Microservices Architecture to Improve the Performance of Machine Learning Applications in eHealth

The objective of this paper is to propose an intergenerational software architecture, designated “µe-Health 4.0,” for the integration and interconnection of traditional eHealth applications utilized within medical institutions. These applications are commonly developed in accordance with the service-oriented architecture architectural style. Furthermore, in conjunction with modern applications, the microservice architecture style has been adapted to accommodate multiple machine learning-based inference systems, which are fed by a large database that integrates diverse and heterogeneous dynamic data sources. Proper standardization of these data sources is achieved using metadata. In contrast to other proposals, our approach prioritizes the replicability, scalability, and interoperability of the specialized services that comprise this architecture, leveraging our versatile, high-performance software features. To demonstrate the feasibility of µeHealth 4.0, it was necessary to adapt this proposal to a real eHealth use case, specifically to the prediction of the occurrence of intradialytic hypotension during a hemodialysis session. These results were validated by quantitative values obtained from rigorous tests, such as response time, efficient use of the infrastructure, and network consumption. In conclusion, it can be posited that µeHealth 4.0 has the potential to create an eHealth ecosystem in conjunction with the integration of multiple machine learning-based inference systems, such as logistic regression, random forest, or XGBoost, to perform specialized decision support tasks in the prevention, monitoring, and treatment of hypotension prediction in hemodialysis. It should be noted that µeHealth 4.0 was developed with the intention of being adapted to other eHealth domains, including diabetes and colon cancer.