Automated Interpretation of Clinical Electroencephalograms Using Artificial Intelligence.

Importance: Electroencephalograms (EEGs) are a fundamental evaluation in neurology but require special expertise unavailable in many regions of the world. Artificial intelligence (AI) has a potential for addressing these unmet needs. Previous AI models address only limited aspects of EEG interpretation such as distinguishing abnormal from normal or identifying epileptiform activity. A comprehensive, fully automated interpretation of routine EEG based on AI suitable for clinical practice is needed. Objective: To develop and validate an AI model (Standardized Computer-based Organized Reporting of EEG-Artificial Intelligence [SCORE-AI]) with the ability to distinguish abnormal from normal EEG recordings and to classify abnormal EEG recordings into categories relevant for clinical decision-making: epileptiform-focal, epileptiform-generalized, nonepileptiform-focal, and nonepileptiform-diffuse. Design, Setting, and Participants: In this multicenter diagnostic accuracy study, a convolutional neural network model, SCORE-AI, was developed and validated using EEGs recorded between 2014 and 2020. Data were analyzed from January 17, 2022, until November 14, 2022. A total of 30 493 recordings of patients referred for EEG were included into the development data set annotated by 17 experts. Patients aged more than 3 months and not critically ill were eligible. The SCORE-AI was validated using 3 independent test data sets: a multicenter data set of 100 representative EEGs evaluated by 11 experts, a single-center data set of 9785 EEGs evaluated by 14 experts, and for benchmarking with previously published AI models, a data set of 60 EEGs with external reference standard. No patients who met eligibility criteria were excluded. Main Outcomes and Measures: Diagnostic accuracy, sensitivity, and specificity compared with the experts and the external reference standard of patients' habitual clinical episodes obtained during video-EEG recording. Results: The characteristics of the EEG data sets include development data set (N = 30 493; 14 980 men; median age, 25.3 years [95% CI, 1.3-76.2 years]), multicenter test data set (N = 100; 61 men, median age, 25.8 years [95% CI, 4.1-85.5 years]), single-center test data set (N = 9785; 5168 men; median age, 35.4 years [95% CI, 0.6-87.4 years]), and test data set with external reference standard (N = 60; 27 men; median age, 36 years [95% CI, 3-75 years]). The SCORE-AI achieved high accuracy, with an area under the receiver operating characteristic curve between 0.89 and 0.96 for the different categories of EEG abnormalities, and performance similar to human experts. Benchmarking against 3 previously published AI models was limited to comparing detection of epileptiform abnormalities. The accuracy of SCORE-AI (88.3%; 95% CI, 79.2%-94.9%) was significantly higher than the 3 previously published models (P < .001) and similar to human experts. Conclusions and Relevance: In this study, SCORE-AI achieved human expert level performance in fully automated interpretation of routine EEGs. Application of SCORE-AI may improve diagnosis and patient care in underserved areas and improve efficiency and consistency in specialized epilepsy centers.

Spike count and morphology in the classification of epileptiform discharges.

Purpose: The purpose of this study is to investigate the impact of Bergen Epileptiform Morphology Score (BEMS) and interictal epileptiform discharge (IED) candidate count in EEG classification. Methods: We included 400 consecutive patients from a clinical SCORE EEG database during 2013-2017 who had focal sharp discharges in their EEG, but no previous diagnosis of epilepsy. Three blinded EEG readers marked all IED candidates. BEMS and IED candidate counts were combined to classify EEGs as epileptiform or non-epileptiform. Diagnostic performance was assessed and then validated in an external dataset. Results: Interictal epileptiform discharge (IED) candidate count and BEMS were moderately correlated. The optimal criteria to classify an EEG as epileptiform were either one spike at BEMS > = 58, two at > = 47, or seven at > = 36. These criteria had almost perfect inter-rater reliability (Gwet's AC1 0.96), reasonable sensitivity of 56-64%, and high specificity of 98-99%. The sensitivity was 27-37%, and the specificity was 93-97% for a follow-up diagnosis of epilepsy. In the external dataset, the sensitivity for an epileptiform EEG was 60-70%, and the specificity was 90-93%. Conclusion: Quantified EEG spike morphology (BEMS) and IED candidate count can be combined to classify an EEG as epileptiform with high reliability but with lower sensitivity than regular visual EEG review.

Accurate identification of EEG recordings with interictal epileptiform discharges using a hybrid approach: Artificial intelligence supervised by human experts.

OBJECTIVE: To evaluate the diagnostic performance of artificial intelligence (AI)-based algorithms for identifying the presence of interictal epileptiform discharges (IEDs) in routine (20-min) electroencephalography (EEG) recordings. METHODS: We evaluated two approaches: a fully automated one and a hybrid approach, where three human raters applied an operational IED definition to assess the automated detections grouped into clusters by the algorithms. We used three previously developed AI algorithms: Encevis, SpikeNet, and Persyst. The diagnostic gold standard (epilepsy or not) was derived from video-EEG recordings of patients' habitual clinical episodes. We compared the algorithms with the gold standard at the recording level (epileptic or not). The independent validation data set (not used for training) consisted of 20-min EEG recordings containing sharp transients (epileptiform or not) from 60 patients: 30 with epilepsy (with a total of 340 IEDs) and 30 with nonepileptic paroxysmal events. We compared sensitivity, specificity, overall accuracy, and the review time-burden of the fully automated and hybrid approaches, with the conventional visual assessment of the whole recordings, based solely on unrestricted expert opinion. RESULTS: For all three AI algorithms, the specificity of the fully automated approach was too low for clinical implementation (16.67%; 63.33%; 3.33%), despite the high sensitivity (96.67%; 66.67%; 100.00%). Using the hybrid approach significantly increased the specificity (93.33%; 96.67%; 96.67%) with good sensitivity (93.33%; 56.67%; 76.67%). The overall accuracy of the hybrid methods (93.33%; 76.67%; 86.67%) was similar to the conventional visual assessment of the whole recordings (83.33%; 95% confidence interval [CI]: 71.48-91.70%; p > .5), yet the time-burden of review was significantly lower (p < .001). SIGNIFICANCE: The hybrid approach, where human raters apply the operational IED criteria to automated detections of AI-based algorithms, has high specificity, good sensitivity, and overall accuracy similar to conventional EEG reading, with a significantly lower time-burden. The hybrid approach is accurate and suitable for clinical implementation.

Electroclinical features and long-term therapeutic response in patients with typical absence seizures.

OBJECTIVE: To characterize in detail the electroclinical features of typical absence seizures and elucidate whether EEG or semiology features, alone or in combination, can predict long-term therapeutic outcome. METHODS: We analysed video-EEG recordings from 213 typical absence seizures from 61 patients with idiopathic generalized epilepsy. We extracted semiological features, in addition to hallmark manifestations (motor/behavioural arrest, non-responsiveness), their location, timing and frequency. We evaluated the duration and frequency of generalized spike-wave discharges and the presence of polyspikes. We used a supervised machine-learning approach (random forest) to search for classifier features for long-term therapeutic outcome (>one year). RESULTS: Besides the hallmark manifestations, additional semiological features were identified in 87% of patients (75% of seizures). The most common additional semiological features were automatisms and eye blinking (observed in 45% and 41.5% of seizures, respectively). Automatisms were associated with longer seizure duration, and oral automatisms occurred earlier compared to limb automatisms (4.03 vs. 6.19 seconds; p=0.005). The mean duration of the ictal spike-wave discharges was nine seconds, and the median frequency was 3 Hz. Polyspikes occurred in 46 seizures (21.6%), in 19 patients (31%). Median follow-up was five years, and 73% of the patients were seizure-free at the end of the follow-up. None of the semiological features, alone or in combination, were predictors of therapeutic outcome. The only significant classifier was the presence of polyspikes, predicting a non-seizure-free outcome with an accuracy of 73% (95% CI: 70-77%), positive predictive value of 92% (95% CI: 84-98%) and negative predictive value of 60% (95% CI: 39-81%). SIGNIFICANCE: Semiological features, in addition to behavioural arrest and non-responsiveness, are common in typical absence seizures, but they do not predict long-term therapeutic outcome. The presence of polyspikes has a high positive predictive value for unfavourable therapeutic outcome, and their presence should therefore be included when reporting EEGs in patients with typical absence seizures.

The operational definition of epileptiform discharges significantly improves diagnostic accuracy and inter-rater agreement of trainees in EEG reading.

To assess whether trainees can learn and implement the operational definition of interictal epileptiform discharges (IEDs) of the International Federation of Clinical Neurophysiology (IFCN), based on six morphological criteria, and whether its implementation improves their diagnostic performance and inter-rater agreement (IRA). Seven trainees evaluated a balanced dataset of 70 EEG samples containing sharp transients (35 from patients with epilepsy and 35 from patients with non-epileptic paroxysmal events). The gold standard was derived from video-EEG recordings of the habitual clinical episodes. The trainees individually reviewed the EEGs, blinded to all other data, in two successive training sessions, three months apart. The second session was preceded by a teaching module about the IFCN criteria, and the trainees implemented them during the second reading session. By implementing the IFCN criteria, trainees significantly improved their specificity (94.29% vs. 77.14%; p=0.01) and overall accuracy (81.43% vs. 64.29%; p=0.01) for identifying IEDs. Sensitivity also improved but did not reach the level of statistical significance (77.14% vs. 60%; p=0.07). IRA improved significantly from fair (k=0.31; 95% CI: 0.22-0.40) to high-moderate (k=0.56; 95% CI:0.46-0.67) beyond-chance agreement. Implementing the IFCN criteria significantly improves the diagnostic performance and IRA of trainees in identifying IEDs. Teaching the IFCN criteria for IEDs will increase specificity in clinical EEG and avoid over-reading, the most common cause of misdiagnosing epilepsy.

The influence of the abundance and morphology of epileptiform discharges on diagnostic accuracy: How many spikes you need to spot in an EEG.

OBJECTIVE: The operational definition of interictal epileptiform discharges (IEDs) of the International Federation of Clinical Neurophysiology (IFCN) described six morphological criteria. Our objective was to assess the impact of pattern-repetition in the EEG-recording, on the diagnostic accuracy of using the IFCN criteria. For clinical implementation, specificity over 95% was set as target. METHODS: Interictal EEG-recordings of 20-minutes, containing sharp-transients, from 60 patients (30 with epilepsy and 30 with non-epileptic paroxysmal events) were evaluated by three experts, who first marked IEDs solely based on expert opinion, and then, independently from the first session evaluated the presence of the IFCN criteria for each sharp-transient. The gold standard was derived from long-term video-EEG recordings of the patients habitual paroxysmal episodes. RESULTS: Presence of at least one discharge fulfilling five criteria provided a specificity of 100% (sensitivity: 70%). For discharges fulfilling fewer criteria, a higher number of discharges was needed to keep the specificity over 95% (5 discharges, when only 3 criteria were fulfilled). A sequential combination of these sets of criteria and thresholds provided a specificity of 97% and sensitivity of 80%. CONCLUSIONS: Pattern-repetition and IED morphology influence diagnostic accuracy. SIGNIFICANCE: Systematic application of these criteria will improve quality of clinical EEG interpretation.

Nutritional Status in patients with Neuro-Behçet's disease.

Background/aim: The aim of this study was to assess the nutritional status and risk factors for malnutrition in Behçet's disease (BD) and neuro-Behçet's disease (NBD) patients. Materials and methods: Medical recordings of 74 patients with BD without neurological involvement (BDWoNI), 72 patients with NBD, and 50 patients with other diseases (carpal tunnel syndrome or lumbar discopathy) were analyzed retrospectively. The serum analyses were performed in the inactive period of disease. Chronic malnutrition was defined as low levels of serum albumin (<3.5 g/dL) with normal sedimentation rate and normal serum CRP levels. Results: Six (8.3%) of the patients in NBD group, 1 (1.4%) of the patients in BDWoNI group, and none of the patients in control group had chronic malnutrition (p = 0.029). Malnutrition rate was higher in NBD than control group (p = 0.036). The mean expanded disability status scale score was 2.92 ± 3.35 (range: 0­8) in patients with malnutrition and 1.44 ± 1.76 (range: 0­9) in patients without malnutrition in NBD group (p = 0.457). The ratio of patients having a progressive disease course was 33.3% and 7.6% in patients with and without malnutrition in NBD group, respectively. The median value of the duration of neurological involvement was 2 years (0­16) in patients with malnutrition and 6.5 years (0­18) in patients without malnutrition in NBD group (p = 0.046). There was no association between malnutrition and medications, disability scores, functional system involvement or findings on brain MRI. Conclusion: Malnutrition was higher in patients with NBD. Higher disability level and progressive disease course may be risk factors for malnutrition in NBD. Malnutrition may be seen more frequently in earlier phases of neurological involvement.

Triphasic Waves Are Generated by Widespread Bilateral Cortical Networks.

PURPOSE: Triphasic waves (TWs) have been observed in the EEG recorded in patients with various types of encephalopathy, yet their genesis and significance is still debated. The aim of this study was to elucidate the localization of the cortical generators of TWs using EEG source imaging. METHODS: In 20 consecutive patients who had encephalopathy with TWs, EEG source imaging of the first negative and the positive phases of the TW was performed. Three different approaches were used: equivalent current dipoles, a distributed source model, and a recently described spatial filtration method for visualizing EEG in source space. RESULTS: Equivalent current dipole models failed to provide valid solutions. The distributed source model and the spatial filtration method suggested that TWs were generated by large, bilateral cortical networks, invariably involving the anterior frontal and the temporo-polar areas. CONCLUSIONS: Source imaging localized TWs to anterior frontal and temporo-frontal structures. Involvement of these regions is consistent with the typical pathophysiological changes of altered consciousness and cognitive changes observed in patients with TW encephalopathy.

Optimized set of criteria for defining interictal epileptiform EEG discharges.

OBJECTIVE: To find and validate the optimal combination of criteria that define interictal epileptiform EEG discharges (IEDs). Our target was a specificity over 95%, to avoid over-reading in clinical EEG. METHODS: We constructed 63 combinations of the six criteria from the operational definition of IEDs, recently issued in the EEG-glossary of the International Federation of Clinical Neurophysiology (IFCN). The diagnostic gold standard was derived from video-EEG recordings. In a testing EEG dataset from 100 patients, we selected the best performing combinations of criteria and then we validated them in an independent dataset from 70 patients. We compared their performance with subjective, expert-scorings and we determined inter-rater agreement (IRA). RESULTS: Without using criteria, the specificity of expert-scorings was lower than the pre-defined threshold (86%). The best performing combination of criteria was the following: waves with spiky morphology, followed by a slow-afterwave and voltage map suggesting a source in the brain. In the validation dataset this achieved a specificity of 97% and a sensitivity of 89%. IRA was substantial. CONCLUSIONS: The optimized set of criteria for defining IEDs has high accuracy and IRA. SIGNIFICANCE: Using these criteria will contribute to decreasing over-reading of EEG and avoid misdiagnosis of epilepsy.

Criteria for defining interictal epileptiform discharges in EEG: A clinical validation study.

OBJECTIVE: To define and validate criteria for accurate identification of EEG interictal epileptiform discharges (IEDs) using (1) the 6 sensor space criteria proposed by the International Federation of Clinical Neurophysiology (IFCN) and (2) a novel source space method. Criteria yielding high specificity are needed because EEG over-reading is a common cause of epilepsy misdiagnosis. METHODS: Seven raters reviewed EEG sharp transients from 100 patients with and without epilepsy (diagnosed definitively by video-EEG recording of habitual events). Raters reviewed the transients, randomized, and classified them as epileptiform or nonepileptiform in 3 separate rounds: in 2, EEG was reviewed in sensor space (scoring the presence/absence of each IFCN criterion for each transient or classifying unrestricted by criteria [expert scoring]); in the other, review and classification were performed in source space. RESULTS: Cutoff values of 4 and 5 criteria in sensor space and analysis in source space provided high accuracy (91%, 88%, and 90%, respectively), similar to expert scoring (92%). Two methods had specificity exceeding the desired threshold of 95%: using 5 IFCN criteria as cutoff and analysis in source space (both 95.65%); the sensitivity of these methods was 81.48% and 85.19%, respectively. CONCLUSIONS: The presence of 5 IFCN criteria in sensor space and analysis in source space are optimal for clinical implementation. By extracting these objective features, diagnostic accuracy similar to expert scorings is achieved. CLASSIFICATION OF EVIDENCE: This study provides Class III evidence that IFCN criteria in sensor space and analysis in source space have high specificity (>95%) and sensitivity (81%-85%) for identification of IEDs.

An artificial intelligence-based EEG algorithm for detection of epileptiform EEG discharges: Validation against the diagnostic gold standard.

OBJECTIVE: To validate an artificial intelligence-based computer algorithm for detection of epileptiform EEG discharges (EDs) and subsequent identification of patients with epilepsy. METHODS: We developed an algorithm for automatic detection of EDs, based on a novel deep learning method that requires a low amount of labeled EEG data for training. Detected EDs are automatically grouped into clusters, consisting of the same type of EDs, for rapid visual inspection. We validated the algorithm on an independent dataset of 100 patients with sharp transients in their EEG recordings (54 with epilepsy and 46 with non-epileptic paroxysmal events). The diagnostic gold standard was derived from the video-EEG recordings of the patients' habitual events. RESULTS: The algorithm had a sensitivity of 89% for identifying EEGs with EDs recorded from patients with epilepsy, a specificity of 70%, and an overall accuracy of 80%. CONCLUSIONS: Automated detection of EDs using an artificial intelligence-based computer algorithm had a high sensitivity. Human (expert) supervision is still necessary for confirming the clusters of detected EDs and for describing clinical correlations. Further studies on different patient populations will be needed to confirm our results. SIGNIFICANCE: The automated algorithm we describe here is a useful tool, assisting neurophysiologist in rapid assessment of EEG recordings.

Clinical characteristics and prognosis of Neuro-Behçet's disease.

OBJECTIVE: Neuro-Behçet's disease (NBD) is a rare manifestation of Behçet's disease (BD) and may cause severe disability. The aim of this study was to evaluate the treatment response in patients with NBD and to investigate the parameters that may influence the prognosis of the disease in patients with severe to mild-moderate disability. METHODS: The files of 60 patients admitted to our outpatient clinic for NBD between January 2007 and June 2014 were retrospectively reviewed. We compared the BD duration, time to NBD, NBD type and course, clinical findings of BD, functional neurological system involvement, localization of lesions on brain MRI, and all the medications between the severe and mild-moderate disability groups. RESULTS: The mean time to the onset of NBD was significantly longer (17.8±4.6 years) and the mean age was significantly higher (50.25±9.1 years) in patients with severe disability than in those with mild-moderate disability (7.5±8.0 years and 37.5±10.9 years; p=0.01 and p=0.03, respectively). Moreover, hemispheric involvement was significantly associated with severe disability (p=0.006). No difference was found with regard toother investigated parameters between the groups. CONCLUSION: We believe that severe neurological disability may be associated with older age at the onset of NBD or longer time to NBD and hemispheric lesions on brain MRI. However, our results should be cautiously evaluated with further research.

Diagnostic utility of distal nerve conduction studies and sural near-nerve needle recording in polyneuropathy.

OBJECTIVE: The electrodiagnosis of polyneuropathy (PNP) may benefit from examination using near-nerve needle technique (NNT) and from inclusion of distal nerves. This study compared the diagnostic utility of distal nerve conduction studies (NCS) and NNT recording. METHODS: Bilateral NNT and surface recording of the sural nerve and surface recording of the dorsal sural and medial plantar nerves were prospectively done in 91 patients with clinically suspected PNP. Distal NCS were additionally done in 37 healthy controls. Diagnostic reference standard was the final clinical diagnosis retrieved from the patients medical records after 1-4years. RESULTS: The clinical follow-up diagnosis confirmed PNP in 68 patients. Equally high sensitivities of the dorsal sural (72%), medial plantar (75%), and sural nerve with NNT recording (77%) were seen, while the sensitivity of conventional surface recording of the sural nerve was lower (60%). Sural NCS with both NNT and surface recording and dorsal sural NCS showed high specificities (85-95%) and positive predictive values (94-98%), while a lower specificity was seen for the medial plantar nerve (68%). CONCLUSION: NCS of distal nerves, especially the dorsal sural nerve, have high diagnostic power equalling sural NNT recording. SIGNIFICANCE: The electrodiagnostic evaluation of patients with suspected PNP benefits from NCS of distal nerves.

Cerebral Venous Thrombosis in Behçet's Disease Patients Compared to Other Causes of Cerebral Venous Thrombosis: a Retrospective Study.

OBJECTIVES: This study aims to report the outcomes of patients with Behçet's disease (BD) with cerebral venous thrombosis (CVT) due to BD compared to patients with CVT due to other causes and to discuss the treatment options. PATIENTS AND METHODS: Files of 47 patients admitted to our outpatient clinic for CVT between January 2007 and November 2014 were retrospectively reviewed. Patients were divided into two groups; group 1 included 21 CVT patients with BD (9 males, 12 females; mean age 47±12 years; range 27 to 69 years) and group 2 included 26 CVT patients without BD (11 males, 15 females; mean age 45±16 years; range 25 to 79 years). We collected data for diagnosis for BD and CVT, duration of all medications, functional system involvement, baseline Expanded Disability Status Scale scores, modified Rankin Scale scores at follow-up, and localizations of lesions in brain magnetic resonance imaging and magnetic resonance venography. RESULTS: Mean follow-up duration was four years in group 1 and two years in group 2. There was no significant difference between the groups regarding age, sex, neurological symptoms/findings and baseline Expanded Disability Status Scale scores. Localizations of occluded sinus in group 1 and group 2 were similar. For acute CVT, patients were treated with intravenous high dose prednisolone in group 1 and with anticoagulant in group 2. Follow-up modified Rankin Scale score was 0 in both groups. CONCLUSION: Our study revealed similar clinical and radiological findings in CVT patients with or without BD. Although medications used for treatment were different between the groups, treatment responses were good in both groups. Treatment with prednisolone may be sufficient and anticoagulation therapy may not be necessary for acute CVT attacks in BD.