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.

Determining the Spike-Wave Index Using Automated Detection Software.

PURPOSE: The spike-wave index (SWI) is a key feature in the diagnosis of electrical status epilepticus during slow-wave sleep. Estimating the SWI manually is time-consuming and is subject to interrater and intrarater variability. Use of automated detection software would save time. Thereby, this software will consistently detect a certain EEG phenomenon as epileptiform and is not influenced by human factors. To determine noninferiority in calculating the SWI, we compared the performance of a commercially available spike detection algorithm (P13 software, Persyst Development Corporation, San Diego, CA) with human expert consensus. METHODS: The authors identified all prolonged EEG recordings for the diagnosis or follow-up of electrical status epilepticus during slow-wave sleep carried out from January to December 2018 at an epilepsy tertiary referral center. The SWI during the first 10 minutes of sleep was estimated by consensus of two human experts. This was compared with the SWI calculated by the automated spike detection algorithm using the three available sensitivity settings: "low," "medium," and "high." In the software, these sensitivity settings are denoted as perception values. RESULTS: Forty-eight EEG recordings from 44 individuals were analyzed. The SWIs estimated by human experts did not differ from the SWIs calculated by the automated spike detection algorithm in the "low" perception mode (P = 0.67). The SWIs calculated in the "medium" and "high" perception settings were, however, significantly higher than the human expert estimated SWIs (both P < 0.001). CONCLUSIONS: Automated spike detection (P13) is a useful tool in determining SWI, especially when using the "low" sensitivity setting. Using such automated detection tools may save time, especially when reviewing larger epochs.

Using sampled visual EEG review in combination with automated detection software at the EMU.

PURPOSE: Complete visual review of prolonged video-EEG recordings at an EMU (Epilepsy Monitoring Unit) is time consuming and can cause problems in times of paucity of educated personnel. In this study we aimed to show non inferiority for electroclinical diagnosis using sampled review in combination with EEG analysis softreferware (P13 software, Persyst Corporation), in comparison to complete visual review. METHOD: Fifty prolonged video-EEG recordings in adults were prospectively evaluated using sampled visual EEG review in combination with automated detection software of the complete EEG record. Visually assessed samples consisted of one hour during wakefulness, one hour during sleep, half an hour of wakefulness after wake-up and all clinical events marked by the individual and/or nurses. The final electro-clinical diagnosis of this new review approach was compared with the electro-clinical diagnosis after complete visual review as presently used. RESULTS: The electro-clinical diagnosis based on sampled visual review combined with automated detection software did not differ from the diagnosis based on complete visual review. Furthermore, the detection software was able to detect all records containing epileptiform abnormalities and epileptic seizures. CONCLUSION: Sampled visual review in combination with automated detection using Persyst 13 is non-inferior to complete visual review for electroclinical diagnosis of prolonged video-EEG at an EMU setting, which makes this approach promising.

Epilepsy monitoring units can be safe places; a prospective study in a large cohort.

OBJECTIVE: No international guideline is available for minimum safety measures at epilepsy monitoring units (EMUs), although recommendations for preferred practices exist. These are mostly based on expert opinion, without evidence of effectiveness. We do not apply all of these preferred practices at our EMU setting. We audited adverse events and diagnostic utility at our EMU over one year. METHODS: From May 2018 to May 2019, we prospectively collected data concerning adverse events and diagnostic utility of all EMU admissions (noninvasive video-electroencephalogram (EEG) recordings); during these admissions, individuals can be ambulant within their EMU room. RESULTS: There were 1062 admissions comprising 1518 EMU days. In 2% of the admissions, a complication occurred, mostly a fall without injury (n = 6). In almost half of the falls, this was from the bed. Complications occurred most often during admissions for presurgical evaluation. Antiseizure medication (ASM) was tapered in 86% of presurgical cases, but no serious injury occurred, and occurring seizures were effectively treated with intranasal midazolam if needed. CONCLUSIONS: The overall adverse event rate was low. Falls are the most common adverse event comparable with previously published fall rates at other EMUs where people are restricted to their bed. We showed that restricted ambulation at a well-monitored EMU is not necessary and possibly unwanted. No serious injury due to tapering of ASM occurred, and intranasal midazolam was shown to be effective as acute seizure treatment.

Improving staff response to seizures on the epilepsy monitoring unit with online EEG seizure detection algorithms.

OBJECTIVE: User safety and the quality of diagnostics on the epilepsy monitoring unit (EMU) depend on reaction to seizures. Online seizure detection might improve this. While good sensitivity and specificity is reported, the added value above staff response is unclear. We ascertained the added value of two electroencephalograph (EEG) seizure detection algorithms in terms of additional detected seizures or faster detection time. METHODS: EEG-video seizure recordings of people admitted to an EMU over one year were included, with a maximum of two seizures per subject. All recordings were retrospectively analyzed using Encevis EpiScan and BESA Epilepsy. Detection sensitivity and latency of the algorithms were compared to staff responses. False positive rates were estimated on 30 uninterrupted recordings (roughly 24 h per subject) of consecutive subjects admitted to the EMU. RESULTS: EEG-video recordings used included 188 seizures. The response rate of staff was 67%, of Encevis 67%, and of BESA Epilepsy 65%. Of the 62 seizures missed by staff, 66% were recognized by Encevis and 39% by BESA Epilepsy. The median latency was 31 s (staff), 10 s (Encevis), and 14 s (BESA Epilepsy). After correcting for walking time from the observation room to the subject, both algorithms detected faster than staff in 65% of detected seizures. The full recordings included 617 h of EEG. Encevis had a median false positive rate of 4.9 per 24 h and BESA Epilepsy of 2.1 per 24 h. CONCLUSIONS: EEG-video seizure detection algorithms may improve reaction to seizures by improving the total number of seizures detected and the speed of detection. The false positive rate is feasible for use in a clinical situation. Implementation of these algorithms might result in faster diagnostic testing and better observation during seizures.

Standardized computer-based organized reporting of EEG: SCORE - Second version.

Standardized terminology for computer-based assessment and reporting of EEG has been previously developed in Europe. The International Federation of Clinical Neurophysiology established a taskforce in 2013 to develop this further, and to reach international consensus. This work resulted in the second, revised version of SCORE (Standardized Computer-based Organized Reporting of EEG), which is presented in this paper. The revised terminology was implemented in a software package (SCORE EEG), which was tested in clinical practice on 12,160 EEG recordings. Standardized terms implemented in SCORE are used to report the features of clinical relevance, extracted while assessing the EEGs. Selection of the terms is context sensitive: initial choices determine the subsequently presented sets of additional choices. This process automatically generates a report and feeds these features into a database. In the end, the diagnostic significance is scored, using a standardized list of terms. SCORE has specific modules for scoring seizures (including seizure semiology and ictal EEG patterns), neonatal recordings (including features specific for this age group), and for Critical Care EEG Terminology. SCORE is a useful clinical tool, with potential impact on clinical care, quality assurance, data-sharing, research and education.

Timing of first event in inpatient long-term video-EEG monitoring for diagnostic purposes.

BACKGROUND: Long-term video-EEG monitoring (LTM) aims to record the habitual event and is a useful diagnostic tool for neurological paroxysmal clinical events. In our epilepsy monitoring unit (EMU) setting, admissions are usually planned to last up to five days. We ascertained time taken for the recording of a first event and determined correlations between different clinical characteristics and timings. METHODS: We retrospectively reviewed diagnostic and classification LTM recording performed at a tertiary epilepsy centre. RESULTS: Sixty-three recordings were reviewed. Most subjects (89%) had events at least once a week prior to admission. In 40 (63%) a habitual event was recorded, mostly (93%) within the first two days. No events were recorded on day four or five. A few characteristics were associated with a trend for events occurring earlier (events more than once a week vs less than once a week, motor symptoms compared with aura or dyscognitive events, and reduction of antiepileptic drugs versus no reduction). CONCLUSIONS: Our finding suggests that, for diagnostic event recording in people with epilepsy or PNEA, a maximum recording time of three days is sufficient in two thirds of them, if event frequency is at least once a week. In the remaining third, prolonged recording up to five days did not result in capturing a clinical event. For these individuals, shorter admission could be planned, for example for 2days rather than 5days.

Value of video monitoring for nocturnal seizure detection in a residential setting.

OBJECTIVE: Following a sudden death at a residential care unit, the Dutch Health and Care Inspectorate advised intensification of the use of video monitoring (VM) at the unit. We assessed whether VM resulted in increased identification of seizures that required clinical intervention. METHODS: The unit provides care for 340 individuals with refractory epilepsy and severe learning disabilities. Acoustic detection systems (ADSs) cover all individuals; 37 people also have a bed motion sensor (BMS) and 46 people with possible nocturnal seizures are now monitored by VM. During a 6-month period, in all cases of a suspected seizure we asked the caregivers to specify which device alerted them and to indicate whether this led to an intervention. Staff costs of VM were estimated using payroll information. RESULTS: We identified 1,208 seizures in 37 individuals: 4 had no nocturnal seizures and 393 (33%) seizures were seen only on video. In 169 (14%) of 1,208 seizures an intervention was made and this included 39 (10%) of 393 seizures seen only on video. When compared to seizures observed with an ADS or BMS, seizures seen only on video were more often tonic seizures (71% vs. 22%, p < 0.001) and occurred mostly in the beginning or at the end of the night (40% vs. 26%, p < 0.001). The extra staff costs of monitoring was 7,035 euro per seizure seen only on video and leading to an intervention. SIGNIFICANCE: VM facilitates nocturnal surveillance, but the costs are high. This underscores the need for development of reliable seizure detection devices.

A 12-year follow-up in sporadic inclusion body myositis: an end stage with major disabilities.

Sporadic inclusion body myositis is considered to be a slowly progressive myopathy. Long-term follow-up data are, however, not yet available. Follow-up data are important with a view to informing patients about their prognosis and selecting appropriate outcome measures for clinical trials. We performed a follow-up study of 64 patients with sporadic inclusion body myositis who participated in a national epidemiological study in the Netherlands. Case histories were recorded, and manual and quantitative muscle tests as well as laboratory tests were performed at baseline and 12 years (median) after the first out-patient visit. Date and cause of death were recorded for all deceased patients. Forty-six patients died during the follow-up period, two patients chose not to participate and one patient was lost to follow-up. The remaining 15 surviving patients had a mean disease duration of 20 years and were clinically evaluated at the second time point. The mean decline in strength was 3.5 and 5.4% per year according to the manual muscle testing and quantitative muscle testing, respectively. This decline was most pronounced in the lower legs, which were also the weakest extremities. Life expectancy was normal at 81 years, but activities of daily life were clearly restricted. At follow-up, all patients were found to be using a wheelchair, seven of them (47%) being completely wheelchair-bound. Disorders of the respiratory system were the most common cause of death. In three patients, euthanasia was requested and in another three, continuous deep sedation was applied. The fact that end-of-life care interventions were used in six patients (13%) reflects the severe disability and loss of quality of life at the end stage of this disease. Sporadic inclusion body myositis is a chronic progressive disorder, leading to major disabilities at the end stage of the disease due to extensive muscle weakness.

Magnetic resonance imaging of skeletal muscles in sporadic inclusion body myositis.

OBJECTIVE: To analyse whether MRI of upper and lower extremity muscles in a large patient group with sporadic IBM (sIBM) is of additional value in the diagnostic work-up of sIBM. METHODS: Thirty-two sIBM patients were included. Magnetic resonance (MR) parameters evaluated in 68 muscles of upper and lower extremity were muscle atrophy, fatty infiltration and inflammation. These findings were correlated with disease duration, weakness and serum creatine kinase (sCK) levels. RESULTS: Fatty infiltration was far more common than inflammation. Muscles most frequently infiltrated with fat were the flexor digitorum profundus (FDP), anterior muscles of the upper leg and all muscles of the lower leg, preferentially the medial part of the gastrocnemius. The rectus femoris was relatively spared compared with other quadriceps muscles as well as the adductors of the upper leg. Inflammation was common in general, but individually sparse, present in 78% of the patients with a median of two inflamed muscles per patient. A statistically significant correlation was found between the amount of fatty infiltration and disease severity, disease duration and sCK. CONCLUSION: We provide a detailed description of the MRI in sIBM and show a distinct pattern of muscle involvement. Relatively severe affliction of the medial compartment of the gastrocnemius, combined with relative sparing of the rectus femoris or involvement of the FDP can be indicative of sIBM. MRI can contribute to the diagnosis in selected patients with clear clinical suspicion, but lacking the mandatory set of muscle biopsy features.

The heart in sporadic inclusion body myositis: a study in 51 patients.

The purpose of this study was to explore the prevalence and nature of cardiac abnormalities in sporadic inclusion body myositis (sIBM). Fifty-one sIBM patients were cross-sectionally studied using history-taking, physical examination, measurements of serum creatine kinase activity, the MB fraction (CK-MB), cardiac troponin T (cTnT) and I (cTnI), a 12-lead electrocardiogram (ECG) and 2-dimensional echocardiography. Present cardiac history was abnormal in 12 (24%) out of 51 patients, 12 (24%) patients had abnormalities on ECG, mostly aspecific, and in 12 (24%) patients the echocardiograph showed abnormalities. Elevated CK-MB was present in 42 (82%) patients and 40 (78%) had an elevated cTnT in the absence of acute cardiac pathology. In contrast, in one patient (2%) cTnI was elevated. There was no apparent association between elevated biomarkers, ECG or echocardiographic abnormalities. The prevalence of cardiac abnormalities in sIBM does not seem to be higher than would be expected in these elderly patients. Elevated CK-MB and cTnT levels are common, in contrast to cTnI, but do not reflect cardiac pathology.