Wheeze detection using fractional Hilbert transform in the time domain

The aim of this paper is to provide an alternative method to detect wheezes, using fractional Hilbert Transform in the time domain instead of traditional spectra domain. Typical waveforms of wheezes show sinusoidal morphological features. To extract the features, fractional Hilbert transforms with a series of different fractional orders have been applied to wheezes, and then a kind of texture images would be generated with dark and lightening strips interlaced with each other. The interlaced dark and lightening strips are corresponding to the sinusoidal patterns of wheezes. Then, by two following processing steps of linear projection of radon transform and local extreme value collection, features of the texture image could be expressed as a feature point in a plane. Judgments have been made according to the positions of the feature points, and the results have been compared to the judgment made by four experienced researchers via expanded-time waveform analysis. The detection accuracy was 90.5%, which validated that the method can be an efficient alternative way to detect wheezes.