Data transformation of unstructured electroencephalography reports by natural language processing: improving data...

Area of Science:

• Medical Informatics
• Computational Neuroscience
• Pediatric Neurology

Background:

• Electroencephalography (EEG) reports are crucial for diagnosing pediatric epilepsy but are often unstructured text, hindering data analysis.
• Extracting clinically relevant information from these reports is challenging due to their textual format.

Purpose of the Study:

• To develop and validate a hierarchical algorithm for transforming unstructured EEG reports of pediatric epilepsy patients into structured data.
• To enhance the extraction and analysis of critical electrographic insights using natural language processing (NLP) techniques.

Main Methods:

• A two-phase algorithm was developed: deep learning for text classification (normal vs. abnormal EEG reports) and rule-based keyword extraction.
• The algorithm systematically identified seizure types (focal/generalized), epileptiform discharges, and anatomical locations.
• A dataset of 17,172 pediatric EEG reports was retrospectively analyzed for algorithm development and validation.

Main Results:

• The algorithm achieved high accuracy in classifying EEG reports and identifying cerebral dysfunction or seizures.
• Accuracy for seizure type determination exceeded 98.5%, and for epileptiform discharge detection, it surpassed 88.5%.
• Validation across multiple institutions confirmed the algorithm's robust performance.

Conclusions:

• The developed hierarchical algorithm effectively structures unstructured EEG reports for pediatric epilepsy patients.
• This NLP-driven approach significantly enhances data accessibility for research and clinical applications in pediatric epilepsy management.
• The methodology streamlines the conversion of clinical notes into structured datasets, facilitating further investigation.