The following article requires a subscription:



(Format: HTML, PDF)

: Lung sounds were recorded from five normal male subjects during tidal breathing. Simultaneous electrocardiograms were recorded and used as index signals to generate simulated heart sounds for digital subtraction from recorded lung sounds to obtain purer lung sounds. Five random breaths from each subject were analyzed. Sound signals were band-pass filtered 25 to 1,000 Hz (antialiasing), digitized at 3,000 Hz, and then subjected to (1) direct fast Fourier transform (FFT) without filtering (NF); (2) digital high-pass filtering at 75 Hz and subsequent FFT (75 HzF); (3) adaptive filtering and subsequent FFT (AF). The FFT algorithms of all lung sounds were characterized by mean, median, and mode frequencies. The mean, median, and mode of NF were lower than those of 75 HzF (64.98 /- 4.04 versus 150.42 /- 17.49, mean /- SE, p< 0.003; 44.57 /- 2.06 versus 111.81.5.78, p < 0.0003; 36.81 /- 1.77 versus 86.16 /- 3.13, p < 0.0001) and those of AF (64.98 /- 4.04 versus 96.87 /- 11.58, p < 0.01; 44.57 /- 2.06 versus 68.23 /- 10.44, p < 0.05; 36.81 /- 1.78 versus 52.24 /- 8.97, p < 0.06). The mean, median, and mode of AF were lower than those of 75 HzF (96.87 /- 11.58 versus 150.42 /- 17.49, p < 0.02; 68.23 /- 10.44 versus 111.81 /- 5.77, p < 0.007; 52.24 /- 8.97 versus 86.16 /- 3.73, p < 0.01). The results indicated that by filtering out low frequency heart sounds, the frequency spectrum of lung sounds was moved upward. We concluded that the adaptive filtering technique is more informative than the traditional high-pass technique of filtering out heart sounds by conserving lower frequency lung sounds, hence more accurate lung sound information. This may lead to a better understanding of lung sounds and better physical diagnosis.

(C) 1989 American Thoracic Society