Medical decision making in intensive care settings relies on complex biomedical information, collected in real time, to assist hospital staff in patient management and intervention decisions. In this context, machine learning (ML) applications applied to healthcare analytics have been shown to provide significant improvement in quality of care and overall efficiency in ICU operations. A number of challenges affect healthcare analytics, these are growing data sizes, increasingly heterogeneous data sources, data assurance, credibility and quality as well as the privacy of patient record interchange. Current approaches to healthcare analytics are predominantly reliant on desktop and cloud computing architectures. These traditional systems exhibit high-latency, high-power requirements, and are overall inefficient for advanced analytic applications. If we hope to leverage the benefits of growing data size and heterogeneity, a new hardware paradigm for real time, secure and reliable ML in the clinic is needed. A potential for great innovation in medical big data applications is to imbue more analytic capability into smaller devices that sit closer to the source of data. This so-called edge computing will be instrumental in future medical applications, where real-time, low-latency, low-power computational tools are likely to dramatically improve the feasibility of remote monitoring and intensive monitoring applications, ultimately improving health outcomes. Lucid Circuit, Inc. is developing a line of purpose-built edge-analytics processors called AstrumTM. AstrumTM is an energy-efficient runtime-adaptable processor for reliable high-performance computing and low-power applications. AstrumTM processors benefit from an adaptable compute fabric that combines runtime-reconfigurable architecture and in-silicon security features. AstrumTM processors are designed to supports evolving analytics algorithms. In this proposal, we seek to optimize state of the art ML algorithms for signal processing, data quality monitoring and safe data interchange in remote and intensive-monitoring medical applications at the edge using AstrumTM. In phase II, we seek to develop an analytics programming ecosystem within which developers can customize signal quality metrics and prediction rules at a high level of abstraction, using popular analytics programming platforms. Ultimately, the work proposed in this application will enable the development of a new generation of edge-based precision intensive care and mobile health devices, together with a programming toolkit for customization, tuning, and development of ML algorithms that best match the needs of researchers, patients, and healthcare providers.
Machine Learning (ML) applications to medical big data, have been shown to provide improved quality of care in intensive care settings. Their implementation in the clinic is currently limited by reliance on obsolete desktop and cloud computing architectures. By leveraging our experience with purpose-built AstrumTM processors, we propose to adapt state of the art analytics for computation at the edge, aiming to deliver edge-based precision intensive care analytics capabilities to a new generation of intensive care and mobile health devices.