Time-Frequency Representations for Continuous Adventitious Lung Sounds

Four different time-frequency representation (TFR) techniques were tested on synthetic wheezes to determine the best one in order to analyze the time-frequency course of this kind of continuous adventitious lung sounds (LS). The best TFR may be help to gain a deeper understanding of the genesis of the wheeze and its relation with lung diseases. In this study, the Spectrogram was included as the classical analysis tool in the field but it has disadvantages when working with nonstationary signals. We also include the Reassigned Spectrogram, the TFR obtained via the Time-Varying Autoregressive Modeling (TVAR), and the more recently developed the Hilbert-Huang Spectrum based on Empirical Mode Decomposition (EMD). Since the theoretical TFR of a synthetic wheeze is known beforehand, performance measurements can be obtained and used to select the appropriate TFR. Performance indexes are based on both the TFR image as well as signal approaches. According to the performance indexes, the Hilbert-Huang Spectrum was the best TFR with ρ equals to 0.9247, ρ mean of 0.9521, NRMSE of 0.0601 and resTF of 2.47×10− 6 . Finally, the Hilbert-Huang Spectrum (HHS) was applied to real wheezes acquired from diffuse interstitial pneumonia patients. TFR with both HHS and Spectrogram were contrasted to point out the meaningful differences.