Classification of nasal inspiratory flow shapes by attributed finite automata

In a significant proportion of individuals, the physiologic decrease of muscle tone during sleep results in increased collapsibility of the upper respiratory airway. At peak inspiratory flow, the pharyngeal soft tissues may collapse and cause airflow limitation or even complete occlusion of the upper airway (sleep apnea). While there are plenty of methods to detect sleep apnea, only a few can be used to monitor flow limitation in sleeping individuals. Nasal prongs connected to pressure sensor provide information of the nasal airflow over time. This paper documents a method to automatically classify each nasal inspiratory pressure profile into one without flow limitation or six flow-limited ones. The recognition of the sample signals consists of three phases: preprocessing, primitive extraction, and word parsing phases. In the last one, a sequence of signal primitives is treated as a word and we test its membership in the attribute grammars constructed to the signal categories. The method gave in practical tests surprisingly high performance. Classifying 94;pc of the inspiratory profiles in agreement with the visual judgment of an expert physician, the performance of the method was considered good enough to warrant further testing in well-defined patient populations to determine the pressure profile distributions of different subject classes.