Abstract:

Stroke has become the world's biggest cause of mortality and long-term disability, with no effective therapy. Deep learning-based techniques may beat current stroke risk prediction algorithms, but they need enormous amounts of well-labeled data. Stroke data is often transferred in tiny bits around multiple institutions due to the strong privacy protection policy in health-care systems. Furthermore, the positive and negative cases of such data are very skewed. Transfer learning may handle minor data issues by using expertise from a related topic, particularly when numerous data sources are available. We present a unique Hybrid Deep Transfer Learning-based Stroke Risk Prediction (HDTL-SRP) approach in this paper to harness the information structure from many correlated sources (i.e., external stroke data, chronic diseases data, such as hypertension and diabetes). The proposed system has been thoroughly evaluated in both synthetic and realworld contexts, and it beats the best stroke risk prediction algorithms currently available. It also demonstrates the feasibility of real-world deployment across many hospitals using 5 G/B5G infrastructure

Description:

stroke risk prediction, transfer learning, generative adversarial networks, active learning, Bayesian optimization