AR2

AR2 reduces muscle artifact contamination in ictal scalp electroencephalogram (EEG) recordings to improve accuracy of seizure-onset localization.

Key Features:

• Muscle artifact reduction: Reduces muscle artifact contamination in ictal scalp EEG to render previously uninterpretable recordings analyzable.
• Head-to-head comparison: Evaluated directly against AR1 (a commercially available method) with superior performance in delineating seizure-onset time and increasing reader confidence.
• Validation dataset: Validated on a blinded study of 23 EEG recordings from eight patients with seizures initially uninterpretable due to muscle artifacts.
• Reader assessment: Processed recordings were reviewed by 26 EEG specialists who assessed seizure-onset time, lateralization, region, and confidence.
• Quantitative performance metrics: Reported outcomes include proportion of readers able to assign seizure-onset time (15/23 with AR2 vs 10/23 with AR1, p<0.01), increased reader confidence (p<0.05), and intra-class correlation (ICC) for seizure-onset time (AR2 0.26, 95% CI 0.21–0.30; AR1 0.15, 95% CI 0.11–0.18).
• Clinical concordance: EEG assignments using AR2 were frequently congruent with behavioral, neurophysiological, and neuro-radiological findings, with left-sided accuracy 95.9% (CI 85.7–98.9%, n=4) versus 91.9% (77.0–97.5%) for AR1.

Scientific Applications:

• Epilepsy diagnosis: Improves diagnostic interpretation of ictal scalp EEG in patients with seizures contaminated by muscle artifact.
• Seizure-onset localization: Enhances accuracy of seizure-onset time determination, lateralization, and regional localization from scalp EEG.
• Integration with clinical data: Produces artifact-reduced EEG interpretations that can be compared with behavioral, neurophysiological, and neuro-radiological findings to inform clinical assessment.

Methodology:

EEG recordings were processed with AR1 and AR2 and subjected to a blinded review by 26 EEG specialists assessing seizure-onset time, lateralization, region, and confidence, with statistical evaluation using proportions (p-values) and intra-class correlation (ICC) with 95% confidence intervals.