The study protocol was approved by the Institutional Review Board, and informed consent was obtained from all patients or their next of kin. Thirty-eight patients (14 men, 24 women) requiring mechanical ventilatory support in the ICU were selected for the study (Table 1). Patients had a mean ± standard deviation (SD) age of 60 ± 16 years, fraction of inspired oxygen (FiO₂) of 0.41 ± 0.05, and positive end-expiratory pressure (PEEP) of 5.2 ± 0.93 cm H₂O and were mechanically ventilated for 5 ± 5 days prior to the recordings. Patients were ventilated with one of several types of ventilators: Puritan Bennett 840 (Tyco Healthcare, Mansfield, MA, USA), Servo 900 C, 300, and 300A and the Servo I (Maquet, Inc., Bridgewater, NJ, USA), and Bird 8400 ST (Bird Products Corp., Palm Springs, CA, USA). The selection of initial ventilator mode was decided by the treating physicians and support staff. The relationship between V_T and flow on lung vibration was demonstrated in four healthy volunteers.

Patients enrolled in the study were adults (18 to 84 years old) who required minimal to moderate mechanical ventilatory support (peak airway pressure of less than or equal to 30 cm H₂O, PEEP of less than or equal to 8 cm H₂O, FiO₂ of less than or equal to 0.5, and RR of less than or equal to 30 breaths per minute), who had no hypotension or severe hypertension, and whose heart rate was in the acceptable range (that is, 60 to 115 beats per minute). Patients with hemodynamic instability requiring vasopressors, chest cage or spine deformity, or skin lesions or hirsutism on the back and any patient deemed unable to be lifted to a near-sitting position with assistance were excluded. Patients judged to have conditions that would make maintenance of near-constant V_T difficult (agitation, anxiety, or unstable pulmonary status) were also excluded.

No patients were enrolled who were paralyzed or who were sedated to the point of inability to interact with the ventilator. All patients were capable of assisting the ventilator. Three patients were judged as poor candidates for stand-alone PS and were studied in VC and PC modes only. The modes used were as follows:

VC: volume-targeted, time- or patient-triggered (based on the frequency of patient respiratory effort), volume-cycled ventilation with constant flow (square/rectangular inspiratory flow waveform per protocol).

PC: pressure-targeted, time- or patient-triggered (based on the frequency of patient respiratory effort), time-cycled ventilation with variable flow (decelerating) and V_T maintained near the desired value by pressure adjustment.

PS: all breaths are pressure-targeted and patient-triggered. Flow (decelerating), volume, and inspiratory time could vary based on patient effort, and protocol targets the pressure adjustment to hold V_T near the desired value.

Because the great majority of ventilated patients included in this study were on VC at the start of the experiment, the first recordings were typically carried out on this mode, followed by PC and then PS. Three patients were unable to trigger the ventilator on PS, so no recording was carried out on this mode. Subgroups of patients who received PC or PS during the first recording (n = 3) or who were re-recorded in VC at the end of the study (n = 6) were used to assess any effect due to the lack of randomization. When switching from VC to PC and PS, the ventilator was set (pressure adjusted) to achieve the target V_T delivered in VC mode. Inspiratory time was unchanged from VC to PC and was determined by the patient on PS. V_T, FiO₂, and PEEP were held constant.

In addition to the ventilated patients, 20 recordings were performed on four non-intubated healthy volunteers at increasing V_T values (range 350 to 1,500 ml). This produced steadily increasing flow rates. V_T values were accurately measured using a CPAP (continuous positive airway pressure) mask and mechanical ventilator. RRs during recordings were kept constant. The sum of the vibration energy in the lungs during each breath cycle (inspiration and expiration) was calculated and matched with each V_T.

The recordings were performed using a VRI device (Deep Breeze Ltd., Or-Akiva, Israel) with two arrays of sensors (six rows by three columns each) or microphones similar to those used in digital stethoscopes. Each array was placed over a lung on the patient's back. The rationale for posterior imaging includes proximity to the lung and difficulty in imaging females anteriorly. To gain access to the patient's back, the patient was lifted to a near-sitting position. The recording was performed during a 20-second period, capturing up to 10 respiratory cycles. Following each recording, the suction was released but the arrays were held in place to ensure no change in array placement for subsequent recordings with different modes. A normalized dynamic image was displayed after each recording, and the raw data were stored digitally on the device for later review and analysis.

The VRI dynamic image is created from a series of gray-scale still images or frames, each of which represents 0.17 seconds of vibration energy recording. The result is a movie depicting a sense of air movement in the lungs. In addition, a graph is produced that represents the average vibration energy as a function of time throughout the respiratory cycle. Artifacts are any distortions in the image which are not related to the condition of the lungs and which are caused by extraneous noises (that is, cough, sneeze, or grunt), vibrations (that is, from stridor or the bed), or excessive motion by the patient during the recording. Artifacts are easily identified in the image, and poor-quality recordings were excluded. Overall, four patients (less than 10%) were excluded due to artifacts. Typical background ICU noise has no effect on VRI recording.

Normalization was applied to a predetermined range of frames. Within a frame, the areas with the highest vibration energy are represented as black in a gray-level scale and the areas with the lowest vibration energy are represented as light gray. Areas of a frame are white if their energy is below a signal-to-noise threshold determined by the VRI software. The software displays a video containing those normalized frames in shades of gray which reflect the intensity of vibration at each stage of the respiratory cycle. The maximal energy frame (MEF) is the frame producing the maximal geographical area of lung vibrations in the selected range of frames. In the present study, this frame was used for analysis. Figure 1 is an image from a recording of a 30-year-old, healthy, male non-smoker (video of this recording is available online as Additional file 1). Recordings are saved as both still MEF and dynamic images, which can be analyzed either as a whole or according to specific regions (left, right, upper, middle, and lower lung).

Although a very large amount of information is available within the 20-second recording, it was necessary to select a method of analysis from among various possibilities. Comparisons of MEF areas and vibration energy were preferred techniques because they provide straightforward quantification. MEFs were extracted from normal, regular, and consistent cycles available within each 20-second recording. Artifact-free MEFs were extracted a priori from these selected cycles according to predefined rules and criteria listed below. The MEF area of the VRI image was measured using the software ImageJ (National Institute of Health, Bethesda, MD, USA) 15. Regional areas were obtained by first separating the image into three regions on the basis of the rows of sensors (upper: rows 1 and 2; middle: rows 3 and 4; and lower: rows 5 and 6). Each segment was then measured with ImageJ. Because the position of the sensors was kept the same for each image recorded on a given patient, the three regions were standardized across studies.

The regional vibration energy, which is not affected by normalization of the image, was also analyzed. Vibration intensity is computed in units of energy (watts × constant), reflecting the acoustic energy associated with respiration. The vibration energy was derived from the signal at each of the 36 sensors as follows: the digitized acoustic signals were bandpass-filtered between 150 and 250 Hz to remove heart and muscle sounds; median filtering was applied to suppress impulse noise, and truncation of samples above an automatically determined signal-to-noise threshold was performed. The resulting signal was down-sampled to produce the vibration energy. The regional distribution of vibration energy was also calculated for the frames of interest (MEFs) by means of proprietary software. The percentage changes in vibration energy within the lower lung region (two lower rows of sensors), the middle lung region (two middle rows), and the upper lung region (two upper rows) were calculated and then compared among different modes of MV. The relative percentage changes within the regions of the lung were also assessed to more clearly demonstrate the shift in vibration energy and were also presented.

Frames were selected a priori from the recordings on the basis of the predefined rules and criteria listed below:

1. To correctly characterize respiratory cycles, the following criteria were applied:

- Vibration intensity is lower between two cycles (from expiration to inspiration) than within a same cycle (from inspiration to expiration).

- The distance between expiration and the next inspiration in the VRI energy graph is greater than the distance between inspiration and expiration within the same cycle.

- The area of rapidly increasing vibration from baseline indicates inspiration.

2. To correctly identify inspiration within a respiratory cycle, these criteria were applied:

- The first dramatic rise of vibration in a cycle is inspiration.

- If there is no separation between inspiration and expiration in the VRI energy graph, inspiration is considered to end at the peak signal.

- If there is more than one peak in the cycle, the first peak is considered the maximal inspiration signal.

- If there is a hint of separation in the form of a shoulder in the VRI energy graph, the shoulder is considered an inspiration.

3. These criteria were applied in choosing the maximal inspiration frame (Figure 2):

- The frame with the maximal energy within inspiration is chosen for analysis.

- If inspiration and expiration are clearly separated, the MEF during inspiration (first peak) is chosen (Figure 2a).

- If inspiration and expiration merge into one peak in the waveform, the frame closest to that peak is chosen from the image (Figure 2b).

- If inspiration and expiration form a plateau, the first frame at zero slope is chosen (Figure 2c).

- If there is no peak and the shoulder is curvilinear, the frame nearest the inflection point is chosen (Figure 2d).

4. The following criteria were applied in choosing the range for normalization of recording:

The dynamic image is produced by proprietary software and is normalized based on a chosen range of frames. The MEF at inspiration is selected for analysis.

- The chosen frame must have the highest energy in the range chosen.

- If there is a peak in the waveform, the chosen range consists of the two frames before and two frames after the peak. If this captures a frame with energy greater than the chosen frame, only frames with energy less than the chosen frame are included.

- If there is no peak and only a shoulder, the chosen range consists of the two frames before and the chosen frame.

The program SPSS (SPSS Inc., Chicago, IL, USA) was used for statistical analysis. Mean ± SD or mean ± standard error of the mean (SEM) are reported. Coefficients of determination for linear regression (R²) were obtained using Microsoft^® Office EXCEL 2003 (Microsoft Corporation, Redmond, WA, USA). The Kolmogorov-Smirnov goodness-of-fit test was used to assess the normal distribution of the samples. The Wilcoxon signed ranks test was used to analyze non-normally distributed data, and paired t tests were performed for normally distributed data. A p value of less than 0.05 was considered statistically significant.

Physiologic effects other than distribution of vibration were not ascertained nor were outcome parameters obtained. The recordings were carried out in rapid succession in order to minimize variables such as changes in patient condition and sensor placement. The inter-patient variations in vibration intensities pose potential difficulties in analyzing data from different patients. To overcome this limitation for analysis of geographic area differences and total vibration energy among the three modes, each patient served as his/her own control and relative percentage change was analyzed over the group. The reason for this large difference is not yet characterized but was not correlated with body mass index in our study patients (r = 0.002, data not shown).

In this study, because not all ventilators used were capable of all waveform selections, we did not compare VC with decelerating inspiratory flow pattern to other modes. It would be useful, however, to compare PC to VC with decelerating waveform. Also, the fact that each recording is normalized to itself may result in an area of the images of similar vibration energy values to be gray for one patient and white for another. Although using each patient as his/her own control aids in alleviating this concern, this approach does not completely eliminate the potential confounding effect of normalization. The analysis of vibration intensity shifts, however, is not influenced by normalization and supports the findings of geographical surface area, making normalization an unlikely confounding factor. The MEF, which displays the peak inspiratory vibration, was selected as the frame providing the most information on the distribution of lung vibrations and on the overall lung condition. However, whether this is the most important period to analyze distribution of vibration remains to be determined. Moreover, the use of a heterogeneous group of ventilated patients with varied diagnoses may hide much greater effects in a subset of similar patients or different effects among patient subgroups, so our conclusions are of a general nature only.

Peak flow data were not collected in this study, which is a potential confounder given that data were not constant across the three modes. If inspiratory time is the same, to deliver the same V_T would require higher peak flows at the start of inspiration in the pressure-targeted modes compared to VC. This may partly explain the higher vibration intensity observed in PC and PS. However, the fact that this change in flow pattern led to greater vibration intensity in lower lung regions still has clinical relevance. Multiple different ventilation types were used in this study. This may have had some influence on flow pattern and intensity. Ventilator specifications available make integration of this type of information into analyses difficult. However, common to all ventilators would be basic tenets of differences and similarities in flow 'patterns' among these three modes.

For technical reasons (need for vacuum sensors that must be free of contact with other objects), the recordings were carried out in the sitting position and would be expected to be different in the supine or intermediate (30° to 45°) position due to shift in fluid and gravity effect in blood flow. However, changes in vibration energy distribution between different modes of MV in the sitting position still have physiologic relevance and are of interest and potential clinical importance.