Calibrated Uncertainty Estimation Based on Bayesian Models for EEG Epileptic Seizure Detection
IEEE Sensors Journal · 01.05.2026 · 26 (9) · 13533-13545
DOI: 10.1109/jsen.2026.3674940 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.05.2026. Verfasst von Hu Jiahao und 2 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 46 Literaturangaben. Gefördert von: King Abdullah University of Science and Technology (KAUST)—KAUST Center of Excellence for Smart Health, KAUST Faculty Baseline Funds.
Abstract (Originalsprache)
Machine learning-empowered electroencephalogram (EEG)-based Epileptic seizure detection methods often overlook issues such as model uncertainty and model calibration. Model calibration is crucial for compensating for the model uncertainty as it ensures that the predicted probabilities reflect the true likelihood of outcomes. This paper introduces a novel end-to-end EEG-based Epileptic seizure detection framework via calibrated model uncertainty learning. In the proposed framework, we combine Bayesian models and calibration methods (i.e., Platt scaling, Isotonic regression, Beta calibration, Spline scaling, and temperature scaling). Three calibrated Bayesian models are tested and compared: Gaussian Process, Monte-Carlo Dropout, and Bayesian Neural Network. Two datasets have been used to test the performance of the proposed framework: the Bonn dataset and the Temple Lemon dataset. The results show that the proposed model calibration approach improves the model accuracy evaluation metrics such as area under the curve (AUC), precision, and sensitivity, as well as the calibration metrics such as expected calibration error (ECE) score, compared with previous benchmark methods. The obtained AUC, precision, sensitivity, ECE scores are given by 0.918, 0.972, 0.9829, 0.0893 (for Bonn dataset), and 0.997, 0.952, 0.987, 0.0385 (for Temple Lemon dataset), respectively. Thus, the results confirm that, under the proposed framework, the models could effectively provide precise uncertainty estimates of their predictions, as well as are capable of doing robust, calibrated detection of EEG Epileptic seizures.
Förderung
- King Abdullah University of Science and Technology (KAUST)—KAUST Center of Excellence for Smart Health
- KAUST Faculty Baseline Funds
Zitieren
APA
Jiahao, H., Rahman, M. M. U., & Laleg-Kirati, T. (2026). Calibrated Uncertainty Estimation Based on Bayesian Models for EEG Epileptic Seizure Detection. IEEE Sensors Journal, 26(9), 13533-13545. https://doi.org/10.1109/jsen.2026.3674940
Vancouver
Jiahao H, Rahman MMU, Laleg-Kirati T. Calibrated Uncertainty Estimation Based on Bayesian Models for EEG Epileptic Seizure Detection. IEEE Sensors Journal. 2026;26(9):13533-13545. doi:10.1109/jsen.2026.3674940
BibTeX
@article{Jiahao2026,
author = {Jiahao, Hu and Rahman, Muhammad Mahboob Ur and Laleg-Kirati, Taous-Meriem},
title = {Calibrated Uncertainty Estimation Based on Bayesian Models for EEG Epileptic Seizure Detection},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {9},
pages = {13533--13545},
doi = {10.1109/jsen.2026.3674940}
}
EUVIMED