Cloud Computing Management Architecture for Digital Health Remote Patient Monitoring

With machine learning, the remote patient monitoring (RPM) devices are no longer just remote data collection devices. In addition to data analytics, data security and systems integration are also core challenges for developers of the next generation of innovative RPM devices. This includes overcoming technological barriers on applying machine learning algorithms to patient data directly on devices and regulatory barriers on patient data privacy. To address these challenges, this study proposed a unified edge-cloud computing architecture to effectively integrate all the RPM devices in use by the individual patient. All the remote patient monitoring data are managed by edge computing, only the latent representations are uploaded to the cloud for AI-assisted decision making. The proposed model has three modules. The edge medical image module used a subspace learning model for anomalies detection and unhealthy signs and symptoms classification. The edge medical time series module used spectral residual for anomalies detection and scattering wavelet network for severity classification. The cloud telehealth management module used convolutional neural network, recurrent neural network and attention model to provide individual patient treatment plan and medicine delivery schedule. The proposed platform has been tested on various RPM devices to provide AI-based anomaly detection and symptoms classifications. The application of the proposed platform has demonstrated that the on-device training model can enable faster and more accurate diagnosis and treatment. For meso-level organizational interoperability on health information exchange, we will only transmit the latent representation instead of the patient's raw data to reduce cyberattacks and ensure confidentiality of health data.