This renewal application for Project Years 4-6 proposes continuation of the development of new computer algorithms for reasoning about time- varying data from underlying causal models. The signals consist of two or more channels of events, each event having several properties including time of occurrence and event shape class. The models can be represented as graphs in which the nodes are digital components, some of which give rise to observable events. The arcs are physical connections associated with time delays. The application domain is the electrocardiogram, in which the P waves and QRS complexes represent two event channels, and optional intracardiac recordings contribute additional channels. The P waves and QRS complexes are the results of the all-or-nothing depolarization of the atria and ventricles, respectively. These cardiac structures are connected anatomically and functionally by a series of other structures, each associated with a characteristic conduction time. This application builds on the results of Project Years 1-2, in which an analytic approach based on a variation of the hypothesize-and-test paradigm was used with a hierarchy of models to track the events in the cardiac rhythm on an event-by-event basis. The output of the system is one or more complete causal explanations of the observed signal, expressed in the standard clinical ladder diagram format and consisting of an instantiated model and event-by-event annotation of causality based on the model. When a signal admits of more than one explanatory model, each is developed and displayed separately. The initial hypothesis, that a cardiac arrhythmia monitor constructed using knowledge-based programming methods can perform substantially better than current clinical arrhythmia monitors, has largely been confirmed. The hypothesis to be tested in this proposal is that the programming techniques being used not only can perform better qualitatively than do current clinical arrhythmia monitors, but that they can do so at a level of performance that is likely to become suitable for the clinical use. This hypothesis will be tested in a trial comparing the program with physicians at several levels of experience. This project is attractive for two reasons. First, it offers new knowledge-based algorithms for model-based reasoning about time-varying signals. This is an unsolved problem in the expert systems field. Second, current cardiac arrhythmia monitors do not perform nearly as well as do expert nurses, technicians, and physicians. The proposed project will contribute to the development of an improved generation of arrhythmia interpretation systems that should result in improved care of patients with disorders of the heart rhythm, particularly in medically underserved settings.
