Unconstrained Respiration States Classification by Detecting Respiratory Cycle Using Autocorrelation

Abstract

Daily sleep monitoring is necessary to make potential patients with sleep apnea syndrome aware of their respiration state during sleep. Although it is desirable to have an unconstrained system for daily monitoring in a home environment, the amplitude of respiration measured by an unconstrained sensor varies depending on the participants’ properties and the recumbent positions. In this study, we propose an algorithm for classifying the respiration state by extracting the respiratory cycle in the signal measured by an unconstrained respiration measurement system as a feature. We confirmed that the respiratory cycle obtained using autocorrelation has different characteristics between the breathing/respiratory arrest periods. By analyzing the respiratory cycle in the biological signal, it was found that the method was resilient to amplitude changes due to differences in the participants’ properties and in the recumbent positions. As a result of cross-validation to evaluate the proposed algorithm, the evaluation indices are all high, and it is confirmed that the algorithm is resilient to variations in the participants’ properties and in the recumbent positions.