Software Credentialed Access
Interpretable quantitative criteria for evaluating interictal epileptiform discharges
Fábio A. Nascimento , Jaden D. Barfuss , A. Jaffe , M. Brandon Westover , Jin Jing
Published: July 9, 2026. Version: 1.0.0
When using this resource, please cite:
(show more options)
Nascimento, F. A., Barfuss, J. D., Jaffe, A., Westover, M. B., & Jing, J. (2026). Interpretable quantitative criteria for evaluating interictal epileptiform discharges (version 1.0.0). Brain Data Science Platform. https://doi.org/10.60508/bp20-rz98.
Abstract
Objective. To evaluate interictal epileptiform discharges (IEDs) using a small set of explicit, interpretable quantitative EEG features, as a transparent complement to black-box spike detectors.
Methods. Six interpretable features (amplitude/prominence, duration, sharpness, after-going slow wave, asymmetry, and field/voltage-map spread) were computed from expert-annotated candidate waveforms and combined in a cross-validated logistic-regression model to classify IED vs non-IED.
Results. The combined six-feature model discriminated IEDs from non-IEDs with AUC ~0.83; individual features ranged from ~0.60 to ~0.78, matching clinical intuition about which morphologic properties are most informative.
Conclusion. A compact set of interpretable features captures much of the signal used to identify IEDs. This project releases the code and de-identified data to reproduce the feature computations, model, and figures.
Background
Automated spike detectors are often black boxes. This project provides an interpretable alternative: a small set of explicit morphologic EEG features, each clinically meaningful, combined in a transparent model to identify interictal epileptiform discharges (IEDs).Software Description
MATLAB per-figure and per-table code plus de-identified data: the feature matrix and labels (step4_output.mat), the fitted model (LRMwithCV.mat, Table2.xlsx), de-identified spike-waveform arrays (SpikesArray_jj.mat) for the feature illustrations, and one example EEG for the interactive scoring GUI. No PHI.Technical Implementation
Six interpretable features were computed from expert-annotated candidate waveforms and combined with cross-validated logistic regression to classify IED vs non-IED; per-feature and combined AUCs were assessed.Installation and Requirements
MATLAB R2016+ with the Statistics and Machine Learning Toolbox. Run Table2-modelPerformance/main_table2 from the repo. See REPRODUCE.md and DATA_SOURCE.md.Usage Notes
main_table2 reproduces the combined-model AUC 0.83 [0.81-0.85] and per-feature AUCs; the Figure* folders regenerate the illustrations and the interactive GUI.Release Notes
First public release: code + de-identified data + reproduction docs.Ethics
De-identified data collected under IRB approval; no identifiers released.Acknowledgements
MGH Clinical Data Animation Center.Conflicts of Interest
See the associated publication (Clin Neurophysiol 2023;146:1-8).Access
Access Policy:
Only credentialed users who sign the DUA can access the files.
License (for files):
BDSP Credentialed Health Data License 1.5.0
Data Use Agreement:
BDSP Credentialed Health Data Use Agreement
Required training:
Discovery
Corresponding Author
Files
- be a credentialed user
- sign the data use agreement for the project