A Real-Time Cardiac Decision Support System for Internet of Medical Things Using Edge AI and DDS

Deep Learning (DL)-based real-time cardiac monitoring systems face a fundamental trade-off between computational accuracy and constraints on edge devices. Although complex Convolutional Neural Networks (CNNs) deliver high diagnostic precision, their deployment on battery-powered Internet of Medical Things (IoMT) devices often results in increased latency and thermal throttling. To address the limitations of traditional IoMT architectures, this study proposes a decentralized Decision-Support System (DSS) that integrates the deterministic Data Distribution Service (DDS) middleware Real-Time Innovations (RTI) Connext with a demographic-specific dual-model Edge Artificial Intelligence (AI) framework. The system utilizes specialized lightweight MobileNetV3 classifiers at the edge inference node to analyze multi-patient ECG data, transmitting only actionable alerts to resource-constrained ward-level Internet of Things (IoT) nodes (Raspberry Pi 5). Evaluation on the MIT-BIH (Adult) and Leipzig (Pediatric) datasets, using a balanced, fixed-size test split (n=105 and n=150 samples per class, respectively), yielded overall test accuracies of 98.39% and 99.43%, respectively. The system maintained robust sensitivities of 93.57% for adults and 97.75% for children. Experimental validation shows that while wireless mobility introduces a 5.5-fold latency penalty relative to a wired baseline, the worst-case response time remains at 99.97 ms, well below the 300 ms medical safety threshold. Furthermore, empirical power profiling shows that the DDS Reliable Quality of Service (QoS) policy yields lower total energy consumption than standard Best Effort delivery, thanks to efficient transmission flow control. These results demonstrate the technical feasibility of the proposed DDS-enabled edge architecture through comprehensive empirical evaluation, providing a clinically relevant proof-of-concept framework for real-time cardiac monitoring and alert dissemination in heterogeneous IoMT environments.