Classification of Lung Sounds during Bronchial Provocation Using Waveform Fractal Dimensions

Lung sounds (LS) of children after bronchoconstriction should differ from baseline LS in terms of amplitude and pattern characteristics. To test these hypotheses, time-domain and fractal based analyses have been applied to LS acquired from eight children ages 9-15 y pre- and post-methacholine challenge (MCh). Change in forced expiratory volume in 1 s after MCh ranged from -4% to -37%, with change proportional to severity of bronchoconstriction. Sounds were recorded over the posterior right lower lung lobe while subjects breathed normally for 60 s with flow measurement, and during 10 s of breath hold (BH). From root-mean-square (RMS) of LS and BH signals, signal-to-noise ratio (SNR) was determined. Two fractal dimension (FD) algorithms were applied, based on signal variance and morphology. Feature vectors for 1-nearest-neighbor classification contained FD and RMS values within flow plateau ranges. Results for LS within 75-600 Hz indicate that the combination of RMS-SNR and morphology-based FD values offers better classification of bronchoconstriction with LS, relative to using RMS-SNR with variance-based FDs and RMS-SNR alone. True positive classification was 90.3%, 63.5% and 58.3% respectively, and false positive classification was 23.4%, 24.9% and 26.1% respectively. Both RMS-SNR and FD values provide useful insight into LS changes post-bronchoconstriction.