Wheeze Detection Algorithm Based on Spectrogram Analysis

Wheeze, which is a type of abnormal lung sounds, is related with pulmonary diseases. It shows a continuous sinusoidal characteristics in time domain and a significant feature of texture in spectrogram. However, currently the clinical diagnosis of wheeze mainly depends on auscultating or X-ray imaging which is subjective or harmful. This paper presents a method to detect wheeze objectively based on time-frequency analysis. Particularly, a new grouping method based on Haas effect is proposed, then the new feature, P_{suspected wheezes}/P_total, is presented in this paper to judge the signals. The simulation result shows that the accuracy (AC) is 90.17%, the sensibility (SE) is 90.00%, the specificity (SP) is 90.48%, and the positive predictive value (PPV) is 94.03%, which manifests that this method could be an efficient way to detect wheeze.