PPubMed28 Feb 2025Cited 1×

Data transformation of unstructured electroencephalography reports by natural language processing: improving data usability for large-scale epilepsy studies

Yoon Gi Chung, Jaeso Cho, Young Ho Kim, Hyun Woo Kim, Hunmin Kim, Yong Seo Koo, Seo-Young Lee, Young-Min Shon

Abstract

1## Introduction Electroencephalography (EEG) is a noninvasive diagnostic modality that is specifically designed to record neuronal activity within the brain. This technique has demonstrated considerable efficacy in identifying clinically significant cortical electrophysiological markers in individuals suffering from neurological disorders. As such, EEG assessments serve as preliminary diagnostic measures for neurological conditions, especially epilepsy (1-3). Typically, EEG reports are composed of unstructured data formatted as free-text, which varies in stylistic presentation depending on th

A figure from Data transformation of unstructured electroencephalography reports by natural language processing: improving data usability for large-scale epilepsy studies — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

A hierarchical NLP algorithm converts pediatric EEG reports into structured data with >99% accuracy, enabling large‑scale epilepsy research.

What’s new

i.Developed a hierarchical algorithm combining BERT/LSTM models with rule‑based keyword extraction to classify EEG reports as normal or abnormal.
ii.Achieved near‑perfect classification performance (sensitivity 100%, specificity 99.90%, accuracy 99.95%) on internal validation.
iii.Validated the approach on an independent hospital dataset, attaining 100% classification accuracy and high keyword‑extraction accuracy.
iv.Created structured, tabular datasets from unstructured EEG reports, facilitating large‑scale epilepsy studies.

How they did it

The researchers retrospectively gathered 17,172 EEG reports from 3,423 pediatric epilepsy patients, manually labeled them as normal or abnormal, and used 5,173 reports for algorithm development. They trained BERT‑with‑LSTM and Clinical BERT‑with‑LSTM models and applied neurologist‑guided keyword rules, employing 6‑fold cross‑validation and external validation on 400 reports from another hospital.

What they found

Classification sensitivity 100%, specificity 99.90%, accuracy 99.95%, AUC 100% (internal).
External validation classification accuracy 100% for normal/abnormal detection.
Keyword extraction accuracy for seizure type: 98.5% (internal) and 100% (external).
IED location extraction accuracy: 97.5% (internal) and 88.5% (external).

Caveats

✕Performance on IED location extraction dropped to 88.5% in external validation, indicating difficulty with channel name variations.
✕Errors arose from misclassifying background activity as IEDs, showing limits of current rule‑based extraction.
✕Findings are based on a limited number of error cases, making broad generalizations difficult.
✕The algorithm relies on pre‑selected clinical keywords, which may limit adaptability to other report styles.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Similar papers in all four archives

The same idea searched in arXiv, PubMed, bioRxiv and chemRxiv at once, so a method can lead you to where it is being applied.

AarXiv4 found

PPubMed4 found

BbioRxiv4 found

CchemRxiv2 found

Searching the other archives…