The fractality of lung sounds: A comparison of three waveform fractal dimension algorithms

Abstract

Abstract Three waveform fractal dimension (FD) analyses of flow-specific lung sounds (LS) have been compared to examine the fractal nature of these signals. LS were recorded from five females, ages 24–40 years, breathing at flows between 0.353 ± 0.063 and 2.723 ± 0.331 L/s (mean ± standard deviation) within a single recording, with simultaneous flow measurement (flow sensor with mouthpiece and nose clip). FDs of time-domain LS were calculated within three running windows of lengths 200, 100, and 50 ms. Three techniques for waveform FD calculation were applied, based on: (1) signal variance; (2) non-normalized signal morphology; and (3) signal morphology normalized along both axes within the windows. Sensitivity to y- and x-axis scalings was evaluated by comparing minimum and maximum FD values within flow-plateau portions of LS with flow and LS intensity (RMS). Minimum and maximum values of both the normalized morphology-based FDs and variance FDs increased with window length, the variance FDs more dramatically. Maximum values of the non-normalized morphology-based FDs decreased overall (no trend for minimum values). With LS amplitude, all FDs increased, though the FD method incorporating morphological normalization exhibited the least increase. Hence, this method may prove useful in the measurement of true changes in LS fractality and deciphering differences between LS in health and disease.