| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Division of Pulmonary Medicine, Rambam Health Care Campus, Technion-Israel
Institute of Technology, 8 Ha'Aliyah St., 35254 Haifa, Israel.
2 Department of Clinical Affairs, Deep Breeze, Or Akiva, Israel.
3 Department of Diagnostic Radiology, Clalit Health Service, Haifa and West
Galilee, Israel.
4 Department of Radiology, University of California at San Diego, La Jolla,
CA.
5 Interstitial Lung Disease Unit, Royal Brompton Hospital, London, United
Kingdom.
6 Department of Medicine, University of Massachusetts Medical School, Worcester,
MA.
Received September 14, 2007;
accepted after revision March 11, 2008.
Deep Breeze paid P. J. Friedman for interpreting the images for this study.
M. Yigla served as a paid consultant for Deep Breeze in 2002; he received
$10,000 for his services. M. Gat is an employee of Deep Breeze. J. M. Madison
received funding support from Deep Breeze for a clinical research project on
vibration response imaging technology. The equipment for the study was
supplied in part by Deep Breeze.
Abstract
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
SUBJECTS AND METHODS. Breath sounds (frequency range, 150–250 Hz) recorded from the backs of 151 healthy asymptomatic subjects (96 nonsmokers and 55 smokers) by the VRIxp device were mapped to create a sequence of 2D images. Three raters interpreted and scored the images for predefined static and dynamic features. In addition, quantitative lung data were analyzed for characteristic regional distributions.
RESULTS. The readers of the images had good inter- and intrarater agreement. Image development in 93% of the evaluations showed an inspiratory and expiratory phase with a progressive and regressive stage that developed bilaterally in a vertical and synchronized manner. Characteristic image features of the maximum energy frame included a smooth, rounded, uninterrupted contour and a planar distribution, area size, and intensity that had right–left symmetry. Quantitative lung data expressed as percentages of the total (100%) vibration energy were normally distributed with mean values (± SD) of 55% ± 6% for the left lung and 45% ± 6% for the right lung. Most of the subjects with images, quantitative lung data, or both lacking these typical features were cigarette smokers or had a history of smoking (p < 0.05).
CONCLUSION. Breath sounds in healthy asymptomatic subjects can be recorded and displayed in a dynamic series of images that have predictable and characteristic features recognizable and complemented by quantitative lung data. Identification and description of these characteristic image features in this study will facilitate future studies of vibration imaging in specific pulmonary diseases.
Keywords: breath sounds • lung function • pulmonary diseases • vibration response imaging device
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The vibration response imaging device (VRIxp, Deep Breeze) (Fig. 1A), which was designed for practical clinical applications, displays breath sounds as static and dynamic gray-scale images [4]. The visual display is complemented by quantitative data: a table showing the distribution of the total vibration energy (Fig. 1B).
|
|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The study population was composed of 198 healthy subjects who attended the general clinical unit for a routine checkup between December 2004 and July 2005. Subjects were enrolled if they had normal findings on chest radiography, spirometry, and chest auscultation on the day of the vibration response imaging recording and if they had no history of chronic lung disease, recent chest infection (< 1 month), or surgical chest procedure.
VRIxp Recordings
Recordings were performed using the VRIxp device as previously described
[4]. Briefly, 40 sensors
assembled in two planar arrays and adhered by vacuum to the posterior chest on
each side of a subject's back are used to acquire lung sound signal data over
both lungs during a single 12-second recording period (Fig.
1A,
1B,
1C). Each row of three sensors
within these two-sensor arrays is held in place by a silicone cup that is
coupled to the back of the patient by a computer-controlled low vacuum.
|
The VRIxp device acquires signal data for all 12 seconds of recording, both inspiration and expiration, and the algorithm extracts the relative intensity of breath sounds or vibrations for each lung and each lung region (i.e., upper, middle, lower). The relative intensities of breath sounds recorded in each region are then aggregated for the 12 seconds of recording and are expressed as a percentage of the total for both lungs combined; these percentages are referred to as the "quantitative data." The quantitative data values (Fig. 1A, 1B, 1C) are displayed in a manner analogous to quantitative lung scintigraphs [17, 18]. To display the images that map the locations of the detected lung sounds, the algorithm creates a series of 2D gray-scale images, a "movie," that can be viewed dynamically or frame by frame (Fig. 2). In addition, the VRIxp device displays a complementary graph that depicts the average intensity of the breath sounds versus time (Fig. 3), and the point in the respiratory cycle of the currently displayed frame is denoted on the graph. These data allow the user to visualize the periodicity of the breathing (inspiration–expira tion) and the current location of the breath cycle.
|
|
The trained raters, one pulmonologist and two radiologists, independently analyzed the dynamic images and scored each subject's images according to a defined list of features (Table 1). The raters were blinded to the subject's medical data and to the other raters' scores. Each reader rated 181 images of one complete breathing cycle, including recordings of 151 individuals and 30 randomly arranged duplicates. Images were presented in the same order to each reader. The readers were instructed to make a final rating of either normal or abnormal on the basis of individual charac teristics as follows: If at least two of the features had a flawed rating, the final assessment was characterized as abnormal; if none of the features was flawed, the final assessment was defined as normal; and if only one of the features was flawed, the final assessment was determined by consensus.
|
The raters' evaluations were coded and ana lyzed to characterize the healthy population's images. Validation of each rater's evaluation was tested by intrarater reliability and interrater agreement. Intrarater reliability was the degree to which a rater consistently evaluated the same image. The percentage of features from the 30 duplicated images that were identically evaluated by each rater was calculated.
Interrater agreement was defined as the level of consensus among the raters' evaluations. For each image and feature (Table 1), the evaluation that appeared most often across raters was counted and specified as number of agreements (Na). The maximum number of agreements for a specific feature was 3 (30 for 10 features). The minimum number of agreements for two categoric features (e.g., yes or no) was 2 and for three categoric features (e.g., right lung [R] = left lung [L]; R < L; or R > L) was 1. For all features, the minimum was 16: [(6 x 2) + (4 x 1)]. For each image and for all 10 features, the number of agreements was summed and adjusted to 0–100%.
The overall interrater agreement for all features and images was calculated
as the average percentage of agreements for 151 images using the following
equation:
 |
For each image and each feature, the rating that appeared most often across raters was analyzed, and the distribution of the features was investigated. In addition, subgroup analyses comparing women with men and smokers with nonsmokers were performed using a two-sided Fisher's exact test and unpaired Student's t tests.
Quantitative data are reported as means ± SD with corresponding 95% CIs. The spread of data and, hence, the intersubject varia tion were calculated using the coefficient of varia tion. The normality of distribution was tested by the Kolmogorov-Smirnov test.
Comparative subgroup analyses of women versus men and heavy smokers versus nonsmokers and mild smokers were performed using unpaired Student's t tests. Pack-years were calculated as the number of years of smoking multiplied by the number of cigarettes per day divided by 20. Heavy smokers were considered subjects with a history of smoking of more than 10 pack-years and nonsmokers and mild smokers, a smoking history of 10 pack-years or less. The equality of variances was calculated with the F test. Linear regression was calculated to test the influence of smoking pack-years on quantitative data values.
All statistical tests were two-sided with a 5% significance level. Analyses of data were performed with statistics software (SPSS version 11.5.1, SPSS).
Maximum-Energy-Frame Image and Pixel-Count Comparison
To quantify maximum-energy-frame area
(Table 1), one of the features
that the raters evaluated in the images, we used an automatic pixel count to
identify the areas of the left and right lungs in the frame during maximum
inspiratory energy (i.e., maximum energy frame). In this manner, we were able
to quantitatively evaluate whether there were differences in area symmetry
between the right and left lungs and to com pare the numbers with the raters'
assessments. Symmetry for the left–right sides of the
maximum-energy-frame image was defined a priori as 50% ± 10% (mean
± SD). In addition, a pixel-count analysis that took into account the
gray-scale value was used to quantify another feature from the rater
evaluation: the intensity symmetry (i.e., maximum-energy-frame intensity)
(Table 1). The percentage of
agreement and McNemar test were conducted between the qualitative evaluations
for symmetry and the pixel-count data.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
Vibration Response Imaging Recordings
The VRIxp device operators reported that 96% of the 151 subjects had an
uncomplicated recording procedure. Reported difficulties (4%) included low
vacuum (1%), subject had a hairy back (1%), subject breathed too fast (1%),
and subject moved shoulder during recording (1%). No adverse events were
reported during the study.
Image Features and Raters' Assessments
The distribution of the image features was analyzed among the study
population (Fig. 4). Most of
the healthy population (n = 134, 89%) had a final assessment of
normal. Of the subjects with a normal final assessment, 86% (115/134) had no
flawed features. In 93% (140/151) of the recordings, the dynamic image showed
an inspiratory and expiratory phase with a progressive and regressive stage
that developed bilaterally in a vertical and synchronized manner. The maximum
energy frame—that is, the image frame having the greatest amount of
detected vibration energy (usually at 
50% of the inspiration
phase)—appeared smooth and rounded and to have an uninterrupted contour.
Planar distribution, area size, and intensity of the right and left lung
images typically appeared to be approximately equal and symmetric in these
healthy subjects.
|
|
Maximum-Energy-Frame Image Analysis Compared with Pixel-Count Data
The readers' visual impression that symmetry in the maximum energy frame is
a characteristic feature of normal images was ascertained by automatic pixel
counts, which confirmed that approximate right–left symmetries for area
and intensity do exist. The raters reported that in 91% of the images
(135/149) there was symmetry between the area of the right and left lungs in
the maximum energy frame (Fig.
6). The pixel count confirmed that indeed there was symmetry in
this group (95%, 128/135). There was high agreement (87%) between qualitative
and pixel-count maximum-energy-frame assessments and no significant
differences (p = 0.2636, McNemar test). Similar findings were
observed for the maximum-energy-frame intensity, with 93% and 95% of the
images showing symmetry by qualitative and pixel-count assessments (based on
pixel count weighted by gray-scale value), respectively. Again there was high
agreement (90%) and no significant differences between the assessments
(p = 0.6056, McNemar test).
|
Quantitative Data Analysis
Although image analysis and pixel counts of the maximum energy frame both
showed right–left symmetry, an analysis of the quantitative data for the
entire inspiratory and expiratory cycle revealed subtle asymmetry not detected
by readers. The quantitative data for the left lung and those for the right
lung were normally distributed (p > 0.05, Kolmogorov-Smirnov) but
were significantly different from each other, with mean total vibration energy
values of 55% ± 6% and 45% ± 6% (95% CI, 54–56% and
44–46%), respectively.
The quantitative data for the left lung were dominant (> 50%) compared with the right lung in 81% of the cases. The mean values for each lung were nearly identical to the median values (Table 3). Intersubject variation calculated for the left lung was 11%. The quantitative data values of the total left lung and of the upper, middle, and lower left lung regions were all significantly higher (p < 0.05, paired Student's t test) than those for the corresponding regions in the right lung. An analysis of the regional quantitative data showed similar mean and median values for the ipsilateral middle and lower regions, whereas the quantitative data value of the upper lung region was less than half those of the middle and lower regions (Table 3).
|
Subpopulation Analysis
Effect of sex—A subpopulation analysis showed no significant
effect of subject sex on the final assessments or scores of the images
(p > 0.05). However, again, quantitative data showed small but
significant differences that were not discerned by the readers. The left lung
quantitative data were significantly higher in female subjects (mean ±
SD, 58% ± 5%; CI, 56–59%) than in male subjects (54% ± 6%;
CI, 53–55%) (p < 0.05, unpaired Student's t test).
In addition, a significant difference between men and women was seen for upper
right, lower right, middle left, and lower left quantitative data values
(p < 0.05, unpaired Student's t test). The intersubject
variability (coefficient of variation) for the left lung was 11% for women and
9% for men. There was no significant difference between the variances of men
and women for quantitative data of either lung or for any of the six lung
regions (p > 0.05).
Effect of smoking—The readings of images showed a significant difference between nonsmokers and smokers (Fig. 7) (p < 0.05). Compared with nonsmokers, smokers were more likely to have dynamic images with the quality judged as disturbed (p < 0.05) and with asymmetries of image size distribution and intensities between the left and right lungs (p < 0.05). No significant differences between smokers and nonsmokers were detected in rates for image frame-by-frame development or for the shape of or missing parts in the maximum energy frame (p > 0.05). Among the subjects with an abnormal final assessment, smokers had a significantly higher rate of abnormalities on dynamic images than nonsmokers. Smokers had abnormalities seen on both dynamic image and static images, whereas nonsmokers had abnormalities seen mainly on static images.
|
|
Pack-years of cigarette smoking had a significant negative correlation with left lung quantitative data (p < 0.05; root mean square, 5.8%). Every 12.5 pack-years of cigarette smoking was associated with a decrease of 1% in the left lung quantitative data and, because it is a percentage, a corresponding 1% increase in the right lung quantitative data. The intersubject variability (coefficient of variation) of the left lung was 11% for heavy smokers and for nonsmokers and mild smokers alike, and there was no difference between the variances of heavy smokers and of nonsmokers and mild smokers for quantitative data of the left and right lungs or for quantitative data of different lung regions.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The main objectives of the present study were to describe the typical features of vibration response images and quantitative data recorded from healthy subjects and to assess the degree to which trained raters could detect and agree about the presence or absence of these features. For this purpose, we used an evaluation form based on findings from preliminary studies [15, 19]. Raters were trained on the normal image dynamics and static maximum-energy-frame features (shape, area, and gray-scale intensity) as described in Figure 4 and Table 1. The results showed that most images had typical dynamic features and maximum-energy-frame features that are characterized by a predictable shape, bilateral symmetry of intensity and area, and synchronized development of the images bilaterally (Fig. 4). Interrater and intrarater reliability scores for the assessment of images were best characterized as substantial. A given rater judged a given image the same 79–87% of the time, and between different raters the overall agree ment was 84%.
Pixel-by-pixel analysis can be used to quantify gray-scale distribution in images. In the present study, pixel-count analysis of one frame, the maximum energy frame, supported the readers' visual impressions that right–left symmetries of area and intensity during inspiration are characteristic features in normal images. Quantitative data that quantified the percentage of total breath sound intensity, averaged for the 12 seconds of recording, for different regions of the right and left lungs proved complementary to visual analysis of the images. A subtle but potentially important difference between the right and left lungs was evident by quantitative data averaged for both inspiration and expiration but not by visual inspection of images that focused on differences during inspiration only; specifically, dominance of the left lung appears to be a characteristic feature of quantitative data values in healthy subjects. With only approximately 10% intersubject variation, the left lung had between 54% and 56% of total breath sounds, which is significantly different from the 44–46% detected over the right lung. This higher breath sound intensity of the left lung compared with the right lung is a finding consistent with prior studies of posteriorly recorded lung sounds [2, 21]. An explanation for this phenomenon may be that the left mainstem bronchus is smaller, angulating more sharply at the carina, and that some major bronchi of the left lung are closer to the posterior chest wall because of the anterior position of the heart [2]. Additional future studies will be needed to determine whether quantitative data measurements can be corrected or adjust ed to predictably reflect regional lung function.
The difference in quantitative data between the right and left lungs was significantly more pronounced in women, with left lung quantitative data values of 56–59%, than in men, with values of 53–55%. This finding agrees well with the results of studies by Gavriely et al. [12] and Gross et al. [22], who reported differences in the lung sounds of healthy men and women. The mechanisms underlying this effect of sex remain to be investigated. However, sex differences in thoracic dimensions, thoracic configuration, and the curvature of the diaphragm are a few qualities that may influence the distribution of lung sounds [23]. This finding suggests that the normal range for quantitative data values should be sex-specific.
The present study extends the scope of previous breath sound measurements in two significant aspects: First, inspiratory and expiratory signals were picked up simultaneously at multiple locations on the posterior chest wall, leading to improved characterization of breath sound patterns in different regions of the lungs; and, second, technical advances in presentation of the data have been incorporated that permit visualization of breath sound distribution patterns in a real-time dynamic image complemented by quantitative data.
Even though all the subjects were healthy, the raters did identify some images as showing abnormal findings (11%), and these abnormalities were mainly in maximum-energyframe size, shape, and intensity. Those results agree well with those of a previous study showing that 8% of healthy subjects have abnormal phonopneumographic findings [24]. Thus, based on our findings and in agreement with those of others, it seems that even among healthy subjects, a small minority may have abnormal lung sound findings. The good intrarater reliability and interrater agreement about these apparent abnormalities strengthen the validity of the raters' overall assessments and descriptions of image features.
A subgroup analysis was performed to assess why 11% of the images from healthy subjects were judged abnormal in the final overall assessment. That analysis revealed a significantly higher proportion of smokers compared with nonsmokers had images rated as abnormal. These abnormalities included flawed static and dynamic features. In addition, the quantitative data supported these visual assessment results. The small but significant difference between breath sound intensity in the right versus left lung is lost in heavy smokers. Our findings that the images and quantitative data of smokers differ from healthy nonsmoking subjects are consistent with those of Ploysongsang [25]. In that study, phase differences in breath sounds between horizontal points of the lung in smokers and nonsmokers showed that smokers tended to be more out of phase during inspiration than nonsmokers.
Our study further showed that the vibration response images of smokers differ from those of nonsmokers in the extent of the flawed features (i.e., dynamics) in both lungs during inspiration and expiration as manifested by intra- and interregional changes (i.e., quantitative data changes). These findings should be investigated further because the significance of modified breath sounds among subjects having no clinical evidence of lung abnormalities is not known. Nevertheless, increased fibrosis, smooth-muscle hypertrophy, inflammation, and increased goblet cells can be found in the lungs of smokers who have no known clinical evidence of disease [26, 27]. Such histologic changes might subtly, but detectably, alter the transmission of sound vibrations through the airways and lungs. Our findings that images and quantitative data are significantly different for otherwise healthy cigarette smokers suggest that the vibration response imaging method of recording lung sounds does, in fact, show significant promise in detecting even subtle disease changes. The localization of these differences opens a new direction for future investigation.
The results of the current study and of previous studies [1, 4–6, 11, 12, 28, 29] suggest that computerized lung sound technologies have potential for improving and enhancing the value of the basic lung sound examination. Because images from healthy subjects have predictable and recognizable features, computerized lung sound technologies might be used in the future to detect the early stages of airways disease [12], pleural effusion [4, 30], pneumonia [30], foreign-body aspiration [31], and airflow obstruction [4, 8]; to measure regional function [32]; and to monitor mechanically ventilated patients [33]. The display of breath sounds as a gray-scale map is an ideal method for presenting such results because it allows rapid reading and is easily comprehended. A limitation of the VRIxp device is that, similar to other imaging systems [34], artifacts appeared in a small number of vibration response images. These artifacts may be caused by extensive background noise or touching of sensors during recording. As the clinical utility of vibration response imaging is further explored, these artifacts will need to be characterized for improved recognition by readers.
In conclusion, the breath sounds of healthy subjects can be recorded and displayed in a dynamic series of images that have predictable and characteristic features recognizable by trained observers and complemented by quantitative data. The characteristic image features identified and described in the current study will facilitate future studies of vibration response imaging in specific pulmonary diseases.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
