The acoustic characteristics of fine crackles predict honeycombing on high-resolution computed tomography

This study included 71 patients with ILD who received medical care at Nagasaki University Hospital (Nagasaki, Japan) between July 2015 and June 2017. All participants presented at their bilateral lower back area fine crackles as determined by chest auscultation by at least two chest physicians and underwent HRCT chest scans for various clinical indications. Exclusion criteria were severe dyspnea, an oxygen saturation < 90% during the lung sound recording, and comorbid conditions (e.g., neurological impairment, severe cognitive impairment, and pregnancy) affecting the recording of respiratory sounds. The HRCT features relating to ILD, independently reviewed by two chest physicians and one radiologist, were as follows: honeycombing, traction bronchiectasis, reticulation, and ground-glass opacity. The study protocol was approved by the Human Ethics Review Committee at Nagasaki University Hospital, and all participants provided written informed consent before enrolment.

Whole-lung CT images were obtained with two 64-detector row CT scanners (Aquilion 64, Toshiba Medical Systems, Otawara, Tochigi, Japan or Somatom Definition, Siemens Healthcare, Erlangen, Germany) using the following settings: 1.0-mm section width with 1.0-mm reconstruction interval, beam pitch 0.828, tube voltage 120 kVp, tube current volume EC, SD11 (Toshiba); beam pitch 0.9, tube voltage 120 kVp, 100–200 mA (Siemens). The images were photographed at the lung parenchymal window setting (window width 1600 HU, window level − 600 HU). Two chest physicians and one radiologist (with 23, 26, and 32 years of experience in chest CT interpretation) who were blinded to the clinical data of the patients independently performed CT findings evaluations including that for honeycombing, traction bronchiectasis, and ground-glass opacities, as defined by the ATS/ERS/JRS/ALAT clinical practice guideline for IPF diagnosis 2018 [1].

Lung sounds were recorded at two sites over the bilateral posterior base (at 6 cm below the inferior angle of the scapula and 4 cm from the paravertebral line), where we could most clearly recognize fine crackles using a hand-held stethoscope. The subjects followed onscreen instructions to breathe deeply for approximately 30 s (eight breaths; one breath consists of 2 s-inspiration and 2 s-expiration) to record the lung sounds over each side. The recording system comprised air-coupled microphones, amplifiers, an analog-to-digital converter (UA-4FX; Shizuoka, Japan), and a personal computer.

The lung sounds acquired by the microphones were amplified and subsequently digitized with 16-bit resolution at a sampling frequency of 44.1 kHz per channel. The lung sounds in the inspiratory phase were analyzed by fast Fourier analysis using a sound spectrometer (Easy-LSA; Fukuoka, Japan) as described previously [10, 11]. The recording system was calibrated using a reference sound pressure (1 kHz; 94 dB [0 dB, 20 mPa]). In order to reduce ambient background noise, the bandwidth for breath sounds was set from 100 to 2000 Hz. The sounds were displayed as a spectrogram, with frequency in Hz on the vertical axis and time on the horizontal axis. The parameters for the sound spectrum were determined by one point of the maximum frequency. The power spectrum (power intensity, with respect to frequency in Hz) is shown in Fig. 1a. The data were automatically calculated using the Easy LSA. The percentile frequencies F99, F95, and F50 below which 99, 95, and 50% of the total signal power are respectively accumulated were measured according to previous reports [12, 13]. The time-expanded waveform analysis was also used to evaluate the predefined features of fine crackles known as 2CD, IDW, LDW [14], and amplitude ratios (A2/A1, A3/A1; Fig. 1b). In accordance with the CORSA guidelines [9], a fine crackle was defined in this study by 2CD < 10 ms. The number of fine crackles was expressed as the average number per inspiratory phase, and the onset timing of fine crackles was divided into three groups (early, mid, and late period of the inspiratory phase).

Continuous variables were expressed as the median with interquartile range. Categorical variables were summarized by frequency. Nominal or ordinal data were contrasted using a Chi-squared test. The Mann-Whitney U test was performed to compare the two ILD groups, with or without honeycombing. Interobserver agreement was determined using Gwet’s AC1 statistics. The agreement by AC1 was assessed as: 0.00–0.20, poor; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, good; and 0.81–1.00, very good. Univariate and multivariate logistic regression analyses were performed to identify the acoustic features of fine crackles in the honeycombing group to distinguish them from those in the non-honeycombing group. Multivariate logistic regression analysis was performed using significant variables chosen based on the univariate logistic regression analysis. If a category contained several strongly correlated variables (e.g., the frequency values), the most representative one (e.g., F99) was used in the multivariate logistic regression analysis. With regard to the continuous variables extracted by multivariate logistic regression analysis (number of fine crackles and their F99 value), the upper left corner coordinate point of the receiver operating characteristic (ROC) curve was used to determine the optimal cutoff level for the discrimination between the presence and absence of honeycombing in HRCT. Statistical analyses were performed using the statistical software package SPSS (IBM SPSS Statistics version 21.0; Chicago, IL, USA) and R version 3. 5. 0 (2018-04-23) [15]. The level of significance was set at 0.05.

The clinical, physiological, and radiological characteristics of the 24 patients with honeycombing (honeycombing group) and the 47 patients without honeycombing (non-honeycombing group) are summarized in Table 1. The interobserver agreement on the presence of honeycombing was substantial (AC1 = 0.88, 95% confidence interval [CI], 0.79–0.97). The median age in the honeycombing group (77.0 years) was higher than that in the non-honeycombing group (67.5 years). The honeycombing group presented less frequently a radiological sign of ground-glass opacity. There were no significant differences in sex, body mass index, lung function, partial pressure of arterial oxygen (PaO₂), oxygen therapy, and biomarkers of ILD between these two groups.

The comparison of the acoustic characteristics revealed significant differences in the onset timing, number, frequency parameters (F99 and F95), and time-expanded waveform parameters (2CD and LDW) of fine crackles (Table 2). Fine crackles began mostly in the early but never in the late period in the honeycombing group, whereas in the non-honeycombing group, they began nearly evenly in all three periods. The number of fine crackles per inspiratory phase was significantly elevated in the honeycombing group compared to the non-honeycombing group (21.5 versus 8.0, P < 0.001). The frequency parameters (F99 and F95) were significantly higher in the honeycombing than in the non-honeycombing group (F99, 841 versus 689 Hz, P = 0.001; and F95, 527 versus 411 Hz, P = 0.003). This means that the fine crackles in the honeycombing group had a higher sound pitch in comparison to the non-honeycombing group. The duration times of 2CD and LDW were shorter in the honeycombing than in the non-honeycombing group (2CD, 5.5 versus 7.0 ms, P = 0.046; LDW, 1.8 versus 2.0 ms, P = 0.004). There were no significant differences in the amplitude ratios (A2/A1, A3/A1).

On multivariate logistic regression analysis, the parameters onset timing, number of fine crackles in the inspiratory phase, F99 value, LDW, and age were independently associated with honeycombing in HRCT (Table 3). Their odds ratios (ORs) with 95% CIs were as follows: onset timing, OR 10.407 (95% CI, 1.366–79.298, P = 0.024); F99 value (unit Hz = 100), OR 5.953 (95% CI, 1.221–28.317, P = 0.029); number of fine crackles in the inspiratory phase (unit number = 5), OR 4.256 (95% CI, 1.098–16.507, P = 0.036); LDW, OR 1.337 (95% CI, 0.214–8.365, P = 0.756); and age, OR 1.074 (95% CI, 0.940–1.228, P = 0.293).

Subsequently, ROC curves were created for the continuous variables extracted by the multivariate logistic regression analysis, i.e., the number of fine crackles in the inspiratory phase and the F99 value (Fig. 2). The cutoff values to discriminate between the honeycombing and the non-honeycombing groups were for the number of fine crackles in the inspiratory phase 13.2 (area under the ROC curve [AUC], 0.913; sensitivity, 95.8%; specificity, 75.6%) and for the F99 value 752 Hz (AUC, 0.911; sensitivity, 91.7%; specificity, 85.2%). The multivariate logistic regression analysis using these cutoff values demonstrated that the odds ratio for the number of fine crackles in the inspiratory phase ≧13.2 was 33.907, whereas the odds ratio for the F99 value ≧752 Hz was 19.397 (Table 4).