Loading...
Please wait, while we are loading the content...
Similar Documents
Knowledge Systems Laboratory August 1996 Report No . KSL – 96 – 25 Guardian : Final Evaluation
| Content Provider | Semantic Scholar |
|---|---|
| Author | Larsson, Jan Eric Hayes-Roth, Barbara Gaba, David |
| Copyright Year | 1996 |
| Abstract | Guardian is a knowledge-based system for monitoring and diagnosis of post-cardiac surgery patients in an intensive-care unit. It is designed as an autonomous agent with a flexible architecture in which several different algorithms cooperate to reach diagnoses under hard realtime conditions. Guardian has been subjected to several different test experiments, and its performance has been compared to that of human physicians, with the help of a patient simulator system. The results show that a system like Guardian can indeed be a valuable companion to the medical personnel in an intensive-care unit. Introduction The Guardian project has been aimed at designing a knowledge-based system which can perform monitoring and diagnosis of patients in an intensive-care unit, using a flexible and adaptive mode of operation, based on cooperative use of several algorithms. During the first phase of the Guardian project, from 1987 to 1993, several different algorithms were developed, implemented, and tested. This resulted in a system with potential capabilities for successful intensive-care unit monitoring and diagnosis. During the second phase, from 1993 to 1996, the knowledge base was increased, tested, and validated, and the system was tested on a set of simulator scenarios. The same scenarios were given to human test subjects, and thus we were able to compare Guardian with human physicians. Our results show that a system like Guardian is clearly valuable as a decisions support tool in an ICU environment. Diagnostic Algorithms Monitoring and diagnosis of intensive-care unit patients take knowledge and skill and also demand correct actions in complex, unexpected, and time-critical situations. Therefore, we believe that this is an area were knowledge-based systems can potentially be of great value. They can hopefully increase both the safety, efficiency, and quality of the care offered. The Guardian system [8, 9] aims at supporting hospital personnel under such circumstances. The Guardian project addresses a wide range of problems in real-time intelligent monitoring and control, such as continuous acquisition and interpretation of data, diagnosis and alarming of complications and diseases, agility in reaction to acute problems, selection of treatment plans and monitoring of their execution, explanation of physiologic phenomena, and closed-loop control of physiological parameters. The system is currently aimed at intensive-care unit patients who have undergone open-heart surgery. The final version of Guardian uses the following algorithms: • An embedded simulator, with which we have performed the test experiments. • Focus, a data selection algorithm that lets Guardian focus its attention on different sets of parameters according to need, and which also reduces the amount and rates of incoming data considerably, (the simulator produces data for some 140 parameters every second; far too much for a knowledge-based system to receive in raw form). • tFPR, (temporal fuzzy pattern recognizer), an algorithm that turns quantitative parameters into discrete sign values, and avoids problems of switching near crisp limits, etc. The patterns are temporal and multivariable; thus, it is possible to define signs for, say, increasing heart rate and a sudden drop in chest tube output [3]. • ReAct, an algorithm that uses action-based hierarchies, (decision trees with actions at internal nodes), to take action in face of a serious deadline, even when there is not enough |
| File Format | PDF HTM / HTML |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
