Smart healthcare systems have been widely applied in the fields of intelligent healthcare, self-monitoring, diagnosis, and emergency. In recent years, there have been growing concerns regarding the privacy of the data collected from the users of the smart healthcare systems. This chapter proposes a light-weight federated learning framework based on multi-key homomorphic encryption for deploying predictive models trained on patient data distributed across multiple healthcare institutions without exchanging them. Two predictive models based on the proposed framework are deployed for in-house mortality prediction from patient data and COVID-19 detection from chest x-ray images. Performance evaluation of these models with standard datasets and comparative analyses show that the proposed models are superior to state-of-the-art approaches. The proposed framework and the models are potential solutions to improve the quality of healthcare across multiple healthcare institutions, protecting the sensitive patient data and ensuring personalization of healthcare.
TopIntroduction
According to Piran and Suh (2019), the novel 6G wireless networks feature a throughput at least three to five times higher, as well as a 10 to 100 times lower latency than that of 5G. It is a new technology whose development and deployment is underway to benefit from the rapid development of Internet of Things (IoT) technology. Such improvements in performance allow for faster data transfers for ubiquitous use of Internet-connected devices. In addition, the scope of 6G networks is expected to be expanded to support emerging technologies such as augmented reality and virtual reality, and various services, including mobility and network access for vehicle-to-vehicle and vehicle-to-infrastructure communications, enhanced mobile broadband, ultra-reliable and low-latency communications, next-generation machine-type communications, and enhanced mobile IoT for machine-to-machine (M2M) communications (Amodu and Othman, 2018).
Smart healthcare applications can be realized with 6G technology because it supports a considerably higher volume of data communication, shorter latency, and higher reliability than those of the existing mobile communication networks. The 6G technologies are required to support not only infrequent and delay-tolerant but also frequent and delay-sensitive operations. For example, healthcare data collection, analysis, and reporting may be delayed when a patient is in hospital. 6G-based healthcare applications may also support M2M communications in order to improve efficiency of the healthcare process and reduce medical costs. In addition, M2M communications can offer telemedicine service, remote healthcare, emergency detection, patient monitoring, etc., and the healthcare industry is focusing on M2M communication technologies for in-hospital medical device network applications to improve the quality of healthcare services. These technologies can be used to obtain patient information and manage medical devices remotely, detect problems in emergency situations, and monitor patient conditions, thereby improving the quality of healthcare services provided to patients.
A review of requirements of data-driven models in 6G-based healthcare systems by Ray et al. (2021) show that they are able to combine many kinds of patient's health data (e.g., biometric signals, images, and continuous health parameters) urgently needed to avoid costly reactive data inflow and support health-aware decision-making. Scarpato et al. (2017) show that intelligent healthcare focuses not only on in-hospital, but also on healthcare data communication (Health IoT), personal healthcare of individual (personal health technology), and service-to-service communications. Various statistical classification techniques, and Natural Language Processing (NLP) are important enablers for 6G-based healthcare.
With the increasing number of patients and demand for novel personalized healthcare services, there is a need to transfer healthcare data for analysis to cloud-based environments. Further, stringent regulations such as the Health Insurance Portability and Accountability Act of 1996 (HIPAA) and other data protection directives have identified patient privacy as a major concern in healthcare data sharing and management (“Act”, 1996). Need for privacy-preserving schemes for sharing patient data is highly pronounced in smart healthcare systems as discussed in the work of Son et al. (2017). Light-weight privacy preserving schemes incurring low data cost and computational overheads, and few communication round trips are highly preferred to improve the quality of care.
Adhering to these requirements, this chapter proposes a novel privacy preserving approach for smart healthcare environments, leveraging Federated Learning (FL) paradigm, the application of which in wireless communications is investigated by Tran et al. (2019) in predictive modelling (Henriksen and Bechmann, 2020). FL aims to reduce the storage of the sensitive data, and processes the model with low communication and computation costs by harnessing the central computing power of the data owners. FL continuously updates the global learning model with the latest data on the local instance for each client, while the global model is not stored on a local instance.