Seizure detection: correlation of human experts

Area of Science:

• Neuroscience
• Medical Technology
• Biomedical Engineering

Background:

• Accurate seizure detection from electroencephalography (EEG) is crucial for epilepsy diagnosis and management.
• Existing comparison methods for seizure detection algorithms may not fully capture the nuances of seizure event marking.

Purpose of the Study:

• To introduce and evaluate a novel overlap-integral comparison method for quantifying human accuracy in marking EEG seizure events.
• To establish human performance benchmarks for the development and testing of automated seizure detection algorithms.

Main Methods:

• Four human experts marked ten 8-hour EEG records from epilepsy patients with diverse seizure types.
• The novel overlap-integral method was applied alongside the traditional any-overlap method.
• Performance metrics including sensitivity, specificity, and correlation were calculated for both methods.

Main Results:

• The overlap-integral method demonstrated high average sensitivity (0.82) and specificity (0.9926), with a correlation of 0.80.
• The traditional any-overlap method yielded an average sensitivity of 0.92 and a false positive rate of 0.117 per hour.
• While human readers showed high interchangeability, certain seizure characteristics (e.g., short/long durations, ambiguous offsets) complicated analysis and reduced correlation in some records.

Conclusions:

• The overlap-integral method provides a sensitive measure for seizure endpoint accuracy, complementing the any-overlap method.
• Human reader variability, influenced by seizure characteristics, can pose challenges for algorithm validation.
• Establishing reliable human performance benchmarks is essential for advancing automated seizure detection technology.