Instantiating and monitoring skeletal treatment plans.

Current emphasis on protocol-based care has stirred interest in planning research as applied to clinical medicine. However, many assumptions made in designing traditional planners and plan-execution algorithms do not hold in medical domains. The problems include the unpredictable nature of the domain, uncertainty and the variability in the utility of available actions, and the need for parallel and continuous execution of treatment plans. In the context of our research on intelligent monitoring and control, we developed an approach for plan instantiation and execution which takes advantage of readily available treatment protocols. The system, named SPIN, instantiates treatment protocols based on current context, executes plans and closed-loop control actions, monitors the execution of plans and actions, and modifies plan execution as necessitated by patient response. The system is incorporated in the Guardian system for intensive-care patient monitoring and control. We address the strengths and limitations of the representation and the execution framework and discuss how the methodology may be improved and used in clinical practice.

Intelligent systems in patient monitoring and therapy management. A survey of research projects.

Although today's advanced biomedical technology provides unsurpassed power in diagnosis, monitoring, and treatment, interpretation of vast streams of information generated by this technology often poses excessive demands on the cognitive skills of health-care personnel. In addition, storage, reduction, retrieval, processing, and presentation of information are significant challenges. These problems are most severe in critical care environments such as intensive care units (ICUs) and operating room (ORs) where many events are life-threatening and thus require immediate attention and the execution of definitive corrective actions. This article focuses on intelligent monitoring and control (IMC), or the use of artificial intelligence (AI) techniques to alleviate some of the common information management problems encountered in health-care environments. This article presents the findings of a survey of over 30 IMC projects. A major finding of the survey is that although significant advances have been made in introducing AI technology in critical care, successful examples of fielded systems are still few and far between. Widespread acceptance of these systems in critical care environments depends on a number of factors, including fruitful collaborations between clinicians and computer scientists, emphasis on evaluation studies, and easy access to clinical information.

Instantiating and monitoring treatment protocols.

This paper presents a system for protocol-based treatment planning, plan execution, and execution monitoring. The approach, named SPIN, is developed as a component of the Guardian system. Guardian is an experimental architecture for intelligent patient monitoring and control. The paper describes and illustrates SPIN in a clinical scenario.

Model-based diagnosis in intensive care monitoring: the YAQ approach.

YAQ is an ontology for model-based reasoning in physiologic domains. YAQ is based on a hybrid algebra of qualitative and numerical values, and is designed to benefit from the rich and ever-changing nature of information available in a critical care monitoring environment. The focus of the project is on diagnosis of clinical conditions, prediction of the effects of therapy, and therapy management assistance. Two models of diagnosis are implemented in YAQ: diagnosis based on associations, and model-based diagnosis. The ontology is applied to the domain of ventilator management in infants with respiratory distress syndrome (RDS). The article describes the diagnostic capabilities of YAQ, illustrates these concepts on examples taken from actual patient records, and reports the results of an evaluation of the diagnostic performance on the RDS/assisted ventilation domain model.

Model-based reasoning in biomedicine.

Significant advances have been achieved in model-based reasoning in the past decade, rendering it one of the most active branches of artificial intelligence. Research on model-based reasoning focuses on two major goals: being able to predict and explain device behavior, mostly using qualitative representations. Viewed from the same perspective, many research issues in biomedicine share the same outlook: being able to predict the behavior of a certain biological system, and being able to explain why the system behaves as it does (or will behave as it is predicted to). Qualitative modeling, with its wide margins for ambiguity, provides an excellent toolset for the simulation of biological systems which do not lend themselves easily to mathematical modeling. This article presents a review of recent advances in model-based reasoning in biomedicine.