Automated information extraction from free-text EEG reports.

Area of Science:

• Neuroscience
• Medical Informatics
• Machine Learning

Background:

• Epilepsy diagnosis relies heavily on interpreting electroencephalogram (EEG) reports.
• Manual review of EEG reports is time-consuming and labor-intensive, hindering large-scale retrospective studies.
• Automated methods are needed to efficiently analyze EEG data for clinical research.

Purpose of the Study:

• To develop and validate a supervised learning system for the automated detection of seizures and epileptiform discharges in EEG reports.
• To improve the efficiency of identifying patient cohorts for epilepsy research.
• To reduce the manual workload associated with analyzing large volumes of neurophysiological data.

Main Methods:

• Manual labeling of 3,277 EEG reports for seizure and epileptiform discharge content.
• Development of a Naïve Bayes classifier incorporating normalization, key sentence extraction, and automated feature selection.
• Utilizing keywords and elastic word sequences (EWS) as candidate features, with sequential backward selection for optimization.
• Employing cross-validation for performance evaluation and out-of-sample prediction.

Main Results:

• The automated system achieved high accuracy in classifying EEG reports.
• Seizure detection accuracy reached an average area under the receiver operating curve (AUC) of 99.05%.
• Epileptiform discharge detection achieved an average AUC of 96.15%.

Conclusions:

• The developed supervised learning methodology effectively automates the identification of seizures and epileptiform discharges in EEG reports.
• This approach significantly streamlines the process of cohort identification for retrospective epilepsy studies.
• The system offers a scalable and accurate solution for analyzing neurophysiological data in epilepsy research.