| Title:
|
Seizure Prediction from Scalp EEG Using Random Forest and Spectral Power Features: A Preliminary Multi-Patient Analysis
|
| Author(s):
|
1
|
| Subject:
|
Brain Mapping Applications in Epilepsy – Across Pediatric and Adult Populations
|
|
|
|
Presenter Author:
|
|
| Name:
|
Mostafa RostampourVajari
|
|
|
|
Author:
|
|
| Name 1:
|
Mostafa RostampourVajari - Presenter
|
| Affiliation 1:
|
Department of Basic Science, Faculty of Veterinary Medicine, KIAU, Karaj, Iran.
|
|
|
| Name 2:
|
Mohammad Rostampour - Corresponding
|
| Affiliation 2:
|
Department of Physiology, School of Medicine, Guilan University of Medical Sciences, Rasht, Iran.
|
|
|
| Name 3:
|
|
| Affiliation 3:
|
|
|
|
| Name 4:
|
|
| Affiliation 4:
|
|
|
|
| Name 5:
|
|
| Affiliation 5:
|
|
|
|
| Name 6:
|
|
| Affiliation 6:
|
|
|
|
| Name 7:
|
|
| Affiliation 7:
|
|
|
|
| Name 8:
|
|
| Affiliation 8:
|
|
|
|
| Name 9:
|
|
| Affiliation 9:
|
|
|
|
| Name 10:
|
|
| Affiliation 10:
|
|
|
|
Abstract
|
Original Research
|
|
introduction :
|
Automated seizure prediction from scalp EEG could improve patient safety by providing early warning before seizure onset. This study aimed to evaluate a machine-learning approach for distinguishing the preictal (pre-seizure) state from the interictal (baseline) state using spectral EEG features, as a preliminary step toward a laboratory-independent methodology feasible without access to specialized neuroimaging equipment.
|
|
Methods:
|
We used the publicly available CHB-MIT Scalp EEG Database (PhysioNet), including recordings from 5 pediatric patients with a total of 27 annotated seizures (23-channel scalp EEG, 256 Hz sampling rate). For each seizure, up to 5 minutes preceding onset were labeled preictal; seizure-free recordings not temporally adjacent to ictal files were labeled interictal. Signals were segmented into non-overlapping 10-second epochs. For each epoch and channel, spectral power in five canonical EEG frequency bands (delta, theta, alpha, beta, gamma) was computed using Welch's method and log-transformed, yielding 115 features per epoch. A Random Forest classifier (200 trees, class-balanced) was trained separately for each patient and evaluated using 5-fold group cross-validation, with folds constructed at the level of the original seizure/interictal recording to prevent information leakage between temporally adjacent epochs.
|
|
Results:
|
Averaged across the 5 patients, the model achieved a mean accuracy of 82.8%, sensitivity of 74.7%, specificity of 86.2%, and AUC of 0.847. Performance varied across patients (accuracy range: 69.3%-99.7%), consistent with known inter-patient variability in seizure electrophysiology. Feature-importance analysis consistently identified gamma-band power across multiple scalp channels as the most informative predictor of the preictal state, in agreement with previous reports implicating high-frequency cortical activity in seizure genesis.
|
|
Conclusion:
|
These preliminary results demonstrate the feasibility of predicting the preictal state from scalp EEG using interpretable spectral features and a computationally lightweight classifier, without requiring intracranial recordings or specialized laboratory infrastructure. The consistent contribution of gamma-band activity offers a physiologically interpretable basis for prediction, which may support clinical trust in future closed-loop seizure-warning systems. Given the modest patient sample, these findings should be considered preliminary; ongoing work aims to extend the analysis to a larger, more diverse patient cohort and evaluate patient-independent generalization as the basis for a full-length manuscript.
|
| Attachment:
|
|
| Keywords:
|
Electroencephalography, Seizures, Epilepsy, Machine Learning, Brain mapping, Signal Processing
|
| Key message:
|
|
| Do you want your abstract to be indexed, if this is selected as one of the top 100
articles?
|
