Classification of Unconstrained Respiratory States Utilising Multidimensional Probability Distribution Based on Respiratory Frequency Information at Each Time Step

Abstract

In this study, an unconstrained respiratory state classification system is proposed at each time step to detect the symptoms of sleep apnoea. An air mattress–type pressure sensor was developed to unconstrainedly measure the respiration signal during sleep. Based on the measurements, an algorithm that can classify respiratory states by applying a multidimensional probability distribution is proposed. Two types of validity experiments were conducted. In the first experiment, it was verified whether the respiration signal could be accurately measured by the developed pressure sensor. The results showed an average absolute error of 0.3 br/min. In the second experiment, the robustness of the classification accuracy to variations in the physical characteristics of the participants and recumbent positions was verified. The results showed an average F-value of 0.83 when extreme value distribution was applied. The classification accuracy of the proposed method outperformed the simple threshold method and the authors’ previous work.