Diagnostic Value of Electroencephalography with Ten Electrodes in Critically Ill Patients.

BACKGROUND: In critical care settings, electroencephalography (EEG) with reduced number of electrodes (reduced montage EEG, rm-EEG) might be a timely alternative to the conventional full montage EEG (fm-EEG). However, past studies have reported variable accuracies for detecting seizures using rm-EEG. We hypothesized that the past studies did not distinguish between differences in sensitivity from differences in classification of EEG patterns by different readers. The goal of the present study was to revisit the diagnostic value of rm-EEG when confounding issues are accounted for. METHODS: We retrospectively collected 212 adult EEGs recorded at Massachusetts General Hospital and reviewed by two epileptologists with access to clinical, trending, and video information. In Phase I of the study, we re-configured the first 4 h of the EEGs in lateral circumferential montage with ten electrodes and asked new readers to interpret the EEGs without access to any other ancillary information. We compared their rating to the reading of hospital clinicians with access to ancillary information. In Phase II, we measured the accuracy of the same raters reading representative samples of the discordant EEGs in full and reduced configurations presented randomly by comparing their performance to majority consensus as the gold standard. RESULTS: Of the 95 EEGs without seizures in the selected fm-EEG, readers of rm-EEG identified 92 cases (97%) as having no seizure activity. Of 117 EEGs with "seizures" identified in the selected fm-EEG, none of the cases was labeled as normal on rm-EEG. Readers of rm-EEG reported pathological activity in 100% of cases, but labeled them as seizures (N = 77), rhythmic or periodic patterns (N = 24), epileptiform spikes (N = 7), or burst suppression (N = 6). When the same raters read representative epochs of the discordant EEG cases (N = 43) in both fm-EEG and rm-EEG configurations, we found high concordance (95%) and intra-rater agreement (93%) between fm-EEG and rm-EEG diagnoses. CONCLUSIONS: Reduced EEG with ten electrodes in circumferential configuration preserves key features of the traditional EEG system. Discrepancies between rm-EEG and fm-EEG as reported in some of the past studies can be in part due to methodological factors such as choice of gold standard diagnosis, asymmetric access to ancillary clinical information, and inter-rater variability rather than detection failure of rm-EEG as a result of electrode reduction per se.

Expert and deep learning model identification of iEEG seizures and seizure onset times.

Hundreds of 90-s iEEG records are typically captured from each NeuroPace RNS System patient between clinic visits. While these records provide invaluable information about the patient's electrographic seizure and interictal activity patterns, manually classifying them into electrographic seizure/non-seizure activity, and manually identifying the seizure onset channels and times is an extremely time-consuming process. A convolutional neural network based Electrographic Seizure Classifier (ESC) model was developed in an earlier study. In this study, the classification model is tested against iEEG annotations provided by three expert reviewers board certified in epilepsy. The three experts individually annotated 3,874 iEEG channels from 36, 29, and 35 patients with leads in the mesiotemporal (MTL), neocortical (NEO), and MTL + NEO regions, respectively. The ESC model's seizure/non-seizure classification scores agreed with the three reviewers at 88.7%, 89.6%, and 84.3% which was similar to how reviewers agreed with each other (92.9%-86.4%). On iEEG channels with all 3 experts in agreement (83.2%), the ESC model had an agreement score of 93.2%. Additionally, the ESC model's certainty scores reflected combined reviewer certainty scores. When 0, 1, 2 and 3 (out of 3) reviewers annotated iEEG channels as electrographic seizures, the ESC model's seizure certainty scores were in the range: [0.12-0.19], [0.32-0.42], [0.61-0.70], and [0.92-0.95] respectively. The ESC model was used as a starting-point model for training a second Seizure Onset Detection (SOD) model. For this task, seizure onset times were manually annotated on a relatively small number of iEEG channels (4,859 from 50 patients). Experiments showed that fine-tuning the ESC models with augmented data (30,768 iEEG channels) resulted in a better validation performance (on 20% of the manually annotated data) compared to training with only the original data (3.1s vs 4.4s median absolute error). Similarly, using the ESC model weights as the starting point for fine-tuning instead of other model weight initialization methods provided significant advantage in SOD model validation performance (3.1s vs 4.7s and 3.5s median absolute error). Finally, on iEEG channels where three expert annotations of seizure onset times were within 1.5 s, the SOD model's seizure onset time prediction was within 1.7 s of expert annotation.