Software To Detect Malfunctions in Anesthesia Equipment

Lyndon B. Johnson Space Center, Houston, Texas

To reduce response times and save lives in operating rooms, community trauma centers, and remote combat care facilities, a team of scientists working on behalf of Lyndon B. Johnson Space Center (JSC) has developed an artificial-intelligence alarm-management software system that detects malfunctions in esophageal intubation and anesthesia machines. This system uses CLIPS [the C-Language Integrated Production System] knowledge-based rules derived from real-time data supplied by a sheep model. Although this system is one of several current alarm-management software systems, it is vastly superior to the commercial software systems designed to perform the same or similar functions.

Commercial software systems of this type monitor faults that occur above or below preset thresholds. Threshold settings may vary, but the monitors are unable to detect faults "intelligently." They are, in effect, hampered by preset values and can mistake threshold issues for life-endangering malfunctions. Because of this possibility of error, an alarm must always be treated as a serious matter, whether or not there is a malfunction. In an operating room, when an alarm warns of a possible malfunction, the surgical procedure can be disrupted. Without intelligent software already in place, surgical team members must verify that a malfunction has indeed occurred, or will occur, thus increasing both their own response time and the risk to the patient. A change is needed.

An artificial-intelligence software system that can detect and define faults accurately while it monitors patients on esophageal intubation and anesthesia machines would decrease response times and reduce risks to patients. CLIPS, the artificial-intelligence software designed at JSC, is a decision-support tool that can effect this change because CLIPS can be taught to determine the parameters of machines.

Basically, an artificial neural network implemented in software in a computer mimics the thought processes of the human brain. CLIPS follows pathways of reasoning, very like human thought processes, to eliminate possibilities. To support CLIPS, data are needed to design a program unique to the esophageal-intubation and anesthesia-machine environment; it is from these data that intelligent software can learn. The task for the team of scientists was to gather these data. For them, gathering data was an integral part of the development of the needed artificial-intelligence software system. The scientists selected a sheep model from which to elicit real-time data; from this model, the team compiled clinical annotations to validate their software system — all without risk to human patients.

Other artificially intelligent systems fail to provide similar annotations. For example, one prototype simulator that enables users to control and monitor an anesthesia delivery system resembles the JSC software. However, because all of its data and faults are simulated, it lacks the validation of real-time data. In short, no animal-model testing was done for this prototype. Thus, the major limitation of this and other currently available systems is the impossibility of validating them in environments in which they can be used without interfering with normal routines or endangering patients. Yet the user-interface design is critical, and artificial-intelligence software can learn from the results of animal-model testing.

During the team effort, advances in artificial-intelligence software were applied to develop an expert system that could provide, to medical personnel, decision support on the condition of the patient and the anesthesia machine. The rules of the knowledge-based domain were defined by data extracted from various sources, including supporting literature, medical experts, and the sheep model. The prototype software that emerged from this process is still in an early stage of development; it will be tested further by users in a laboratory. The data generated by the sheep model will be analyzed and, if the analysis warrants, the design of the user interface will be modified. The inventors believe that more data should be collected, and hope to test 20 to 30 additional sheep to refine the quality of the expert-system rules being written. This would enable the validation of the software in real time to determine sensitivity and specificity in recognizing alarm conditions.

CLIPS-based software can detect malfunctions in esophageal intubation and anesthesia machines because the software uses the sheep model to provide real-time data from which knowledge-based rules are built. Because the reported malfunctions will be genuine, this system will reduce response times. The use of the real-time data from the sheep model sets this software system apart from other, commercially available artificial-intelligence software systems of this type. Once the sheep model has been fully demonstrated, the scientists anticipate that the system will be tested on human patients — in a way that does not interfere with patient care and adheres to standard medical practices.

This work was done by Todd T. Schlegel of Johnson Space Center and Karin C. Loftin of KRUG Life Sciences, Inc.; Travis A. Moebes of Science Applications International; Jurine Adolf of Lockheed Martin; Donald J. Deyo of the University of Texas Medical Branch; and Jeffrey M. Feldman of the Albert Einstein Medical Center. MSC-22675