Secure Ultra-Low Latency Data Transport for High-Speed Edge–Cloud Networks

The emergence and evolution of data-driven clinical decision support (CDS) systems have marked a rapid progress of computer-aided diagnosis and disease prognosis. But the centralized aggregation and labeling of sensitive medical data present serious challenges to patient privacy, data security, and compliance with regulation. Furthermore, insufficient labeled clinical samples might lead to low generalization ability for traditional supervised deep learning models. In order to solve above problems, in this paper we present the Privacy-Preserving Disease Diagnosis Framework Based on Self-Supervised Deep Learning (PP-SSDL). The proposed method makes use of self-supervised representation learning to learn clinically-interpretable representations from unlabeled medical data with reduced dependence on labeled samples. The approach relies on the use of contrastive and reconstruction-based pretext tasks to learn complex latent embeddings from multimodal medical data, such as images and physiological signals, while being robust to outliers. An alphanumeric vector is then fine-tuned with a slim classifier for the downstream detection of diseases. Privacy is maintained using decentralized training and encrypted updates of models, thereby enabling clinical sites to retain local copies of sensitive patient data. Experimental analysis is performed on standard medical datasets, including chest X-ray and ECG, with non-IID data distributions in a realistically sensitive setting. By contrast, the proposed PP-SSDL framework yields a mean diagnostic accuracy of 97.6% with precision, recall and$F 1$-score of$96.9 \%, 97.3 \%$, and 97.1%, respectively, demonstrating superiority to the conventional supervised DLMs by a margin ranging between [4]–[7]%. Further, compared with centralized learning methods, the model can save 65% on the labeling cost and reduce risks of privacy leakage. Our results show that self-supervised deep learning can provide an effective, scalable privacy-preserving solution to the next generation of medical diagnosis systems, especially when dealing with data-limited and privacy-sensitive healthcare regimes.