Trust-Aware Adaptive Searchable Encryption Using Federated Graph Neural Networks for Secure Healthcare Blockchain Systems
Abstract
Hospital insiders such as doctors, patients, and administrative staff who access records without clinical justification are hard to detect under fixed access rules. Doctors, patients, and bedside medical devices form relationships that flat machine learning features do not capture, and clinics cannot easily share raw data. TA-ASE-FGNN combines federated graph neural networks (GNNs) with adaptive searchable encryption and Hyperledger Fabric audit logs. Each clinic builds a heterogeneous trust graph from consultations, prescriptions, and Internet of Medical Things (IoMT) monitoring, and a federated GNN estimates trust locally without sharing patient sequence data. Trust scores set the Advanced Encryption Standard (AES) key strength and keyword search scope. On a synthetic benchmark and on MIMIC-III, the F1 scores were 0.99 and 0.92, above those of a centralized graph convolutional network (GCN). Adaptive encryption reduced the insider detection latency from 18.4 h to 6.8 h. Federated averaging (FedAvg) gave consistent agreement across sites.