TAWB-SFL: Trust-Aware Blockchain-Orchestrated Split Federated Learning for Intrusion Detection in 6G Healthcare IoT

The ongoing shift toward 6G-enabled healthcare IoT is making privacy-preserving collaborative intelligence a practical requirement rather than a research luxury. Federated Learning (FL) is a natural fit for this setting, yet it remains vulnerable when untrusted devices participate in training, as attackers can inject poisoned updates, distort aggregation, or increase the risk of sensitive information leakage. In this paper, we present TAWB-SFL, a trust-aware framework that combines Split Federated Learning (SFL), blockchain, and edge intelligence to build a secure intrusion detection pipeline for healthcare IoT. The proposed design splits the deep model across the device and edge: lightweight CNN layers on IoT nodes extract spatial patterns from patient-related data, while an edge-side BiLSTM models temporal behavior to identify attacks. SFL keeps raw data on devices, blockchain smart contracts handle device registration/authentication and provide auditable logging of update transactions, and trust scores are used to weight client contributions during aggregation. To support frequent and timely update commitments, we further adopt the TAWB-SFL consensus design to reduce latency and improve scalability. Experimental results on the ToN-IoT and IoT-Healthcare datasets achieve 99.9% detection accuracy with a consensus time of 250 ms, indicating that the proposed approach can deliver accurate threat detection while enabling low-latency, accountable model updates in 6G healthcare IoT environments.