FED-LIFE: Ghost LinkNet Enabled Federated Learning for Anomaly Detection in Smart Intensive Care Unit based on IOMT

Intensive Care Unit (ICU) patient monitoring plays a vital role in ensuring the safety and well-being of critically ill patients by providing continuous and real-time insights into their health status. The integration of Internet of Medical Things (IoMT) devices in ICU including wearable sensors and remote monitoring tools, enables the seamless collection and transmission of patient data, allowing for real-time tracking of vital signs. Federated learning (FL) enhances this process by utilizing decentralized patient data to improve model generalization while maintaining data privacy. However, FL-based ICU monitoring faced challenges including high delays in decision-making due to centralized data processing, and significant execution time caused by the need to transfer large volumes of patient data. This research proposes a novel FEDerated learning-based LIFE saving ICU system (FED-LIFE) for effective tracking and providing timely health services to patients. The FED-LIFE system initially trains local models utilizing Ghostnet combined Enhanced LinkNet (Ghost_EliNet) which combines GhostNet and LinkNet, for tuning the Ghost_EliNet model a Red Deer Optimization (RDO) algorithm is employed for accurate service allocation. The suggested approach is implemented in Python programming. The efficacy of the developed approach is evaluated utilizing several metrics namely Precision, recall, f1-score, accuracy, delay, throughput, and execution time. The proposed method achieves the lowest delay of 22 seconds for 50 patients. Whereas the existing FEDSDM, Deep-CFL, and FL-IRL attain 45 seconds, 37 seconds, and 35 seconds for 50 patients respectively.