Abstract
Modern healthcare multimedia Internet of Things (IoT) systems are systematically integrating face video, respiratory audio, and wearable physiological signals for remote clinical monitoring. However, this evidence can be replayed, spoofed, substituted, or temporally desynchronized before reaching clinical inference. To address these risks, we propose a physiological real-time verification framework that facilitates an edge-level admissibility firewall for secure healthcare multimedia IoT. Our proposed model first verifies whether multimodal evidence is biologically consistent across video-derived rPPG, ECG–PPG timing, respiratory rhythm, SpO 2 trends, and respiratory-audio events. A controlled multimodal integrity benchmark is constructed by integrating video-physiology evidence from UBFC-rPPG and PURE, wearable and ICU waveform evidence from BIDMC and MIMIC-IV Waveform, and respiratory-audio evidence from the ICBHI 2017 Respiratory Sound Database. We have generated clean samples using physiological compatibility matching, and attacked samples are created using video replay, respiratory-audio replay, wearable spoofing, crossmodal desynchronization, physiological substitution, and respiratory-event contradiction. Experimental evaluation shows that the proposed model achieves an attacked F1-score of 0.914 ± 0.013 , AUROC of 0.956 ± 0.009 , attack detection rate of 93.4% ± 1.2% , and false clinical acceptance rate of 5.6% ± 0.9%. Our model reduces physiological integrity error to 0.084 ± 0.007. Edge analysis shows that the 20 s evidence window provides the best trade-off and requires 89.6 ± 6.9 ms inference latency and 1.24 ± 0.12 J per verification cycle. These results highlight the ability of the proposed framework to provide robust physiological evidence authentication before clinical decision-making.