.png){kind=icon}
.png){kind=icon}
Return
An adaptive epilepsy detection system using federated learning
.png){kind=icon}
PRE
AI
H
H
G
DOI：10.1007/s11042-026-21858-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Epilepsy is a prevalent neurological condition characterized by recurrent seizures, requiring accurate and effective detection technologies to enhance patient care and quality of life. Traditional machine learning (ML) approaches often rely on centralized data, posing significant challenges related to data silos, data heterogeneity (non-IID distribution), and stringent patient privacy laws (e.g., GDPR, CCPA). To overcome these limitations, this paper introduces an Adaptive Federated Learning (FL) system designed to enable collaborative model training across 14 distributed clinical clients using the Federated Averaging (FedAvg) aggregation strategy. The system employs dedicated Sequential Convolutional Neural Network (CNN) architectures specifically optimized for each modality. For temporal EEG signals (CHB-MIT dataset), a custom 2D-CNN was developed to capture spatio-temporal features. Similarly, a specialized CNN was tailored for structural MRI images (OpenNeuro dataset). These models are processed independently within a unified FL system to ensure modality-specific optimization. Experimental results demonstrate that the FL approach significantly exceeds centralized benchmarks, achieving a high detection accuracy of 98.53% for EEG and 91.35% for MRI datasets. Beyond accuracy, the system exhibits high clinical relevance through robustness to non-IID data and a low false-alarm rate (0.11% FPR for EEG and 3% FPR for MRI). Crucially, the FL approach achieved a substantial reduction in training latency compared to centralized learning: EEG model training was reduced by 56% (from 7h 43m to 3h 24m), and MRI model training time decreased by 80% (from 5h 9m to 59m). These findings confirm the system’s scalability and its efficacy as a robust, privacy-preserving, and computationally efficient solution for multi-modal epilepsy detection.
Keywords:
Epilepsy
Federated learning
Seizure detection
Medical imaging
Electrical signals
Data privacy
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W
Organization
Cited Papers