Automated estimation of frequency and spatial extent of periodic and rhythmic epileptiform activity from continuous electroencephalography data

Abstract

This datasets and code supports our study on the development of an automated algorithm estimating the frequency and spatial extent of periodic and rhythmic epileptiform discharges from countinous electroencephalography (EEG) data.

We provide 1087 segments of EEG recordings classified as lateralized or generalized periodic discharge and rhythmic delta activity, expert annotations from three neurophysiologists on their frequency and spatial extent, along with the code developed for the automatic estimation. 

We evaluated the algorithms on segments with 100% agreement on event classification (n=389) and on the full cohort of 1087 segments (including disagreements). For the first subset, top algorithms matched or exceeded expert agreement on frequency and spatial extent. For the full cohort, inter-rater reliability (IRR) declined, but expert-algorithm inter-rater-reliability remained comparable or superior to experts. 

 

Background

Rhythmic and periodic patterns (RPP) are harmful brain activity observed on EEG recordings of critically ill patients. Periodic discharges (PD) are defined as repetitive waveforms with clear inter-discharge background activity occurring at regular intervals. Rhythmic delta activity (RDA) is characterized by repetitive slower waves between 0.5-4Hz. American Clinical Neurophysiology Society (ACNS) further characterizes these patterns as lateralized or generalized periodic discharges (LPD, GPD) and lateralized or generalized rhythmic delta activity (LRDA, GRDA), and the ictal-interictal continuum (IIC). The IIC is defined as a combination of LPD, GPD or LRDA observed in 10 seconds of EEG recordings.

RPPs and the IIC have been directly linked to a higher likelihood of mortality and worse functional and cognitive outcomes in patients with acute brain injuries such as stroke and trauma. One of the major limitations in pursuing large scale clinical research studies to examine the impact of anti-seizure treatment of these patterns if the labor-intensive task of manual review of EEG recordings collected in the critical care setting. Automated methods of identifying RPPs and IIC patterns and estimating their frequency and spatial extent, can help clinical patient monitoring as well as conducting research studies on large patient population to identify links between treatments and outcomes.

Methods

A total of six rule-based algorithms were proposed for the automatic annotation of frequency and spatial extent of periodic discharges and rhythmic delta activity. The best performing algorithms are highlighted.

• RDA 1a - based on Fourier Transform and frequency modeling of aperiodic components, delta peak detection
• RDA 1b - based on EEG channel selector using data variance quality checks, Fourier Transform, frequency modeling of aperiodic components, delta peak detection 
• RDA 2 - based on Hilbert Transform frequency analysis
• PD 1 - based on fixed thresholds for peak detection
• PD 2a - based on an adaptive first derivative peak detector
• PD 2b - based on a z-score first derivative peak detector

For validation, we used a dataset comprised of 10 second EEG recordings previously labeled as LPD, GPD, LRDA and GRDA by 20 clinical experts. Segments where the majority of annotators agreed on a single pattern type were selected. A total of 296 LPD, 296 GPD, 210 LRDA and 285 GRDA segments were further selected for expert annotation. 

Three clinical neurophysiologists performed independent annotations of the pattern type, classification, frequency f event and spatial extent. The participants were instructed to annotated the segments using the ACNS terminology in a custom interface.

Results were evaluated based on expert-algorithm IRR compared to expert-expert IRR. The metrics for IRR evaluation were the intra-class correlation coefficient and percent agreement. 

More details on the applied methods are available in the original publication.

Data Description

The repository contains the code for running the algorithms and reproducing the data analysis pipeline for validation, along with the data used in the validation study and the annotations of the neurophysiologists.

• Code
  • Algorithms
    • package: pd_detector - PD1
    • package: pd_detector_alternate - PD2a & PD2b
    • package: rda_detector - RDA1a, RDA1b, RDA2
    • example code for usage: extract_frequency_spatial_extent.py
  • Data visualization
    • package: imageGeneration
    • example code for usage: visualize_output.py
  • Data analysis
    • irr_analysis_fulldataset.ipynb
    • irr_analysis_onagreement.upynb
  • Dependencies: environment.yml
  • How to get started: readme.txt
• Data
  • EEG recordings
  • Annotations
• Results
  • Algorithm output: .csv files for LPD, GPD, LRDA, GRDA
  • Visualizes results: results_figures

Usage Notes

Code to generate all results and figures from the publication are provided here. 
Data is provided through an AWS access point (see below). 

Ethics

This study was conducted under protocols approved by the institutional review boards of Massachusetts General Hospital (protocols #2023P000487, #2024P002630) and Beth Israel Deaconess Medical Center (protocols #2022P000481, #2022P000417), which waived the requirement for informed con-sent for this retrospective analysis.

Acknowledgements

This work was supported by grants from the NIH (R01NS131347, R01NS126282).

Conflicts of Interest

Dr. Westover if a co-founder, scientific advisor, consultant to, and has personal equity in Beacon Biosignals. Dr. Zafar receives publishing royalties from Springer and Wolter Kluwer.

References

1. Tăuțan AM, Jing J, Basovic L, Hadar PN, Sartipi S, Fernandes MP, Kim J, Struck AF, Westover MB, Zafar SF. Automated estimation of frequency and spatial extent of periodic and rhythmic epileptiform activity from continuous electroencephalography data. J Neural Eng. 2025 Dec 16;22(6). doi: 10.1088/1741-2552/ae2716. PMID: 41330044.