A Trust-Aware Hybrid Unsupervised Framework for Robust ECG Anomaly Detection in IoMT Monitoring Networks

A reliable anomaly detection mechanism is paramount for ensuring trustworthy electrocardiogram (ECG) monitoring within the Internet of Medical Things (IoMT) ecosystems, where diverse noise sources and acquisition variability frequently compromise signals. This study introduces a hybrid unsupervised framework that integrates a Long Short-Term Memory (LSTM) autoencoder with an Isolation Forest to enable robust signal-trust assessment under unlabeled and noisy conditions. The proposed methodology synergizes temporal reconstruction learning with distribution-based outlier isolation through a trust-driven decision logic. This integration enables the explicit differentiation of true pathological abnormalities from noise-induced distortions, addressing a critical gap in conventional diagnostic systems. The framework was rigorously evaluated using the MIT-BIH Arrhythmia benchmark dataset in a single-modal, label-free learning environment. Experimental results demonstrate superior performance over both standalone architectures and contemporary state-of-the-art methods, achieving an accuracy of 99.71%, sensitivity of 99.92%, specificity of 99.46%, and an ROC-AUC of 99.96%. Ablation studies confirm that the hybrid synergy significantly enhances detection reliability compared to individual LSTM-AE or Isolation Forest implementations. Furthermore, the model exhibits a substantial reduction in both false-negative risk (FNR: 0.08) and false-alarm rate (FPR: 0.54%), thereby bolstering clinical confidence in continuous monitoring scenarios. The findings suggest that the proposed trust-based framework provides a robust and operationally efficient solution for ECG anomaly detection without requiring extensive labeled datasets, supporting sustainable long-term monitoring in real-world IoMT healthcare applications.