MediNet: Self-Supervised Framework for Multimodal Analysis and Patient Care in IoMT

Medical data analysis presents major challenges due to multimodal characteristics, real-time processing demands, and scalability constraints. Existing methods face limitations in managing data heterogeneity and generating timely outputs, which restrict their applicability in clinical environments. This study introduces MediNet, a self-supervised framework that addresses these challenges by integrating multimodal analysis, temporal monitoring, and decentralized processing within the Internet of Medical Things (IoMT). MediNet employs an adaptive self-supervised learning strategy, cross-modal contrastive learning, and edge computing to enable efficient real-time analytics. Temporal data streams are processed using dynamic monitoring modules for anomaly detection, while multimodal integration supports personalized patient care. The framework was implemented in PyTorch and evaluated on the BraTS 2021, CheXpert, ISIC 2018, and MIMIC-IV datasets using an NVIDIA A100 GPU. Experimental results show that MediNet consistently outperforms baseline models, achieving a generalization performance of 94.8% on BraTS 2021 and an inference latency of 74 ms on ISIC 2018.