کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Multi-Agent Large Language Model Framework for Comprehensive ECG Signal Analysis: Delineation, Classification, and Arrhythmia Subtyping with Temporal Hierarchical Reasoning

نویسندگان :

Mohsen Avesta¹ Maedeh Avesta² Asma Yousefian Baboukani³ Bashir Najafabadian⁴ Mohaddeseh Behjati⁵

1- Faculty of Electrical Engineering, Shahid Beheshti University, Tehran, Iran 2- Department of Biomedical Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Iran 3- Department of Biomedical Engineering, Sheikh Bahaei University, Isfahan, Iran 4- Department of Biomedical Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran 5- Interventional Cardiology Research Center, Cardiovascular Research Institute, Isfahan University of Medical Sciences, Isfahan, Iran

کلمات کلیدی :

ECG analysis،multi-agent systems،arrhythmia classification،large language models،deep learning،signal delineation،temporal hierarchical reasoning،uncertainty quantification

چکیده :

This paper introduces a novel multi-agent framework leveraging large language models (LLMs) and deep neural networks for comprehensive ECG signal analysis. The framework comprises four specialized agents working in hierarchical coordination: (1) ECG delineation via attention-enhanced U-Net++ with adaptive boundary refinement, (2) AAMI EC57-compliant beat classification using hybrid ResNet-101-BiLSTM architecture with focal loss optimization, (3) Arrhythmia type detection through disease-specific attention networks (DANet) with clinical rule integration, and (4) Sub-arrhythmia classification with multimodal temporal attention and cross-lead correlation analysis. We formulate the multi-agent coordination as a Markov Decision Process (MDP) with uncertainty-aware communication protocols. The system processes raw ECG signals through a hierarchical feature extraction pipeline where each agent contributes specialized diagnostic capabilities. Extensive validation on MIT-BIH, PTB-XL, and CPSC2018 datasets demonstrates state-of-the-art performance: 98.73% delineation accuracy (mean absolute error of 3.8ms for R-peaks), 96.41% EC57 classification F1-score, 94.82% arrhythmia type recognition accuracy, and 93.17% subarrhythmia classification accuracy. Ablation studies reveal that the multi-agent architecture provides 7.32% performance gain over monolithic approaches through specialized decomposition and cross-agent knowledge transfer. The framework reduces false positives by 34.8% compared to existing methods while maintaining real-time processing capabilities (15.3ms per 10- second segment on NVIDIA A100 GPU). Clinical validation with 12 cardiologists confirms 96.2% diagnostic agreement on complex arrhythmia cases.