This Small Business Innovation Research (SBIR) Phase I project proposes to research improvements in rule engine technologies that enable the development of autonomous "smart systems" that can identify when and how they must adapt to unexpected environmental conditions. Reasoning technologies are important in a variety of commercial systems providing valuable control, monitoring, and data analysis capabilities. Within healthcare, they provide indispensable services for patient monitoring, disease surveillance, and decision support, etc., allowing such systems to operate reliably and predictably. Nevertheless, real-world environments are characterized by unexpected events that autonomous systems find difficult to recognize and compensate for without human assistance. These challenges suggest the need for a high-level Management Component capable of identifying when operational conditions change and when the system must re-configure or re-train its control algorithms. We will research the development of adaptive hybrid intelligent systems combining two or more reasoning technologies into a general-purpose management framework that promises to improve an autonomous system's responsiveness and adaptability.
The broader impact/commercial potential of this project can be readily illustrated within the healthcare market where the volume of patient data now being generated cannot reliably be analyzed by any one provider. New clinical decision support technologies and products are desperately needed to process and reason over complex patient data and to assist clinicians in making appropriate decisions. If successful, the proposed hybrid intelligent system architecture is potentially applicable to many medical devices. For example, gas blenders that are currently adjusted manually to maintain adequate patient oxygen saturation could be servo controlled with algorithms that safely detect and alarm when simply increasing gas delivery is not appropriate. Furthermore, the utility of hybrid intelligent system architectures is not specific to the medical domain. Any use case where contextual awareness is required for optimal performance of a control algorithm or model could be a potential candidate. Hybrid control architectures promise to reduce the dependency that such systems have on human oversight, and if appropriately applied, could reduce human error, improve cost-effectiveness and yet still maintain quality standards.
Reasoning technologies (expert systems predictive models, optimization algorithms, etc.) are increasingly used to provide valuable control, monitoring, and data analysis capabilities. Within the healthcare domain, they provide indispensible services for anesthesia machines, patient monitoring devices, disease surveillance, and clinical decision support systems. Their use enables systems to operate reliably and predictably within a variety of complex environments, helping to improve patient safety, evidence-based clinical practice, and organizational efficiency. Nevertheless, the real-world is characterized by variance in input data and unexpected changes in the operational environment that require autonomous systems recognize when their operational context has changed and a different control model may be needed. For this NSF effort, Cognitive Medical Systems (Cognitive) extended a commercial rule-engine system (JBoss Drools) and developed a prototype framework for recognizing contextual change. This framework, hereto referred to as the Adaptive Management Framework, is a hybrid intelligent system capable of (1) identifying when changes in the operational environment make a current closed-loop control model unfit, and (2) deciding whether to re-configure/re-train the existing model or adopt an entirely new model. The engineering research in adaptive management of runtime systems was performed using a sophisticated simulation tool that emulated the complex interactions between a closed-loop mechanical ventilator, a respiratory disease, and an intubated patient’s physiologic condition as measured by arterial blood gases. The tool allowed us to simulate realistic changes in the autonomous system’s operational context and develop a platform that could respond appropriately. The benefits of using such adaptive management frameworks to solve complex real-world control problems have implications not only for healthcare, but also for a number of other industries that employ closed-loop devices or systems.
