Respiratory onset detection using variance fractal dimension

Recently a non-invasive acoustical method has been developed to detect respiratory phases without airflow measurement, in which the average power of tracheal breath sounds is used to detect the onset of breaths. We improved the accuracy of the breath onsets detection by applying variance fractal dimension D_σ. For the sake of a comparison, the same set of data was used. Data included tracheal breath sound recorded simultaneously with airflow from nine healthy subjects. Variance fractal dimension was used to detect the onset of breaths directly from the time domain tracheal sound signals. Result shows that onsets can be detected by the peaks of the variance fractal dimension, with an accuracy of 40±9 ms. Comparing to the accuracy reported in the previous method (41.5±34.7 ms), this study slightly improves the average error but also is more robust in term of standard deviation. It also provides an alternative approach to analyze breath sound signals in time domain. The result increases the reliability of acoustical phase detection algorithm and paves the way for further analysis such as actual amount of airflow estimation.