Intelligent Multimodal Biomedical Signal Fusion Using IoT and Deep Learning for Early Disease Prediction

The combination of Internet of Things (IoT), changing technologies like wearable sensing, and Deep Learning offers the potential for an "always-on" and intelligent healthcare monitoring system. There are multiple existing systems, however, that require signals from individual biomedical modalities making them potentially insufficient in utilizing complementary physiological information. This paper presents an IoT-based multimodal biomedical signal fusion framework for early disease predictions based on Deep Learning. The proposed system combines wearable IoT motion sensors, electrocardiography (ECG), photoplethysmography (PPG), blood oxygen saturation (SpO₂), respiration and body temperature sensors data. We use a modality-specific convolutional neural network (CNN) encoder for every modality, combined with a bidirectional gated recurrent unit (BiGRUs) for temporal representation learning. Adaptive attention-based fusion mechanism: combines the modality specific representations while weighing the signal reliability and a missing modality will not increase the contribution to the final prediction. Cloud infrastructure is utilised for storage and model management, while edge processing system complements the capabilities for filtering, synchronisation, signal quality assessment and low latency inference. The framework is intended to forecast disease risk based on disease indicators continually derived from changing physiological patterns and not just based on snapshot measurements. The following considerations are given for the different stages of validation of the following: PTB-XL, MIMIC waveform resources, WESAD The proposed architecture aims to deliver an integrated research platform that integrates multimodal biomedical sensing, IoT communication, the edge intelligence, and deep-learning-based predictive analytics for smart healthcare monitoring.