An Interpretable Multidimensional Acoustic Physiology Map for COPD Using Digital Lung Sounds

Lung sound analysis may capture chronic obstructive pulmonary disease (COPD) related physiology, but many methods are hard to interpret clinically. We developed a multidimensional acoustic physiology map using 4 indices from digital lung sounds: median respiratory frequency (MRF), long-term energy index (LTEI), subharmonic index (SubH), and harmonicity deviation index (HDI).

In this single-center retrospective study, 235 adults were classified as Healthy (n=62), Stable COPD (n=85), or COPD Exacerbation (n=88). We analyzed 1403 posterior thoracic 15-second recordings. Between-group differences in the 4 indices were tested with the Kruskal–Wallis test; when significant, Dunn posthoc pairwise comparisons were performed with Holm adjustment. Multidimensional separation was evaluated in the z-scored 4-index space using principal component analysis and permutational multivariate analysis of variance. All tests were 2-sided with a P-value < 0.05 considered significant.

Overall group differences were significant for MRF (P-value =1.36×10^-5), SubH (P-value =3.29×10^-6), and HDI (P-value =4.74×10^-8), whereas LTEI did not show a statistically significant overall effect (P-value =0.086). Posthoc analyses indicated that MRF and SubH primarily separated Stable COPD from both Healthy and COPD Exacerbation, while HDI primarily separated COPD Exacerbation from Healthy and Stable COPD. Group distributions were visualized with triangular heatmaps and summarized in a 3-axis model.

Complementary acoustic indices reflect distinct domains of COPD-related sound generation and transmission. Although standalone classification performance was limited, the proposed map provides an interpretable framework for digital auscultation phenotyping and future composite scoring.