Electroencephalogram data platform for application of reduction methods

Abstract

Long-term electroencephalogram (EEG) monitoring (≥24-h) is a resourceful tool for properly diagnosis sparse life-threatening events like non-convulsive seizures and status epilepticus in Intensive Care Unit (ICU) inpatients. Such EEG data requires objective methods for data reduction, transmission and analysis. This work aims to assess specificity and sensibility of HaEEG and aEEG methods in combination with conventional multichannel EEG when achieving seizure detection. A database architecture was designed to handle the interoperability, processing, and analysis of EEG data. Using data from CHB-MIT public EEG database, the reduced signal was obtained by EEG envelope segmentation, with 10 and 90 percentiles obtained for each segment. The use of asymmetrical filtering (2-15 Hz) and overall clinical band (1-70 Hz) was compared. The upper and lower margins of compressed segments were used to classify ictal and non-ictal epochs. Such classification was compared with the corresponding specialist seizure annotation for each patient. The difference between medians of instantaneous frequencies of ictal and non-ictal periods were assessed using Wilcoxon Rank Sum Test, which was significant for signals filtered from 2 to 15 Hz (p = 0.0055) but not for signals filtered from 1 to 70 Hz (p = 0.1816).