Johannes Koren

Automatic multimodal detection for long-term seizure documentation in epilepsy

OBJECTIVE This study investigated sensitivity and false detection rate of a multimodal automatic seizure detection algorithm and the applicability to reduced electrode montages for long-term seizure documentation in epilepsy patients. METHODS An automatic seizure detection algorithm based on EEG, EMG, and ECG signals was developed. EEG/ECG recordings of 92 patients from two epilepsy monitoring units including 494 seizures were used to assess detection performance. EMG data were extracted by bandpass filtering of EEG signals. Sensitivity and false detection rate were evaluated for each signal modality and for reduced electrode montages. RESULTS All focal seizures evolving to bilateral tonic-clonic (BTCS, n=50) and 89% of focal seizures (FS, n=139) were detected. Average sensitivity in temporal lobe epilepsy (TLE) patients was 94% and 74% in extratemporal lobe epilepsy (XTLE) patients. Overall detection sensitivity was 86%. Average false detection rate was 12.8 false detections in 24h (FD/24h) for TLE and 22 FD/24h in XTLE patients. Utilization of 8 frontal and temporal electrodes reduced average sensitivity from 86% to 81%. CONCLUSION Our automatic multimodal seizure detection algorithm shows high sensitivity with full and reduced electrode montages. SIGNIFICANCE Evaluation of different signal modalities and electrode montages paces the way for semi-automatic seizure documentation systems.

Applicability of NeuroTrend as a bedside monitor in the neuro ICU

OBJECTIVE To assess whether ICU caregivers can correctly read and interpret continuous EEG (cEEG) data displayed with the computer algorithm NeuroTrend (NT) with the main attention on seizure detection and determination of sedation depth. METHODS 120 screenshots of NT (480h of cEEG) were rated by 18 briefly trained nurses and biomedical analysts. Multirater agreements (MRA) as well as interrater agreements (IRA) compared to an expert opinion (EXO) were calculated for items such as pattern type, pattern location, interruption of recording, seizure suspicion, consistency of frequency, seizure tendency and level of sedation. RESULTS MRA as well as IRA were almost perfect (80-100%) for interruption of recording, spike-and-waves, rhythmic delta activity and burst suppression. A substantial agreement (60-80%) was found for electrographic seizure patterns, periodic discharges and seizure suspicion. Except for pattern localization (70.83-92.26%), items requiring a precondition and especially those who needed interpretation like consistency of frequency (47.47-79.15%) or level of sedation (41.10%) showed lower agreements. CONCLUSIONS The present study demonstrates that NT might be a useful bedside monitor in cases of subclinical seizures. Determination of correct sedation depth by ICU caregivers requires a more detailed training. SIGNIFICANCE Computer algorithms may reduce the workload of cEEG analysis in ICU patients.

Monitoring burst suppression in critically ill patients: Multi-centric evaluation of a novel method.

OBJECTIVE To develop a computational method to detect and quantify burst suppression patterns (BSP) in the EEGs of critical care patients. A multi-center validation study was performed to assess the detection performance of the method. METHODS The fully automatic method scans the EEG for discontinuous patterns and shows detected BSP and quantitative information on a trending display in real-time. The method is designed to work without setting any patient specific parameters and to be insensitive to EEG artifacts and periodic patterns. For validation a total of 3982 h of EEG from 88 patients were analyzed from three centers. Each EEG was annotated by two reviewers to assess the detection performance and the inter-rater agreement. RESULTS Average inter-rater agreement between pairs of reviewers was κ=0.69. On average 22% of the review segments included BSP. An average sensitivity of 90% and a specificity of 84% were measured on the consensus annotations of two reviewers. More than 95% of the periodic patterns in the EEGs were correctly suppressed. CONCLUSION A fully automatic method to detect burst suppression patterns was assessed in a multi-center study. The method showed high sensitivity and specificity. SIGNIFICANCE Clinically applicable burst suppression detection method validated in a large multi-center study.

Automated Long-Term EEG Review: Fast and Precise Analysis in Critical Care Patients

Background: Ongoing or recurrent seizure activity without prominent motor features is a common burden in neurological critical care patients and people with epilepsy during ICU stays. Continuous EEG (CEEG) is the gold standard for detecting ongoing ictal EEG patterns and monitoring functional brain activity. However CEEG review is very demanding and time consuming. The purpose of the present multirater, EEG expert reviewer study, is to test and assess the clinical feasibility of an automatic EEG pattern detection method (Neurotrend). Methods: Four board certified EEG reviewers used Neurotrend to annotate 76 CEEG datasets à 6 h (in total 456 h of EEG) for rhythmic and periodic EEG patterns (RPP), unequivocal ictal EEG patterns and burst suppression. All reviewers had a predefined time limit of 5 min (± 2 min) per CEEG dataset and were compared to a predefined gold standard (conventional EEG review with unlimited time). Subanalysis of specific features of RPP was conducted as well. We used Gwets AC1 and AC2 coefficients to calculate interrater agreement (IRA) and multirater agreement (MRA). Also, we determined individual performance measures for unequivocal ictal EEG patterns and burst suppression. Bonferroni-Holmes correction for multiple testing was applied to all statistical tests. Results: Mean review time was 3.3 min (± 1.9 min) per CEEG dataset. We found substantial IRA for unequivocal ictal EEG patterns (0.61–0.79; mean sensitivity 86.8%; mean specificity 82.2%, p < 0.001) and burst suppression (0.68–0.71; mean sensitivity 96.7%; mean specificity 76.9% p < 0.001). Two reviewers showed substantial IRA for RPP (0.68–0.72), whereas the other two showed moderate agreement (0.45–0.54), compared to the gold standard (p < 0.001). MRA showed almost perfect agreement for burst suppression (0.86) and moderate agreement for RPP (0.54) and unequivocal ictal EEG patterns (0.57). Conclusions: We demonstrated the clinical feasibility of an automatic critical care EEG pattern detection method on two levels: (1) reasonable high agreement compared to the gold standard, (2) reasonable short review times compared to previously reported EEG review times with conventional EEG analysis.

Automatic ictal onset source localization in presurgical epilepsy evaluation

OBJECTIVE To test the diagnostic accuracy of a new automatic algorithm for ictal onset source localization (IOSL) during routine presurgical epilepsy evaluation following STARD (Standards for Reporting of Diagnostic Accuracy) criteria. METHODS We included 28 consecutive patients with refractory focal epilepsy (25 patients with temporal lobe epilepsy (TLE) and 3 with extratemporal epilepsy) who underwent resective epilepsy surgery. Ictal EEG patterns were analyzed with a novel automatic IOSL algorithm. IOSL source localizations on a sublobar level were validated by comparison with actual resection sites and seizure free outcome 2 years after surgery. RESULTS Sensitivity of IOSL was 92.3% (TLE: 92.3%); specificity 60% (TLE: 50%); positive predictive value 66.7% (TLE: 66.7%); and negative predictive value 90% (TLE: 85.7%). The likelihood ratio was more than ten times higher for concordant IOSL results as compared to discordant results (p = 0.013). CONCLUSIONS We demonstrated the clinical feasibility of our IOSL approach yielding reasonable high performance measures on a sublobar level. SIGNIFICANCE Our IOSL method may contribute to a correct localization of the seizure onset zone in temporal lobe epilepsy and can readily be used in standard epilepsy monitoring settings. Further studies are needed for validation in extratemporal epilepsy.

Platform Session – Evoked potentials & NIOM: Reduced electrode arrays for the automated detection of rhythmic and periodic patterns in the intensive care unit: Frequently tried, frequently failed?

Introduction To investigate the effect of systematic electrode reduction from a common 10–20 EEG system on pattern detection sensitivity (SEN). Methods Two reviewers rated 17,130 one-minute segments of 83 prospectively recorded cEEGs according to the ACNS standardized critical care EEG terminology (CCET), including burst suppression patterns (BS) and unequivocal electrographic seizures. Consensus annotations between reviewers were used as a gold standard to determine pattern detection SEN and specificity (SPE) of a computational algorithm (baseline, 19 electrodes). Electrodes were than reduced one by one in four different variations. SENs and SPEs were calculated to determine the most beneficial assembly with respect to the number and location of electrodes. Results High automated baseline SENs (84.99–93.39%) and SPEs (90.05–95.6%) were achieved for all patterns. Best overall results in detecting BS and CCET patterns were found using the “hairline + vertex” montage. While the “forehead + behind ear” montage showed an advantage in detecting ictal patterns, reaching a 15% drop of SEN with 10 electrodes, all montages could detect BS sufficiently if at least nine electrodes were available. Conclusion For the first time an automated approach was used to systematically evaluate the effect of electrode reduction on pattern detection SEN in cEEG. Prediction of the expected detection SEN of specific EEG patterns with reduced EEG montages in ICU patients.

T58. Automatic seizure detection based on multimodal signal quantification

Introduction Quantitative analysis and automated seizure detection is able to increase efficiency of EEG review. However, acceptance of software assisted review is often low because results are inaccurate in real world patient cohorts and the reason for false detections cannot be deduced. The graphical software tool encevis visualizes detections of fast rhythmic activity and patterns defined by the ACNS critical care EEG terminology. Based on these detections as well as quantitative information of EEG, ECG, and EMG a multimodal seizure detection algorithm was developed. Simple rule based classification is utilized that facilitates easy interpretability. Aim of this work was to assess detection performance of different modalities and patient groups. Methods Our computer algorithm automatically detects seizures including rhythmic EEG patterns that show an increased amplitude compared to baseline. EMG signal is extracted by bandpass filtering EEG (30–60 Hz) to measure line length (LL) for detection of sustained and excessive ictal EMG activity of generalized tonic-clonic seizures (GTCS). High absolute values of LL and an increase of 500% to baseline trigger detections. Heart rate is calculated from ECG signals to detect ictal tachycardia (ITC) with more than 100 beats per minute and an increase of over 30% compared to baseline. To assess sensitivity and false detection rate a retrospective study was conducted including EEG/ECG recordings of 92 patients from two epilepsy monitoring units. Inclusion criteria were an age above 18 and at least one recorded epileptic seizure. EEGs were used without modification or manual editing of any kind. Automatic seizure detection was calculated for all 11,978 h of EEG (min = 23 h, max = 547 h). In total 410 manual seizure annotations were compared to automatic detections to define sensitivity (SE) and false detection in 24 h (FD/24 h). Results Combination of all three seizure detection methods (EMG + ECG + EEG) resulted in SE = 88% with 10.5 FD/24 h on average. By using only EMG based detections 100% of GTCS (n = 49) were found with an average false detection rate of 3.39 FD/24 h. Seizure detection solely based on ECG yielded SE = 31% with 1.35 FD/24 h. Analysis of the temporal lobe epilepsy patients showed SE = 93.3% and 6.75 FD/24 h, the extra temporal lobe patient group resulted in SE = 80% at 15.3 FD/24 h. Conclusion We showed that automatic seizure detection based on multimodal signal quantification can reach high sensitivity. The low false detection rate results in an average of 20 false detections per week than can be validated quickly by using EEG and time synchronized quantitative screens in parallel. By visualizing quantitative information that is the source of automatic seizure detection the interpretability of results is improved. Our proposed approach to automatic EEG analysis will raise efficiency of post hoc analysis compared to the current state of the art.

ID 145 – NeuroTrend: Prospective validation of rhythmic and periodic pattern detection method for Scalp

NeuroTrend is a computational method that automatically analyses long-term scalp EEGs in the ICU according to ACNS standardized critical care EEG terminology (CCET). At present, it is a screening aid to facilitate the review process and optimize resources. A prospective multi-center study was performed in two neuro-ICUs including 68 patients who were subjected to continuous video-EEG (cEEG). Two reviewers independently annotated the first minute of each hour in the cEEG according to CCET and electrographic seizures. The matching annotations (2911 segments) were then used as gold standard condition to test sensitivity and specificity of the rhythmic and periodic pattern detection of NeuroTrend. Inter-rater agreement was substantial for main term 1 and main term 2 of the CCET. The overall detection sensitivity of NeuroTrend was 94% with high detection rates for periodic discharges (PD 81%) and rhythmic delta activity (RDA 82%). The overall specificity was 67% due to false positive detections of RDA in cases of general slowing. In contrast, for PDs a detection specificity of 88% was reached. NeuroTrend is suitable as a screening tool for cEEG in the ICU and will raise the efficiency of long-term EEG-monitoring in the ICU. Pattern differentiation between RDA and general slowing still needs improvement.