Temporal Pattern Matching Using Fuzzy Templates

摘要

The identification of temporal patterns plays an important role in many medical diagnostic applications. Template systems that identify events and landmark points directly from time-series information have been shown to work well in various applications and in various forms. However, few such systems directly account for the uncertainty and vagueness often associated with medical decision-making. This paper describes a template system that uses fuzzy set theory to provide a consistent mechanism of accounting for uncertainty in the existence of events, as well as vagueness in their starting times and durations. Fuzzy set theory allows the creation of fuzzy templates from linguistic rules. The fuzzy template system that is introduced in this paper can accommodate multiple time signals, relative or absolute trends, and obviates the need to also design a regression formula for pattern matching (a requirement in non-fuzzy template systems)—the system automatically generates a normalised 'goodness of fit’ score. Our target application for the fuzzy template system is anaesthesia monitoring. Initial testing using both simulated and recorded patient data has been encouraging. Results are presented showing the diagnosis using various temporal relationships of a number of problems.