Agentic AI-Based Medical Support System for Cardiovascular Diseases Using IoMT Sensors

Large Language Models (LLMs) have recently enabled the development of agentic systems capable of coordinating complex workflows through natural language and external tool invocation. However, existing healthcare applications primarily focus on clinical reasoning after patient data have been acquired, with limited support for real-time physiological sensing and IoMT device orchestration. This paper presents an Agentic AI-based medical support system for cardiovascular care that integrates multiple LLM-driven agents with an Internet of Medical Things (IoMT) infrastructure for patient-guided electrocardiogram (ECG) acquisition, physiological signal processing, and interpretable cardiovascular risk assessment. The proposed architecture combines an Executor-mediated orchestration layer, explicit state management, and autonomous recovery mechanisms to supervise ECG acquisition while coordinating interactions between conversational agents, medical devices, and backend services. A hardware prototype based on an ESP32 and AD8232 ECG sensor is developed to demonstrate real-time signal acquisition, while NeuroKit2 is employed for ECG quality assessment and feature extraction. Cardiovascular risk scores are subsequently interpreted through a fuzzy inference layer to provide transparent decision support while preserving established clinical scoring models. The framework is evaluated through hardware monitoring, functional verification of the multi-agent workflow, and a scenario-based evaluation of the IoT Agent across representative acquisition conditions using an independent evaluation agent. The experimental results demonstrate consistent workflow execution, effective failure recovery, stable hardware performance, and reliable state synchronization under controlled experimental scenarios. These findings indicate the feasibility of integrating LLM-based agents with IoMT devices to support end-to-end cardiovascular monitoring and decision-support workflows.