Answers are grounded in this paper and cite their sources.
Introduction Healthcare Internet of Things (IoT) deployments primarily rely on data–driven intrusion detection, without adequately addressing the trustworthiness of the participating devices or users. Passive intrusions that leave no evidence of their existence, such as tampering, injecting, or replaying records, therefore pose a critical risk to HIoT systems. This work proposes an efficient intrusion detection framework for sensitive systems like healthcare IoT using effective deep learning approaches to detect intrusion and offer resilience against sophisticated authentication attempts, ensuring zero–knowledge proof. Methods A postquantum authenticated trust–score module is used to gate device observations via record–level authentication, while an optional privacy–preserving proof is included for credential verification under classical assumptions. Each device observation is represented as a compact record commitment and authenticated using CRYSTALS–Dilithium (ML–DSA) signatures. Only those records which satisfy signature–verification and freshness constraints are admitted to the trust–score pipeline, while failed records are conservatively assigned TS e f = 0 for auditability. The Schnorr–style non–interactive zero–knowledge proof (NIZK) component is retained for proof–of–knowledge and privacy–related protocol functionality, but it relies on classical cryptographic assumptions and is not itself claimed to be postquantum secure. A hybrid version of the Synthetic Minority Over–Sampling Technique (SMOTE–Tomek) resampling technique is used to handle class imbalances in the dataset. Adaptive threshold tuning is enforced to handle dynamic attack scenarios. Results A comprehensive evaluation using standard healthcare IoT attack datasets has demonstrated significant improvements in various performance parameters, including detection accuracy, robustness, throughput, and cryptographic assurance, compared to other state–of–the–art models. Using the WUSTL EHMS 2020 dataset, the proposed framework has achieved strong detection performance with accuracy up to 0.937 with improved precision and sensitivity relative to baseline models. The model incurs low per–record cryptographic overhead in the evaluation environment with average signing time of 1.7 ms and verification time of 0.9 ms, over 99.2% valid admissions Discussions The results indicate that integrating deep–learning intrusion detection with record–level postquantum authentication and trust–score gating can improve detection performance while maintaining low cryptographic overhead. The framework offers a scalable and security–aware approach for protecting HIoT systems against passive record–level attacks; further evaluation across additional datasets and adversarial scenarios is required to establish broader robustness claims.
No takes yet. Share an insight, caveat, or question.
Mathew et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: