Automatic classification of inspiratory flow limitation assessed non-invasively during sleep

Detection of inspiratory flow limitation (IFL) is being recognized of increasing importance in order to diagnose pathologies related to sleep disordered breathing. Currently, IFL is usually identified with the help of invasive esophageal pressure measurement, still considered the gold-standard reference to assess respiratory effort. But the invasiveness of esophageal pressure measurement and its impact on sleep discourages its use in clinical routine. In this study, a new noninvasive automatic system is proposed for objective IFL classification. First, an automatic annotation system for IFL based on pressure/flow relationship was developed. Then, classifiers (Support Vector Machines and adaboost classifiers) were trained with these gold-standard references in order to objectively classify breaths non-invasively, solely based on the breaths' flow contours. The new non-invasive automatic classification system seems to be promising, as it achieved a sensitivity of 0.92 and a specificity of 0.89, outperforming prior classification results obtained by human experts.