SUPPLEMENTAL MATERIAL

Spectrogram Screening of Adult Electroencephalograms is Sensitive and Efficient:

Lidia M.V.R. Moura, MD; Mouhsin M. Shafi, MD, PhD; Marcus Ng, MD; Sandipan Pati, MD; Sydney S. Cash, MD, PhD; Andrew J. Cole, MD; Daniel Brian Hoch, MD, PhD; Eric S. Rosenthal, MD; M. Brandon Westover, MD, PhD.

A. CSA-guided and conventional cEEG review protocols

A variety of “quantitative EEG” trending techniques are available as part of commercial cEEG review software packages, including measures of “seizure probability”, alpha-delta ratio, asymmetry, “amplitude-integrated EEG”, and rhythmicity. We opted for CSA (spectrograms) because this display involves a straightforward, nonproprietary transformation of the underlying data, and as a single modality provides data in a simple format compatible with rapid visual review. For additional details of the implementation of the CSA analysis, see sections A and B below.

1. CSA guided cEEG review protocol (Figures e-1 and e-2): At each cEEG workstation, readers had three computer screens where CSA displays and EEG raw data could be reviewed simultaneously. cEEG analysis began with reviewing the reported findings from the first 30 minutes of monitoring,to simulate the standard clinical practice of first performing a detailed review and reporting on 20-60 minutes of EEG data at the outset of monitoring. This initial review provides context in which the ensuing cEEG data is interpreted. CSA displays were used to guide subsequent review. Clicking anywhere within the CSA display automatically displayed a standard 10-second view of the corresponding segment of EEG on a separate monitor, allowing rapid “back-and-forth” inspection of the CSA and raw cEEG data. Reviewers were allowed to page back and forth around the raw EEG data segment displayed to them, up to a maximum of 1 minute in either direction. The location of each viewed EEG segment was recorded by placing a cursor on the review screen. During the review process, the fellows were able to make annotations and flag identified seizures on the raw EEG data. Review times for CSA-guided review of each record were recorded. In this way electrographic seizures and other cEEG patterns of interest were marked. After cEEG analysis was completed a 15-item form was used to report (a) the number of seizures, (b) presence or absence of the following patterns: periodic epileptiform discharges (PEDs), epileptiform discharges (EDs), focal slowing (FS), generalized slowing (GS) and rhythmic delta activity (RDA). PEDs and RDA were defined according to standard ACNS criteria 1, and electrographic seizures were defined using standard criteria2. These patterns were selected as the most critical elements in daily clinical of cEEG reporting. Interpreters using-CSA guided review were blinded to the results of the gold standard interpretation (conventional cEEG review by readers D and E described below), and vice versa.

2. Conventional cEEG review protocol: The same 118 cEEGs were independently reviewed by the same readers for the CSA-protocol (readers A-C). Readerswere assigned different cEEGs than the ones they reviewed under the CSA-guided protocol to ensure that they were not familiar with the recorings. Standard protocol at our institution is for every cEEG recording to be visually reviewed page-by-page in its entirety by a fellow in training, and then a more focused review is conducted by an attending level neurophysiologist. In the present study, to measure the time taken for “conventional review” we employed fellows (readers A-C), the same readers whose times were measured for CSA-guided review.

A second independent round of conventional review was performed by two experienced attending-physician EEGers (readers D and E), for the purpose of establishing a “gold standard” for findings contained in each EEG recording (e.g. number of seizures, presence of periodic discharges, etc). Readers D and E did not record their review times, as the emphasis of their review was on detecting all findings with high confidence, rather than efficiency. Any questions about EEG pattern classification were resolved by discussion of readers D and E until reaching consensus.

We emphasize that ‘conventional review’ in this study refers in either case to aneurophysiologist looking at the at the entire recording without the aid of CSA or any other automated computer analysis, as opposed to reviewing only timed samples or events identified and labeled by a technologist, physician, or any other assistant.

Note that the estimated time reported for conventional CSA review was not the time taken by the attending neurophysiologists to review the EEGs. Rather, to provide a relevant comparison, we estimated the time required for conventional visual analysis of the same EEGs (without CSA-guidance) by EEG fellows (readers A-C). Findings of readers D and E were used only as a gold standard against which to assess the sensitivity of readers A-C for detecting critical EEG patterns.

B. Technical Details of EEG recording and CSA Displays:

cEEG data was recorded using 19 silver chloride electrodes, affixed to the scalp according to the International 10-20 System. CSA displays (i.e. spectrograms) were computed by the Fast Fourier Transform (FFT) algorithm available in Magic Marker clinical software (Persyst, San Diego, CA). Epoch sizes for FFT calculations were 2 seconds duration. The cEEG sampling rate was 512 Hz. CSA displays were configured to show the average spectrogram on a square-root microvolt scale displayed as a colormap (Figure 1). Two hour segments of CSA power data were displayed from top-to-bottom in a montage based on the following electrode groups: “left lateral” (Fp1-F7, F7-T3, T3-T5, T5-O1),“left parasagittal” (Fp1-F3,F3-C3,C3-P3,P3-O1), “right lateral” (Fp2-F8,F8-T4,T4-T6, T6-O2), “right parasagittal” (Fp2-F4,F3-C4,C4-P4,P4-O2) and “relative asymmetry index” (difference between spectrograms for corresponding left and right EEG channels). For all CSA displays, the y-axis represented frequency (0 to 20Hz) and the x-axis represented time. For the first four displays the z-axis represented power as a range of colors with black representing low power and blue, green, orange, pink, and white representing successively higher power. The z-axis of the relative asymmetry index represented the power difference between left and right hemispheres at each frequency, with darker red corresponding to increased right-sided power and darker blue corresponding to increased left-sided power.

C. Analysis of Differences in Detection Performance Among CSA-guided cEEGReviwers

The minor seizure detection rate disparities between neurophysiologists A-C using CSA-guided review are detailed in Table e-3.Reader A had the highest detection rates for seizures (91.6%) and epileptiform discharges (96.5%). One missed seizure consisted of focal rhythmic delta activity at 1-2 Hz with admixed epileptiform discharges. (The reviewer marked this event but misclassified it as a PED pattern.) Reviewer A missed seizures in only 5 EEGs, in which the proportions missed were 2/54, 1/28, 1/4, 1/53 and 16/111.

Reader B showed the lowest detection rate for epileptiform discharges at 57.8%. On subsequent questioning, this neurophysiologist reported ignoring triphasic waves and sharp waves, and reported only discharges considered to be definite epileptiform spikes. This reviewer was also responsible for false-positive identification (relative to conventional review) of one seizure on a single study. The CSA display for this event is shown in Figure 1 where all 46 seizures were accurately marked.

Reader C achieved high sensitivity for PEDs (100%), RDA (93.3%) and FS (100%). In one cEEG this reviewer identified GS, FS, and PEDs but missed brief periods of rhythmic delta activity. Reader C missed 37 of 57 seizures in a record in which the CSA display was largely obscured by artifact (Figure 4).

Spearman’s rho test revealed a statistically significant negative relationship between the number of seizures per cEEG study and seizure detection sensitivity for reader C (R2[13] = -0.68, p=0.0096). This suggests that the percentage of seizures identified by reader C decreased as the number of seizures per cEEG study increased. For Readers A and B Spearman’s rho test did not show significant values (reader A: R2[9] = -0.3, p=0.42 and reader B R2[18] = -0.1, p=0.54).

References:

1. Hirsch LJ, SM L, Gaspard N, et al. American Clinical Neurophysiology Society’s Standardized Critical Care EEG Terminology: 2012 version. 2012;

2. Young GB, Jordan KG, Doig GS. An assessment of nonconvulsive seizures in the intensive care unit using continuous EEG monitoring: an investigation of variables associated with mortality. [Internet]. Neurology 1996;47(1):83–9.Available from: