View in EDS

Artificial intelligence‐assisted detection of epileptic spasms using electroencephalographic–video analysis.

Saved in:
Bibliographic Details
Title: Artificial intelligence‐assisted detection of epileptic spasms using electroencephalographic–video analysis.
Authors: Wan, Lin (AUTHOR), Lin, Nan (AUTHOR), Wang, Wen (AUTHOR), Liang, Zi (AUTHOR), Dong, Yisu (AUTHOR), He, Haibo (AUTHOR), Chen, Jian (AUTHOR), Fan, Yuying (AUTHOR), Liu, Tong (AUTHOR), Wang, Hongjie (AUTHOR), Cheng, Weike (AUTHOR), Lu, Guangshuang (AUTHOR), Wang, Juan (AUTHOR), Zhang, Bo (AUTHOR), Lu, Qiang (AUTHOR), Yang, Guang (AUTHOR)
Source: Epilepsia (Series 4). Jun2026, Vol. 67 Issue 6, p3009-3022. 14p.
Subjects: Electroencephalography, Video processing, Diagnosis, Artificial intelligence, Diagnostic services, Childhood epilepsy, Infantile spasms, Machine learning
Abstract: Objective: This study was undertaken to develop and validate an artificial intelligence (AI) diagnostic tool using hybrid electroencephalographic (EEG)–video signals for automatic epileptic spasms (ES) detection. Methods: This retrospective cohort study with internal cross‐validation and multicenter external validation was conducted from July 2022 to May 2025. It included 252 patients with ES from Chinese PLA General Hospital and 60 from three other medical centers. We developed a multimodal fusion approach combining video and electrophysiological signals. All EEG data were segmented into continuous 4‐s pages. The internal cohort consisted of 212 patients (723.4 h video‐EEG, 7348 ES, and 643 215 non‐ES segments). Clinical validation involved 100 patients across four datasets (212.2 h, 5709 ES and 185 207 non‐ES segments) plus 78 controls without ES (218.1 h, 196 249 segments, used for false alarm rate analysis). Primary outcomes included sensitivity, specificity, accuracy, precision, and F1 score versus electroencephalographer interpretation. Results: In internal cross‐validation, the hybrid EEG–video model achieved superior performance compared to current EEG‐only models (area under the precision–recall curve =.7334, 95% confidence interval [CI] =.7118–.7539, area under the receiver operating characteristic curve =.9820, 95% CI =.9782–.9856). In the clinical validation dataset, the model demonstrated diagnostic sensitivity (.735, 95% CI =.673–.822) and specificity (.995, 95% CI =.992–.996) comparable to experienced electroencephalographers. With AI assistance, all electroencephalographers showed improved sensitivity trends, with one rater showing a marked improvement from.620 (95% CI =.532–.699) to.792 (95% CI =.757–.821). Notably, specificity did not decline significantly for any of the raters. In samples containing subtle ES, machine‐assisted recognition significantly improved sensitivity by 16%–21% for all electroencephalographers (p <.01). The model maintained low false alarm rates (.16‰) across different patient populations including healthy controls and epilepsy patients without ES. Significance: This AI diagnostic tool achieves clinical performance comparable to senior electroencephalographers when applied independently for ES detection, with greater robustness in detecting subtle ES. When used collaboratively with clinicians, it could enhance diagnostic sensitivity while maintaining high specificity. The technology addresses critical diagnostic challenges in pediatric epilepsy care and shows promise to reduce health care inequities in resource‐limited settings lacking specialized expertise. [ABSTRACT FROM AUTHOR]
Copyright of Epilepsia (Series 4) is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Psychology and Behavioral Sciences Collection
Description
Abstract:Objective: This study was undertaken to develop and validate an artificial intelligence (AI) diagnostic tool using hybrid electroencephalographic (EEG)–video signals for automatic epileptic spasms (ES) detection. Methods: This retrospective cohort study with internal cross‐validation and multicenter external validation was conducted from July 2022 to May 2025. It included 252 patients with ES from Chinese PLA General Hospital and 60 from three other medical centers. We developed a multimodal fusion approach combining video and electrophysiological signals. All EEG data were segmented into continuous 4‐s pages. The internal cohort consisted of 212 patients (723.4 h video‐EEG, 7348 ES, and 643 215 non‐ES segments). Clinical validation involved 100 patients across four datasets (212.2 h, 5709 ES and 185 207 non‐ES segments) plus 78 controls without ES (218.1 h, 196 249 segments, used for false alarm rate analysis). Primary outcomes included sensitivity, specificity, accuracy, precision, and F1 score versus electroencephalographer interpretation. Results: In internal cross‐validation, the hybrid EEG–video model achieved superior performance compared to current EEG‐only models (area under the precision–recall curve =.7334, 95% confidence interval [CI] =.7118–.7539, area under the receiver operating characteristic curve =.9820, 95% CI =.9782–.9856). In the clinical validation dataset, the model demonstrated diagnostic sensitivity (.735, 95% CI =.673–.822) and specificity (.995, 95% CI =.992–.996) comparable to experienced electroencephalographers. With AI assistance, all electroencephalographers showed improved sensitivity trends, with one rater showing a marked improvement from.620 (95% CI =.532–.699) to.792 (95% CI =.757–.821). Notably, specificity did not decline significantly for any of the raters. In samples containing subtle ES, machine‐assisted recognition significantly improved sensitivity by 16%–21% for all electroencephalographers (p <.01). The model maintained low false alarm rates (.16‰) across different patient populations including healthy controls and epilepsy patients without ES. Significance: This AI diagnostic tool achieves clinical performance comparable to senior electroencephalographers when applied independently for ES detection, with greater robustness in detecting subtle ES. When used collaboratively with clinicians, it could enhance diagnostic sensitivity while maintaining high specificity. The technology addresses critical diagnostic challenges in pediatric epilepsy care and shows promise to reduce health care inequities in resource‐limited settings lacking specialized expertise. [ABSTRACT FROM AUTHOR]
ISSN:00139580
DOI:10.1002/epi.70194