting with machine learning and artificial intelligence could fundamentally change the clinical care of people with seizures, but there are multiple validation steps necessary to rigorously demonstrate their benefits and costs, relative to the current standard.

Machine Learning and Artificial Intelligence Applications to Epilepsy: a Review for the Practicing Epileptologist

PURPOSE OF REVIEW

Machine Learning (ML) and Artificial Intelligence (AI) are data-driven techniques to translate raw data into applicable and interpretable insights that can assist in clinical decision making. Some of these tools have extremely promising initial results, earning both great excitement and creating hype. This non-technical article reviews recent developments in ML/AI in epilepsy to assist the current practicing epileptologist in understanding both the benefits and limitations of integrating ML/AI tools into their clinical practice.

RECENT FINDINGS

ML/AI tools have been developed to assist clinicians in almost every clinical decision including (1) predicting future epilepsy in people at risk, (2) detecting and monitoring for seizures, (3) differentiating epilepsy from mimics, (4) using data to improve neuroanatomic localization and lateralization, and (5) tracking and predicting response to medical and surgical treatments. We also discuss practical, ethical, and equity considerations in the development and application of ML/AI tools including chatbots based on Large Language Models (e.g., ChatGPT). ML/AI tools will change how clinical medicine is practiced, but, with rare exceptions, the transferability to other centers, effectiveness, and safety of these approaches have not yet been established rigorously. In the future, ML/AI will not replace epileptologists, but epileptologists with ML/AI will replace epileptologists without ML/AI.

Can Presurgical Interhemispheric EEG Connectivity Predict Outcome in Hemispheric Surgery? A Brain Machine Learning Approach

OBJECTIVES

Hemispherotomy (HT) is a surgical option for treatment of drug-resistant seizures due to hemispheric structural lesions. Factors affecting seizure outcome have not been fully clarified. In our study, we used a brain Machine Learning (ML) approach to evaluate the possible role of Inter-hemispheric EEG Connectivity (IC) in predicting post-surgical seizure outcome.

METHODS

We collected 21 pediatric patients with drug-resistant epilepsy; who underwent HT in our center from 2009 to 2020; with a follow-up of at least two years. We selected 5-s windows of wakefulness and sleep pre-surgical EEG and we trained Artificial Neuronal Network (ANN) to estimate epilepsy outcome. We extracted EEG features as input data and selected the ANN with best accuracy.

RESULTS

Among 21 patients, 15 (71%) were seizure and drug-free at last follow-up. ANN showed 73.3% of accuracy, with 85% of seizure free and 40% of non-seizure free patients appropriately classified.

CONCLUSIONS

The accuracy level that we reached supports the hypothesis that pre-surgical EEG features may have the potential to predict epilepsy outcome after HT.

SIGNIFICANCE

The role of pre-surgical EEG data in influencing seizure outcome after HT is still debated. We proposed a computational predictive model, with an ML approach, with a high accuracy level.

Clinical MRI morphological analysis of functional seizures compared to seizure-naïve and psychiatric controls

PURPOSE

Functional seizures (FS), also known as psychogenic nonepileptic seizures (PNES), are physical manifestations of acute or chronic psychological distress. Functional and structural neuroimaging have identified objective signs of this disorder. We evaluated whether magnetic resonance imaging (MRI) morphometry differed between patients with FS and clinically relevant comparison populations.

METHODS

Quality-screened clinical-grade MRIs were acquired from 666 patients from 2006 to 2020. Morphometric features were quantified with FreeSurfer v6. Mixed-effects linear regression compared the volume, thickness, and surface area within 201 regions-of-interest for 90 patients with FS, compared to seizure-naïve patients with depression (n = 243), anxiety (n = 68), and obsessive-compulsive disorder (OCD, n = 41), respectively, and to other seizure-naïve controls with similar quality MRIs, accounting for the influence of multiple confounds including depression and anxiety based on chart review. These comparison populations were obtained through review of clinical records plus research studies obtained on similar scanners.

RESULTS

After Bonferroni-Holm correction, patients with FS compared with seizure-naïve controls exhibited thinner bilateral superior temporal cortex (left 0.053 mm, p = 0.014; right 0.071 mm, p = 0.00006), thicker left lateral occipital cortex (0.052 mm, p = 0.0035), and greater left cerebellar white-matter volume (1085 mm³, p = 0.0065). These findings were not accounted for by lower MRI quality in patients with FS.

CONCLUSIONS

These results reinforce prior indications of structural neuroimaging correlates of FS and, in particular, distinguish brain morphology in FS from that in depression, anxiety, and OCD. Future work may entail comparisons with other psychiatric disorders including bipolar and schizophrenia, as well as exploration of brain structural heterogeneity within FS.

Abnormal phase-amplitude coupling characterizes the interictal state in epilepsy

OBJECTIVE

 Diagnosing epilepsy still requires visual interpretation of electroencephalography and magnetoencephalography (MEG) by specialists, which prevents quantification and standardization of diagnosis. Previous studies proposed automated diagnosis by combining various features from electroencephalography and MEG, such as relative power (Power) and functional connectivity. However, the usefulness of interictal phase-amplitude coupling (PAC) in diagnosing epilepsy is still unknown. We hypothesized that resting-state PAC would be different for patients with epilepsy in the interictal state and for healthy participants such that it would improve discrimination between the groups.

METHODS

 We obtained resting-state MEG and magnetic resonance imaging in 90 patients with epilepsy during their preoperative evaluation and in 90 healthy participants. We used the cortical currents estimated from MEG and magnetic resonance imaging to calculate Power in the δ (1-3 Hz), θ (4-7 Hz), α (8-13 Hz), β (13-30 Hz), low γ (35-55 Hz), and high γ (65-90 Hz) bands and functional connectivity in the θ band. PAC was evaluated using the synchronization index (SI) for eight frequency band pairs: the phases of δ, θ, α, and β and the amplitudes of low and high γ. First, we compared the mean SI values for the patients with epilepsy and the healthy participants. Then, using features such as PAC, Power, functional connectivity, and features extracted by deep learning individually or combined, we tested whether PAC improves discrimination accuracy for the two groups.

RESULTS

 The mean SI values were significantly different for the patients with epilepsy and the healthy participants. The SI value difference was highest for θ/low γ in the temporal lobe. Discrimination accuracy was the highest, at 90%, using the combination of PAC and deep learning.

SIGNIFICANCE

 Abnormal PAC characterized the patients with epilepsy in the interictal state compared with the healthy participants, potentially improving the discrimination of epilepsy.

Analysis of power spectrum and phase lag index changes following deep brain stimulation of the anterior nucleus of the thalamus in patients with drug-resistant epilepsy: A retrospective study

OBJECTIVES

The mechanisms underlying the anterior nucleus of the thalamus (ANT) deep brain stimulation (DBS) for the treatment of drug-resistant epilepsy (DRE) have not been fully explored. The present study aimed to measure the changes in whole-brain activity generated by ANT DBS using interictal electroencephalography (EEG).

MATERIALS AND METHODS

Interictal EEG signals were retrospectively collected in 20 DRE patients who underwent ANT DBS surgery. Patients were classified as responders or non-responders depending on their response to ANT DBS treatment. The power spectrum (PS) and Phase Lag Index (PLI) were determined and data analyzed using a paired sample t-test to evaluate activity differences between pre-and-post-treatment on different frequency categories. Student's t-test, Mann-Whitney test (non-parametric test) and Fisher exact test were used to compare groups in terms of clinical variables and EEG metrics. P values < 0.05 were considered statistically significant, and FDR-corrected values were used for multiple testing.

RESULTS

PS analysis revealed that whole-brain spectral power had a significant decrease in the beta (p = 0.005) and gamma (p = 0.037) bands following ANT DBS treatment in responders. The analysis of scalp topographic images of all patients showed that ANT DBS decreases PS in the beta band at the F3, F7 and Cz electrode sites. The findings indicated a decrease in PS in the gamma band at the Fp2, F3, Cz, T3, T5 and Oz electrode sites. After ANT DBS treatment, PLI analysis showed a significant decrease in PLI between Fp1 and T3 in the gamma band in responders.

CONCLUSION

The findings showed that ANT DBS induces a decrease in power in the left frontal lobe, left temporal lobe and midline areas in the beta and gamma bands. Lower whole-brain power in the beta and gamma bands can be used as biomarkers for a favorable therapeutic response to ANT DBS, and decreased synchronization between the left frontal pole and temporal lobe in the gamma band can also be used as a biomarker for effective clinical stimulation to guide postoperative programming.

Neural Decoding of EEG Signals with Machine Learning: A Systematic Review

Electroencephalography (EEG) is a non-invasive technique used to record the brain's evoked and induced electrical activity from the scalp. Artificial intelligence, particularly machine learning (ML) and deep learning (DL) algorithms, are increasingly being applied to EEG data for pattern analysis, group membership classification, and brain-computer interface purposes. This study aimed to systematically review recent advances in ML and DL supervised models for decoding and classifying EEG signals. Moreover, this article provides a comprehensive review of the state-of-the-art techniques used for EEG signal preprocessing and feature extraction. To this end, several academic databases were searched to explore relevant studies from the year 2000 to the present. Our results showed that the application of ML and DL in both mental workload and motor imagery tasks has received substantial attention in recent years. A total of 75% of DL studies applied convolutional neural networks with various learning algorithms, and 36% of ML studies achieved competitive accuracy by using a support vector machine algorithm. Wavelet transform was found to be the most common feature extraction method used for all types of tasks. We further examined the specific feature extraction methods and end classifier recommendations discovered in this systematic review.

A minority of patients with functional seizures have abnormalities on neuroimaging

OBJECTIVE

Functional seizures often are managed incorrectly as a diagnosis of exclusion. However, a significant minority of patients with functional seizures may have abnormalities on neuroimaging that typically are associated with epilepsy, leading to diagnostic confusion. We evaluated the rate of epilepsy-associated findings on MRI, FDG-PET, and CT in patients with functional seizures.

METHODS

We studied radiologists' reports from neuroimages at our comprehensive epilepsy center from a consecutive series of patients diagnosed with functional seizures without comorbid epilepsy from 2006 to 2019. We summarized the MRI, FDG-PET, and CT results as follows: within normal limits, incidental findings, unrelated findings, non-specific abnormalities, post-operative study, epilepsy risk factors (ERF), borderline epilepsy-associated findings (EAF), and definitive EAF.

RESULTS

Of the 256 MRIs, 23% demonstrated ERF (5%), borderline EAF (8%), or definitive EAF (10%). The most common EAF was hippocampal sclerosis, with the majority of borderline EAF comprising hippocampal atrophy without T2 hyperintensity or vice versa. Of the 87 FDG-PETs, 26% demonstrated borderline EAF (17%) or definitive EAF (8%). Epilepsy-associated findings primarily included focal hypometabolism, especially of the temporal lobes, with borderline findings including subtle or questionable hypometabolism. Of the 51 CTs, only 2% had definitive EAF.

SIGNIFICANCE

This large case series provides further evidence that, while uncommon, EAF are seen in patients with functional seizures. A significant portion of these abnormal findings are borderline. The moderately high rate of these abnormalities may represent framing bias from the indication of the study being "seizures," the relative subtlety of EAF, or effects of antiseizure medications.

Epilepsy, dissociative seizures, and mixed: Associations with time to video-EEG

PURPOSE

Video-electroencephalographic monitoring (VEM) is a core component to the diagnosis and evaluation of epilepsy and dissociative seizures (DS)-also known as functional or psychogenic seizures-but VEM evaluation often occurs later than recommended. To understand why delays occur, we compared how patient-reported clinical factors were associated with time from first seizure to VEM (TVEM) in patients with epilepsy, DS or mixed.

METHODS

We acquired data from 1245 consecutive patients with epilepsy, VEM-documented DS or mixed epilepsy and DS. We used multivariate log-normal regression with recursive feature elimination (RFE) to evaluate which of 76 clinical factors interacting with patients' diagnoses were associated with TVEM.

RESULTS

The mean and median TVEM were 14.6 years and 10 years, respectively (IQR 3-23 years). In the multivariate RFE model, the factors associated with longer TVEM in all patients included unemployment and not student status, more antiseizure medications (current and past), concussion, and ictal behavior suggestive of temporal lobe epilepsy. Average TVEM was shorter for DS than epilepsy, particularly for patients with depression, anxiety, migraines, and eye closure. Average TVEM was longer specifically for patients with DS taking more medications, more seizure types, non-metastatic cancer, and with other psychiatric comorbidities.

CONCLUSIONS

In all patients with seizures, trials of numerous antiseizure medications, unemployment and non-student status was associated with longer TVEM. These associations highlight a disconnect between International League Against Epilepsy practice parameters and observed referral patterns in epilepsy. In patients with dissociative seizures, some but not all factors classically associated with DS reduced TVEM.

Validation of a predictive calculator to distinguish between patients presenting with dissociative versus epileptic seizures

Dissociative seizures (also known as psychogenic nonepileptic seizures) are a common functional neurological disorder that can be difficult to distinguish from epileptic seizures. Patients with dissociative seizures provide diagnostic challenges, leading to delays in care, inappropriate care, and significant healthcare utilization and associated costs. The dissociative seizure likelihood score (DSLS) was developed by Kerr and colleagues at UCLA to distinguish between patients with epileptic seizures and dissociative seizures based on clinical and medication history as well as features of seizure semiology. We validated this calculator at the University of Colorado, which is a Level 4 National Association of Epilepsy Center. The DSLS accurately predicted the diagnosis in 81% of patients, despite local variability in the factors associated with epileptic versus dissociative seizures between the two populations. The DSLS can be a useful tool to assist with history taking and may have important utility for clinical decision making with these difficult to distinguish patient populations.

Factors associated with delay to video-EEG in dissociative seizures

PURPOSE

While certain clinical factors suggest a diagnosis of dissociative seizures (DS), otherwise known as functional or psychogenic nonepileptic seizures (PNES), ictal video-electroencephalography monitoring (VEM) is the gold standard for diagnosis. Diagnostic delays were associated with worse quality of life and more seizures, even after treatment. To understand why diagnoses were delayed, we evaluated which factors were associated with delay to VEM.

METHODS

Using data from 341 consecutive patients with VEM-documented dissociative seizures, we used multivariate log-normal regression with recursive feature elimination (RFE) and multiple imputation of some missing data to evaluate which of 76 clinical factors were associated with time from first dissociative seizure to VEM.

RESULTS

The mean delay to VEM was 8.4 years (median 3 years, IQR 1-10 years). In the RFE multivariate model, the factors associated with longer delay to VEM included more past antiseizure medications (0.19 log-years/medication, standard error (SE) 0.05), more medications for other medical conditions (0.06 log-years/medication, SE 0.03), history of physical abuse (0.75 log-years, SE 0.27), and more seizure types (0.36 log-years/type, SE 0.11). Factors associated with shorter delay included active employment or student status (-1.05 log-years, SE 0.21) and higher seizure frequency (0.14 log-years/log[seizure/month], SE 0.06).

CONCLUSIONS

Patients with greater medical and seizure complexity had longer delays. Delays in multiple domains of healthcare can be common for victims of physical abuse. Unemployed and non-student patients may have had more barriers to access VEM. These results support earlier referral of complex cases to a comprehensive epilepsy center.

Objective score from initial interview identifies patients with probable dissociative seizures

OBJECTIVE

To develop a Dissociative Seizures Likelihood Score (DSLS), which is a comprehensive, evidence-based tool using information available during the first outpatient visit to identify patients with "probable" dissociative seizures (DS) to allow early triage to more extensive diagnostic assessment.

METHODS

Based on data from 1616 patients with video-electroencephalography (vEEG) confirmed diagnoses, we compared the clinical history from a single neurology interview of patients in five mutually exclusive groups: epileptic seizures (ES), DS, physiologic nonepileptic seizure-like events (PSLE), mixed DS plus ES, and inconclusive monitoring. We used data-driven methods to determine the diagnostic utility of 76 features from retrospective chart review and applied this model to prospective interviews.

RESULTS

The DSLS using recursive feature elimination (RFE) correctly identified 77% (95% confidence interval (CI), 74-80%) of prospective patients with either ES or DS, with a sensitivity of 74% and specificity of 84%. This accuracy was not significantly inferior than neurologists' impression (84%, 95% CI: 80-88%) and the kappa between neurologists' and the DSLS was 21% (95% CI: 1-41%). Only 3% of patients with DS were missed by both the fellows and our score (95% CI 0-11%).

SIGNIFICANCE

The evidence-based DSLS establishes one method to reliably identify some patients with probable DS using clinical history. The DSLS supports and does not replace clinical decision making. While not all patients with DS can be identified by clinical history alone, these methods combined with clinical judgement could be used to identify patients who warrant further diagnostic assessment at a comprehensive epilepsy center.

Bring on the Machines: Could Machine Learning Improve the Quality of Patient Education Materials? A Systematic Search and Rapid Review

PURPOSE

Clear and trustworthy information is essential for people who are ill. People with cancer, in particular, are targeted with vast quantities of patient education material, but of variable quality. Machine learning technologies are popular across industries for automated tasks, like analyzing language and spotting readability issues. With the experience of patients with cancer in mind, we reviewed whether anyone has proposed, modeled, or applied machine learning technologies for the assessment of patient education materials and explored the utility of this application.

METHODS

We systematically searched the literature to identify English-language articles published in peer-reviewed journals or as conference abstracts that proposed, used, or modeled the use of machine learning technology to assess patient education materials. Specifically, we searched MEDLINE, Web of Science, CINAHL, and Compendex. Two reviewers assessed study eligibility and performed study screening.

RESULTS

We identified 1,570 publications in our search after duplicate removal. After screening, we included five projects (detailed in nine articles) that proposed, modeled, or used machine learning technology to assess the quality of patient education materials. We evaluated the utility of each application across four domains: multidimensionality (2 of 5 applications), patient centeredness (1 of 5 applications), customizability (0 of 5 applications), and development stage (theoretical, 1 of 5 applications; in development, 3 of 5 applications; complete and available, 1 of 5 applications). Combining points across each domain, the mean utlity score across included projects was 1.8 of 5 possible points.

CONCLUSION

Given its potential, machine learning has not yet been leveraged substantially in the assessment of patient education materials. We propose machine learning systems that can dynamically identify problematic language and content by assessing the quality of patient education materials across a range of flexible, customizable criteria. Assessment may help patients and families decide which materials to use and encourage developers to improve materials overall.

An objective score to identify psychogenic seizures based on age of onset and history

OBJECTIVE

Psychogenic nonepileptic seizure (PNES) is a common diagnosis after evaluation of medication resistant or atypical seizures with video-electroencephalographic monitoring (VEM), but usually follows a long delay after the development of seizures, during which patients are treated for epilepsy. Therefore, more readily available diagnostic tools are needed for earlier identification of patients at risk for PNES. A tool based on patient-reported psychosocial history would be especially beneficial because it could be implemented in the outpatient clinic.

METHODS

Based on the data from 1375 patients with VEM-confirmed diagnoses, we used logistic regression to compare the frequency of specific patient-reported historical events, demographic information, age of onset, and delay from first seizure until VEM in five mutually exclusive groups of patients: epileptic seizures (ES), PNES, physiologic nonepileptic seizure-like events (PSLE), mixed PNES plus ES, and inconclusive monitoring. To determine the diagnostic utility of this information to differentiate PNES only from ES only, we used multivariate piecewise-linear logistic regression trained using retrospective data from chart review and validated based on data from 246 prospective standardized interviews.

RESULTS

The prospective area under the curve of our weighted multivariate piecewise-linear by-sex score was 73%, with the threshold that maximized overall retrospective accuracy resulting in a prospective sensitivity of 74% (95% CI: 70-79%) and prospective specificity of 71% (95% CI: 64-82%). The linear model and piecewise linear without an interaction term for sex had very similar performance statistics. In the multivariate piecewise-linear sex-split predictive model, the significant factors positively associated with ES were history of febrile seizures, current employment or active student status, history of traumatic brain injury (TBI), and longer delay from first seizure until VEM. The significant factors associated with PNES were female sex, older age of onset, mild TBI, and significant stressful events with sexual abuse, in particular, increasing the likelihood of PNES. Delays longer than 20years, age of onset after 31years for men, and age of onset after 40years for women had no additional effect on the likelihood of PNES.

DISCUSSION

Our promising results suggest that an objective score has the potential to serve as an early outpatient screening tool to identify patients with greater likelihood of PNES when considered in combination with other factors. In addition, our analysis suggests that sexual abuse, more than other psychological stressors including physical abuse, is more associated with PNES. There was a trend of increasing frequency of PNES for women during childbearing years and plateauing outside those years that was not observed in men.

Identifying psychogenic seizures through comorbidities and medication history

OBJECTIVE

Low-cost evidence-based tools are needed to facilitate the early identification of patients with possible psychogenic nonepileptic seizures (PNES). Prior to accurate diagnosis, patients with PNES do not receive interventions that address the cause of their seizures and therefore incur high medical costs and disability due to an uncontrolled seizure disorder. Both seizures and comorbidities may contribute to this high cost.

METHODS

Based on data from 1,365 adult patients with video-electroencephalography-confirmed diagnoses from a single center, we used logistic and Poisson regression to compare the total number of comorbidities, number of medications, and presence of specific comorbidities in five mutually exclusive groups of diagnoses: epileptic seizures (ES) only, PNES only, mixed PNES and ES, physiologic nonepileptic seizurelike events, and inconclusive monitoring. To determine the diagnostic utility of comorbid diagnoses and medication history to differentiate PNES only from ES only, we used multivariate logistic regression, controlling for sex and age, trained using a retrospective database and validated using a prospective database.

RESULTS

Our model differentiated PNES only from ES only with a prospective accuracy of 78% (95% confidence interval =72-84%) and area under the curve of 79%. With a few exceptions, the number of comorbidities and medications was more predictive than a specific comorbidity. Comorbidities associated with PNES were asthma, chronic pain, and migraines (p < 0.01). Comorbidities associated with ES were diabetes mellitus and nonmetastatic neoplasm (p < 0.01). The population-level analysis suggested that patients with mixed PNES and ES may be a population distinct from patients with either condition alone.

SIGNIFICANCE

An accurate patient-reported medical history and medication history can be useful when screening for possible PNES. Our prospectively validated and objective score may assist in the interpretation of the medication and medical history in the context of the seizure description and history.

Diagnostic implications of review-of-systems questionnaires to differentiate epileptic seizures from psychogenic seizures

OBJECTIVE

Early and accurate diagnosis of patients with psychogenic nonepileptic seizures (PNES) leads to appropriate treatment and improves long-term seizure prognosis. However, this is complicated by the need to record seizures to make a definitive diagnosis. Suspicion for PNES can be raised through knowledge that patients with PNES have increased somatic sensitivity and report more positive complaints on review-of-systems questionnaires (RoSQs) than patients with epileptic seizures. If the responses on the RoSQ can differentiate PNES from other seizure types, then these forms could be an early screening tool.

METHODS

Our dataset included all patients admitted from January 2006 to June 2016 for video-electroencephalography at UCLA. RoSQs prior to May 2015 were acquired through retrospective chart review (n=405), whereas RoSQs from subsequent patients were acquired prospectively (n=190). Controlling for sex and number of comorbidities, we used binomial regression to compare the total number of symptoms and the frequency of specific symptoms between five mutually exclusive groups of patients: epileptic seizures (ES), PNES, physiologic nonepileptic seizure-like events (PSLE), mixed PNES plus ES, and inconclusive monitoring. To determine the diagnostic utility of RoSQs to differentiate PNES only from ES only, we used multivariate logistic regression, controlling for sex and the number of medical comorbidities.

RESULTS

On average, patients with PNES or mixed PNES and ES reported more than twice as many symptoms than patients with isolated ES or PSLE (p<0.001). The prospective accuracy to differentiate PNES from ES was not significantly higher than naïve assumption that all patients had ES (76% vs 70%, p>0.1).

DISCUSSION

This analysis of RoSQs confirms that patients with PNES with and without comorbid ES report more symptoms on a population level than patients with epilepsy or PSLE. While these differences help describe the population of patients with PNES, the consistency of RoSQ responses was neither accurate nor specific enough to be used solely as an early screening tool for PNES. Our results suggest that the RoSQ may help differentiate PNES from ES only when, based on other information, the pre-test probability of PNES is at least 50%.

Diagnostic delay in psychogenic seizures and the association with anti-seizure medication trials

PURPOSE

The average delay from first seizure to diagnosis of psychogenic non-epileptic seizures (PNES) is over 7 years. The reason for this delay is not well understood. We hypothesized that a perceived decrease in seizure frequency after starting an anti-seizure medication (ASM) may contribute to longer delays, but the frequency of such a response has not been well established.

METHODS

Time from onset to diagnosis, medication history and associated seizure frequency was acquired from the medical records of 297 consecutive patients with PNES diagnosed using video-electroencephalographic monitoring. Exponential regression was used to model the effect of medication trials and response on diagnostic delay.

RESULTS

Mean diagnostic delay was 8.4 years (min 1 day, max 52 years). The robust average diagnostic delay was 2.8 years (95% CI: 2.2-3.5 years) based on an exponential model as 10 to the mean of log10 delay. Each ASM trial increased the robust average delay exponentially by at least one third of a year (Wald t=3.6, p=0.004). Response to ASM trials did not significantly change diagnostic delay (Wald t=-0.9, p=0.38).

CONCLUSION

Although a response to ASMs was observed commonly in these patients with PNES, the presence of a response was not associated with longer time until definitive diagnosis. Instead, the number of ASMs tried was associated with a longer delay until diagnosis, suggesting that ASM trials were continued despite lack of response. These data support the guideline that patients with seizures should be referred to epilepsy care centers after failure of two medication trials.

Interval analysis of interictal EEG: pathology of the alpha rhythm in focal epilepsy

The contemporary use of interictal scalp electroencephalography (EEG) in the context of focal epilepsy workup relies on the visual identification of interictal epileptiform discharges. The high-specificity performance of this marker comes, however, at a cost of only moderate sensitivity. Zero-crossing interval analysis is an alternative to Fourier analysis for the assessment of the rhythmic component of EEG signals. We applied this method to standard EEG recordings of 78 patients divided into 4 subgroups: temporal lobe epilepsy (TLE), frontal lobe epilepsy (FLE), psychogenic nonepileptic seizures (PNES) and nonepileptic patients with headache. Interval-analysis based markers were capable of effectively discriminating patients with epilepsy from those in control subgroups (AUC~0.8) with diagnostic sensitivity potentially exceeding that of visual analysis. The identified putative epilepsy-specific markers were sensitive to the properties of the alpha rhythm and displayed weak or non-significant dependences on the number of antiepileptic drugs (AEDs) taken by the patients. Significant AED-related effects were concentrated in the theta interval range and an associated marker allowed for identification of patients on AED polytherapy (AUC~0.9). Interval analysis may thus, in perspective, increase the diagnostic yield of interictal scalp EEG. Our findings point to the possible existence of alpha rhythm abnormalities in patients with epilepsy.

Multimodal data and machine learning for surgery outcome prediction in complicated cases of mesial temporal lobe epilepsy

BACKGROUND

This study sought to predict postsurgical seizure freedom from pre-operative diagnostic test results and clinical information using a rapid automated approach, based on supervised learning methods in patients with drug-resistant focal seizures suspected to begin in temporal lobe.

METHOD

We applied machine learning, specifically a combination of mutual information-based feature selection and supervised learning classifiers on multimodal data, to predict surgery outcome retrospectively in 20 presurgical patients (13 female; mean age±SD, in years 33±9.7 for females, and 35.3±9.4 for males) who were diagnosed with mesial temporal lobe epilepsy (MTLE) and subsequently underwent standard anteromesial temporal lobectomy. The main advantage of the present work over previous studies is the inclusion of the extent of ipsilateral neocortical gray matter atrophy and spatiotemporal properties of depth electrode-recorded seizures as training features for individual patient surgery planning.

RESULTS

A maximum relevance minimum redundancy (mRMR) feature selector identified the following features as the most informative predictors of postsurgical seizure freedom in this study's sample of patients: family history of epilepsy, ictal EEG onset pattern (positive correlation with seizure freedom), MRI-based gray matter thickness reduction in the hemisphere ipsilateral to seizure onset, proportion of seizures that first appeared in ipsilateral amygdala to total seizures, age, epilepsy duration, delay in the spread of ipsilateral ictal discharges from site of onset, gender, and number of electrode contacts at seizure onset (negative correlation with seizure freedom). Using these features in combination with a least square support vector machine (LS-SVM) classifier compared to other commonly used classifiers resulted in very high surgical outcome prediction accuracy (95%).

CONCLUSIONS

Supervised machine learning using multimodal compared to unimodal data accurately predicted postsurgical outcome in patients with atypical MTLE.

Multimodal diagnosis of epilepsy using conditional dependence and multiple imputation

The definitive diagnosis of the type of epilepsy, if it exists, in medication-resistant seizure disorder is based on the efficient combination of clinical information, long-term video-electroencephalography (EEG) and neuroimaging. Diagnoses are reached by a consensus panel that combines these diverse modalities using clinical wisdom and experience. Here we compare two methods of multimodal computer-aided diagnosis, vector concatenation (VC) and conditional dependence (CD), using clinical archive data from 645 patients with medication-resistant seizure disorder, confirmed by video-EEG. CD models the clinical decision process, whereas VC allows for statistical modeling of cross-modality interactions. Due to the nature of clinical data, not all information was available in all patients. To overcome this, we multiply-imputed the missing data. Using a C4.5 decision tree, single modality classifiers achieved 53.1%, 51.5% and 51.1% average accuracy for MRI, clinical information and FDG-PET, respectively, for the discrimination between non-epileptic seizures, temporal lobe epilepsy, other focal epilepsies and generalized-onset epilepsy (vs. chance, p<0.01). Using VC, the average accuracy was significantly lower (39.2%). In contrast, the CD classifier that classified with MRI then clinical information achieved an average accuracy of 58.7% (vs. VC, p<0.01). The decrease in accuracy of VC compared to the MRI classifier illustrates how the addition of more informative features does not improve performance monotonically. The superiority of conditional dependence over vector concatenation suggests that the structure imposed by conditional dependence improved our ability to model the underlying diagnostic trends in the multimodality data.

Balancing Clinical and Pathologic Relevence in the Machine Learning Diagnosis of Epilepsy

The application of machine learning to epilepsy can be used both to develop clinically useful computer-aided diagnostic tools, and to reveal pathologically relevant insights into the disease. Such studies most frequently use neurologically normal patients as the control group to maximize the pathologic insight yielded from the model. This practice yields potentially inflated accuracy because the groups are quite dissimilar. A few manuscripts, however, opt to mimic the clinical comparison of epilepsy to non-epileptic seizures, an approach we believe to be more clinically realistic. In this manuscript, we describe the relative merits of each control group. We demonstrate that in our clinical quality FDG-PET database the performance achieved was similar using each control group. Based on these results, we find that the choice of control group likely does not hinder the reported performance. We argue that clinically applicable computer-aided diagnostic tools for epilepsy must directly address the clinical challenge of distinguishing patients with epilepsy from those with non-epileptic seizures.

Computer-Aided Diagnosis and Localization of Lateralized Temporal Lobe Epilepsy Using Interictal FDG-PET

Interictal FDG-PET (iPET) is a core tool for localizing the epileptogenic focus, potentially before structural MRI, that does not require rare and transient epileptiform discharges or seizures on EEG. The visual interpretation of iPET is challenging and requires years of epilepsy-specific expertise. We have developed an automated computer-aided diagnostic (CAD) tool that has the potential to work both independent of and synergistically with expert analysis. Our tool operates on distributed metabolic changes across the whole brain measured by iPET to both diagnose and lateralize temporal lobe epilepsy (TLE). When diagnosing left TLE (LTLE) or right TLE (RTLE) vs. non-epileptic seizures (NES), our accuracy in reproducing the results of the gold standard long term video-EEG monitoring was 82% [95% confidence interval (CI) 69-90%] or 88% (95% CI 76-94%), respectively. The classifier that both diagnosed and lateralized the disease had overall accuracy of 76% (95% CI 66-84%), where 89% (95% CI 77-96%) of patients correctly identified with epilepsy were correctly lateralized. When identifying LTLE, our CAD tool utilized metabolic changes across the entire brain. By contrast, only temporal regions and the right frontal lobe cortex, were needed to identify RTLE accurately, a finding consistent with clinical observations and indicative of a potential pathophysiological difference between RTLE and LTLE. The goal of CADs is to complement - not replace - expert analysis. In our dataset, the accuracy of manual analysis (MA) of iPET (∼80%) was similar to CAD. The square correlation between our CAD tool and MA, however, was only 30%, indicating that our CAD tool does not recreate MA. The addition of clinical information to our CAD, however, did not substantively change performance. These results suggest that automated analysis might provide clinically valuable information to focus treatment more effectively.