On Using Causal Knowledge to Recognize Vital Signals: Knowledge-Based Interpretation of Arrhythmias

Abstract

We have proposed a knowledge-based system for the recognition of time varying vital signals, such as electrocardiograms. This paper discusses its causal model approach. A prototype system exhibits the efficacy of the method of knowledge base stratification, where each knowledge base (KB) represents a distinct perspective to the phenomenon, such as the observable waveform knowledge and the causal knowledge of the physiological entity. Projection links in our frame representation relate corresponding concepts in different KBs, e.g., abnormalities in shape or temporality are mapped into abnormalities in physiological causality. The role of projections in the recognition process is to transduce established waveform hypotheses into corresponding event hypotheses and to form more global hypotheses using the causal model of entity. Several types of one-shot causal links have been introduced to represent causal relationships among underlying physiological events. A causal link includes the existential dependency and the implicit temporal constraints between the effect and the cause events. Using the causal knowledge with event statistics, the recognition system makes expectations for unseen events in relation to already-observed events when partial input information is given. Statistical information defined coherently through metaclasses of the PSN language supports a default reasoning process. The overall recognition framework is based on the hypothesize-and-test paradigm and the specializationand-aggregation of hypotheses using similarity links in IS-A hierarchies and causal links in PART-OF hierarchies. 1