A post-quantum verified self-healing framework for healthcare IoT systems: deep anomaly detection, severity...

Stronger defense mechanisms are needed for the safety of healthcare IoT systems, for faster detection and smooth recovery from cyber-attacks, without disrupting services. This study addresses the need to safeguard healthcare IoT systems through efficient detection of potential anomalies leveraging the capacities of deep learning, and also rolling back the system to the safest checkpoint in the recent past, verified using post-quantum verification using ML-DSA/CRYSTALS-Dilithium. A two-tier architecture is proposed in this study with deep anomaly detection and attack severity analysis in the first tier, and post-quantum verification for selecting a policy-driven safe rollback point based on multi-factor assessment, with due consideration of risks. The reinforcement learning-based framework is used in the checkpoint selection phase that selects the checkpoint with the best Q-Score. The model is implemented over two prominent HIoT network datasets-WUSTL EHMS 2020 and CIC IoMT 2024. The model is evaluated using anomaly detection metrics (attack probability, confidence score, severity analysis, and recovery uptime), trust score, verification latency, recovery time, rollback quality, and performance metrics (F1-score, false-positive rate, precision, and recall). The proposed model achieved strong overall performance across both healthcare IoT datasets, with average values of 59.23% attack probability, 94.00% confidence, 5.61% anomaly score, 69.50% severity, 26.61 ns recovery time, 75.86% trust score, 98.37% uptime, 72.93% rollback quality, 11.28 ns verification latency, 98.37% F1-score, 98.28% recall, 98.4% precision, and 0.045% false positive rate. The proposed model effectively and accurately detects an attack and proposes the suitable recovery method-micro rollback, medium rollback, or complete rollback- based on the attack severity analysis. Better system uptime is ensured by selecting recent and trustworthy checkpoints to roll back based on post-quantum verification and combined trust score calculation.