A Novel Decentralized Federated Incremental Learning Framework for ECG and EEG Signal Analysis
Year: 2025
Type: article
Abstract: The increasing use of artificial intelligence (AI) in healthcare has revolutionized medical diagnostics, particularly in cardiology and neurology, where electrocardiograms (ECG) and electroencephalograms (EEG) play a crucial role in diagnosing conditions like heart attacks and epilepsy. However, the sensitive nature of medical data poses significant privacy concerns, limiting data sharing between institutions for AI model training. Federated learning (FL) offers a solution by enabling collaborative learning without sharing raw data. Traditional FL approaches rely on centralized servers, which introduce risks such as single points of failure and communication bottlenecks. To address these limitations, we propose a decentralized federated learning (DFL) system combined with incremental learning (IL), allowing continuous adaptation to new data streams while preserving patient privacy. Our architecture utilizes a CNN-BiLSTM model for physiological signal analysis, trained locally at each institution. Model weights are exchanged in a ring topology using an incremental federated averaging algorithm (IncFedAvg), ensuring efficient weight aggregation without a central server. The proposed system demonstrates high accuracy in both ECG arrhythmia classification and EEG seizure detection. Moreover, the incremental learning capability allows the model to adapt to real-time data while maintaining performance. This approach effectively addresses the challenges of privacy preservation and dynamic healthcare data processing, offering a scalable solution for medical institutions.
Authors: Rostom Mennour, Sofiane Blikaze, Samy Benaissa
Institution: University Frères Mentouri Constantine 1
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4415646342