With the improved monitoring capabilities in intensive care units (ICU's) there is a growing need to provide enhanced data-processing capabilities both for better patient care and more effective use of ICU personnel. A number of biomedical monitors provide alarms (e.g. for apnea), however there is a need for a system to alert ICU personnel before emergencies occur by responding to more subtle physiological indicators. We propose to develop two microcomputer- based prototype programs to detect certain acute respiratory abnormalities in the ICU. These programs will make use of respiratory data from a multibed gas analysis and flow measurement system, as well as data from blood gas analysis. One of the programs will be based on recent developments in recursive partitioning for pattern classification. The other program will be developed using an expert system shell and criteria tables which will allow considerable flexibility in the construction of the production rules. In Phase II of the project the best features of both programs will be combined and the resulting program will be designed to run in real-time in the ICU. The commercial applications include early diagnosis of acute respiratory abnormalities and use as a teaching tool for ICU personnel.
