Results indicate that a federated TinyML architecture with lightweight patient-state tracking, validation-based ensemble filtering, differential privacy, and post-quantum-secure communication can support privacy-aware and attack-resilient ICU monitoring experiments in resource-constrained IoMT settings.
Abstract
Resource-constrained medical sensing devices are increasingly expected to support local intelligence, privacy-preserving collaboration, and secure communication in Internet of Medical Things (IoMT) environments. However, deploying federated learning in ICU monitoring remains challenging because hospital data are often non-IID, model updates may be adversarially poisoned, and emerging quantum-security threats require stronger communication protection. This paper presents a federated TinyML framework with an edge-hosted patient-state Digital Twin layer for ICU monitoring. The Digital Twin component is implemented as a lightweight patient-state representation rather than a full physiological simulator. It maintains recent physiological observations, temporal risk trends, predicted status labels, and interpretable alert information at the hospital gateway. Patient-specific adaptation is therefore achieved through individualized temporal state tracking and risk-history synchronization, while the predictive model is collaboratively learned across hospitals. The suggested framework employs decision-tree ensembles due to their lightweight nature, efficient inference capabilities, and interpretability inherent to ESP32-class devices. As decision tree structures are not amenable to average aggregation like neural network parameters, the aggregation of local client models is done via the validation-based ensemble fusion. In order to enhance the resistance of the framework to attacks based on poisoning clients, the Performance-Based Filtering (PBF) method analyzes the performance of each local ensemble against a trusted validation set and excludes low-quality local ensembles from the subsequent global ensemble aggregation. The secure model exchange process utilizes ML-KEM-512 key encapsulation scheme in combination with AES-256-GCM authenticated encryption. The extra overhead of the crypto layer is 0.09 ms per update exchange. The framework is evaluated using a clinician-reviewed synthetic ICU monitoring dataset distributed across three non-IID hospital clients and externally benchmarked using a PhysioNet-based critical-care dataset constructed from shared physiological features. The prediction task uses five status categories: normal, mild, moderate, critical, and an outlier/anomaly category; these labels represent synthetic severity-status classes rather than prospectively validated ICU endpoints. Under a targeted label-flipping attack in which one client relabels critical cases as normal, the proposed PBF strategy achieves 86.79% accuracy, 0.782 macro-F1, 0.904 critical-class recall, 0.837 critical-class F1-score, 0.947 macro-AUROC, and 0.799 macro-AUPRC on the synthetic ICU evaluation. External benchmarking at [Formula: see text] achieves 98.04% accuracy, 0.959 macro-F1, 0.977 critical-class recall, 0.987 critical-class F1-score, 0.999 macro-AUROC, and 0.983 macro-AUPRC. Comparative evaluation against FedAvg, Krum, Trimmed Mean, and Coordinate Median shows that PBF provides stronger protection of critical-class recall under adversarial conditions, although full prospective clinical validation remains outside the scope of this study. Overall, the results indicate that a federated TinyML architecture with lightweight patient-state tracking, validation-based ensemble filtering, differential privacy, and post-quantum-secure communication can support privacy-aware and attack-resilient ICU monitoring experiments in resource-constrained IoMT settings. The study should be interpreted as a technical feasibility and robustness evaluation rather than a fully integrated clinical Digital Twin deployment.
The growing use of interconnected and digital systems in clinical settings has increased the necessity of smart and robust protection systems that can assume extremely rigid privacy and reliability requirements. This paper presents the GuardianMesh: Anomaly-Resilient Federated Orchestration (GM-ARFO) a new AI-based threat prevention model that can be used to provide security to the world of distributed healthcare information ecosystems. The method proposed will allows collaborative intelligence between heterogeneous medical nodes and does not present sensitive patient information or centralised control. GuardianMesh (GM) works by using local clinical and system cues to create compact privacy preserving representations in the edge and then a detection of anomalous behaviors is possible early on. These depictions are jointly trained in an effective federated orchestration system that is resilient to adversarial manipulation and communication inefficiently. A dec-layer adjudication layer is what is used to package distributed evidence of anomalies to facilitate swift and automatic response procedures with have minimum impact to clinical processes. Moreover, adaptive monitoring adapts to behavior drift and the changing attack plans all the time, ensuring the reliability of detection over a long period. Thorough tests in various conditions of operation and adversary show that GM-ARFO has a high level of detection, low false alarms, lower response time in addition to maintaining data confidentiality. The findings support the fact that the suggested GuardianMesh framework offers a scalable, future-restaurant, and privacy-aware platform of ensuring the safety of next-generation healthcare information infrastructures. The suggested method attains an overall detection accuracy of 96.8%, indicating a highly dependable identification of anomalous behaviours in remote healthcare systems.
Ramgopal Kashyap, Vrince Vimal, Vikalp Sharma et al.· 2026 International Conferenc...· 0 citations
The rapid integration of Internet of Things (IoT) in the healthcare domain has led to the emergence of the Internet of Medical Things (IoMT), which introduces significant benefits in patient monitoring and real‐time medical services. However, IoMT networks are inherently vulnerable due to resource constraints, heterogeneous devices, and sensitivity of medical data. In this paper, we propose a novel federated learning‐based anomaly detection system (Fed‐ADS) designed specifically for IoMT networks. Our system leverages local training of lightweight ML models on resource‐constrained IoMT devices and employs secure model aggregation at the gateway to preserve privacy and avoid centralized data collection. To address real‐world challenges, we implement and evaluate our system on a real IoMT testbed using Raspberry Pi devices under various attack scenarios. Furthermore, we examine the impact of privacy‐preserving techniques such as differential privacy on detection accuracy and system overhead. The runtime evaluation shows that our approach achieves high detection accuracy (over 94%) with minimal CPU and memory usage (under 3%), making it suitable for practical deployment in medical environments.
Mahdi Ajdani, Maziar Asmani, Asif Ali Laghari· International Journal of Com...· 0 citations
The Internet of Medical Things (IoMT) enables remote patient monitoring, providing timely diagnostic assessments crucial for saving lives. Despite these benefits, IoMT systems face significant security challenges that endanger patient privacy and data integrity. Existing centralized machine learning (ML) and deep learning (DL) solutions, while useful, are “black boxes” that raise critical concerns about data privacy and ethical compliance, as sharing sensitive patient information in a centralized system may violate confidentiality, integrity, and availability, particularly in healthcare sectors where patient data is highly confidential. Motivated by these challenges, we propose a novel federated learning (FL) explainable deep recurrent neural network (DRNN)-augmented long short-term memory (LSTM) framework for anomaly intrusion in IoMT systems, named federated recurrent short-term memory. The proposed decentralized framework processes data on local devices, sharing only learned model weights with the central server to safeguard data integrity. Training the DRNN-LSTM presents a non-convex optimization problem that is computationally expensive. To address this issue, we propose a new Bayesian Optimization (BO) method, which models the objective function stochastically and intelligently selects the most promising hyperparameter configurations to identify the hyperparameter vector that maximizes the objective function. The Adam optimizer was implemented to address gradient descent resource issues and improve the speed of convergence with a learning rate of 0.001. We introduced Shapley Additive Explanations (XAI) to ensure security decisions are transparent, a key advantage over existing black-box approaches that helps meet ethical guidelines such as HIPAA and GDPR. The framework was evaluated on the Washington University in St. Louis Enhanced Healthcare Monitoring System (WUSTL-EHMS), as well as the new CIC-IoMT2024 (IoMT-specific) and BoT-IoT (general IoT) benchmark datasets. The results show strong detection capabilities, with high detection rates for critical attack types such as DoS with 99.94%, DoS SYN with 100%, and DoS TCP with 100%. The model also demonstrated exceptionally high precision for classes like DoS SYN, DDoS TCP, and ARP spoofing. Our proposed federated recurrent short-term memory framework achieves a robust balance between detection fidelity and computational efficiency, making it a reliable solution for intrusion detection in resource-constrained IoMT infrastructures.
Internet of Medical Things (IoMT) applications require collaborative learning across healthcare institutions while ensuring patient data privacy. Traditional centralized learning approaches require sharing sensitive medical records, increasing privacy and security risks. Federated Learning (FL) enables distributed model training by exchanging model parameters instead of raw data, but its performance is affected by client dropouts and communication failures. This paper proposes a Resilience-Enhanced Federated Learning Framework for IoMT applications that improves the reliability of collaborative learning under unstable network conditions. The framework incorporates Federated Averaging (FedAvg), resilient aggregation using historical model updates with staleness decay, and quantized model updates to reduce communication overhead. The proposed model was evaluated using a heart disease dataset distributed across multiple healthcare clients. Experimental results demonstrate that the framework achieves 95.72% accuracy while maintaining stable model convergence during client failures. The proposed approach provides a secure, privacypreserving, and fault-tolerant solution for distributed healthcare applications.
Potharam Shiva Kumar, O. Ramanaiah· International Journal of Inn...· 0 citations
Background The rapid expansion of Internet of Medical Things (IoMT) and telehealth platforms has generated vast amounts of patient data suitable for training diagnostic Artificial Intelligence (AI) models. However, strict privacy regulations (HIPAA, GDPR) and the risk of data breaches prevent the centralization of this sensitive information. While Federated Learning (FL) allows for collaborative training without sharing raw patient data, it introduces a critical “trust deficit”: central aggregators cannot verify the integrity of local model updates without inspecting the private data, leaving the system vulnerable to model poisoning and malicious actors. Methods We introduce TeleZK-FL, a privacy-preserving framework designed specifically for resource-constrained telehealth environments. Unlike existing Zero-Knowledge Proof (ZKP) systems that require high-performance computing or offload proof generation to trusted edge servers, TeleZK-FL integrates (1) Post-Training Quantization (PTQ) to compress model gradients from 32-bit floating-point to 8-bit integers, and (2) optimized Look-Up Table (LUT) arguments to generate cryptographic proofs of client-side training integrity directly on edge devices. We evaluated the framework on two clinical modalities—CheXpert (chest X-rays) and PTB-XL (12-lead ECGs)—using simulated medical edge gateways (Raspberry Pi 4) under both homogeneous and heterogeneous data distributions. Results TeleZK-FL accelerates cryptographic proof generation by approximately 25× compared to standard ZK-SNARK implementations, generating full-model proofs in approximately 84 milliseconds per client on edge hardware. It cuts the communication payload by 75% (a 4.0× reduction) while maintaining a diagnostic Area Under the Curve (AUC) of 0.877 on CheXpert and 0.891 on PTB-XL, representing a degradation of only 0.1% and 0.3% respectively compared to the unquantized baselines of 0.878 and 0.894. Conclusion TeleZK-FL establishes the feasibility of verifiable, trustless federated learning on commodity telehealth hardware. By eliminating the computational bottlenecks of server-side proof generation while incurring only 0.1%–0.3% AUC degradation, it provides an efficient, regulation-aligned method for building scalable and secure decentralized medical AI networks. We further discuss an inherent trade-off: the edge-efficient KZG-based construction is classically—not post-quantum—secure, which we position explicitly against recent lattice-based alternatives.
P. Jayaraman, R. Delhibabu· Frontiers in Digital Health· 0 citations
AI-driven Internet of Medical Things (IoMT) services increasingly rely on federated learning (FL) for privacy-preserving remote patient monitoring; however, existing frameworks often neglect the longitudinal battery sustainability required for persistent clinical care. This paper presents BattFL, an energy-sustainable federated AI framework that reframes distributed healthcare intelligence as a battery-budgeted service. By integrating explicit per-client battery evolution modeling with computation-communication-security energy decomposition, BattFL supports role-heterogeneous participation (attack-only, health-only, and hybrid) via masked multi-task learning. Furthermore, we incorporate risk-driven adaptive sensing to regulate workload intensity based on predicted clinical risk. Experimental results reveal a pronounced sustainability asymmetry: high-workload attack clients experience up to 41.2% battery depletion within five FL rounds, while health-only clients remain near initial capacity. Across extended horizons $(R=5-60)$, we identify diminishing energy-accuracy returns and participation decay, motivating the need for energy-aware stopping and orchestration mechanisms. By explicitly coupling battery dynamics, security overhead, and federated optimization, BattFL provides a systems-level foundation for secure, resilient, and long-lived AI-driven healthcare services in next-generation ubiquitous IoMT networks.
C. A. Nnadiekwe, S. Ajakwe, J. Isong et al.· International Conference on...· 0 citations