A new respiratory monitor system for four-dimensional computed tomography by measuring the pressure change on the back of body

Abstract

Objective:

A novel respiratory monitoring method based on the periodical pressure change on the patient’s back was proposed and assessed by applying to four-dimensional CT (4DCT) scanning.

Methods:

A pressure-based respiratory monitoring system is developed and validated by comparing to real-time position management (RPM) system. The pressure change and the RPM signal are compared with phase differences and correlations calculated. The 4DCT images are reconstructed by these two signals. Internal and skin artifacts due to mismatch between CT slices and respiratory phases are evaluated.

Results:

The pressure and RPM signals shows strong consistency (R = 0.68±0.19 (1SD)). The time shift is 0.26 ± 0.51 (1SD) s and the difference of breath cycle is 0.02 ± 0.17 (1SD) s. The quality of 4DCT images reconstructed by two signals is similar. For both methods, the number of patients with artifacts is eight and the maximum magnitudes of artifacts are 20 mm (internal) and 10 mm (skin). The average magnitudes are 8.8 mm (pressure) and 8.2 mm (RPM) for internal artifacts, and 5.2 mm (pressure) and 4.6 mm (RPM) for skin artifacts. The mean square gray value difference shows no significant difference (p = 0.52).

Conclusion:

The pressure signal provides qualified results for respiratory monitoring in 4DCT scanning, demonstrating its potential application for respiration monitoring in radiotherapy.

Advances in knowledge:

Pressure change on the back of body is a novel and promising method to monitor respiration in radiotherapy, which may improve treatment comfort and provide more information about respiration and body movement.

Introduction

Respiratory motion is a major concern in external beam radiation therapy treatment of tumors in lung and liver. Without proper management, such motion could cause insufficient delivery of radiation dose to the tumor and severe radiation damage to the surrounding healthy tissue.^1,2 Several strategies have been in use clinically to minimize the respiratory motion effects such as forced shallow breathing,^3,4 breath-holding techniques^5,6 and respiratory gating techniques^7,8 etc.

Respiratory gating radiotherapy (RGRT) only turns on the treatment beam during phases of respiratory cycle when the tumor is within a predefined region and is widely used to tackle this respiratory motion issue in radiotherapy.^9,10 Four-dimensional CT (4DCT) has been used to evaluate the tumor motion for the planning of treatment with or without respiratory gating.^11–13 During scanning, a respiratory monitoring system is used to measure the breathing signal and detect the patient’s respiratory phase in real time.¹⁴ By using the breathing signal, the original images acquired by the CT scanner are sorted into different phases. Either phase-based binning method or amplitude-based binning method can be applied to reconstruct these 4DCT images.^15,16 As free breathing causes uncertainty in the phase and amplitude of the respiratory signal, there are often mismatches of the CT slices between adjacent couch position, which could make artifacts in the 4DCT images.

There are several existing respiration monitoring systems in use or under study today. The real-time position management (RPM, Varian Co., Palo Alto, CA, USA) system measures the abdominal height by using a camera and infrared reflective markers placed on the abdomen.¹⁷ The Anzai belt (Anzai AZ-733V, Anzai Medical Co., Tokyo, Japan) also measures the abdominal movement by a pressure sensor.¹⁸ Additionally, some studies used the time-of-flight camera¹⁹ or radars²⁰ to measure the breathing signal.

Ballistocardiogram (BCG) is a recording of the recoil forces of body in reaction to blood ejection and is used to assess cardiac functions.²¹ It can be measured as a displacement, acceleration or pressure signal and is known to include movements in all three axes: head-to-foot, anteroposterior and left–right.^22–24 In anteroposterior measurement, a pressure sensor is usually installed on a bed right below the subjects with tight contact between the pressure sensor and the back.^25–31 As a result, the sensor can measure not only the BCG but also record the respiratory motion. Therefore, a mixed signal of BCG and respiration is captured as shown in Figure 1.

Figure 1.

(a) The raw signal recorded by the pressure sensor which consists of respiratory motion and BCG. (b) The respiratory signal filtered from the raw signal. (c) BCG signal. BCG, ballistocardiogram.

The respiratory signal has been commonly regarded as an interference of BCG and was removed by filters in cardiac applications, while such signals, as shown in Figure 1(b), could potentially be quite useful for respiratory motion management. As 4DCTs are commonly used to determine the tumor positon before treatment, it is important to know whether the respiratory signal measured by the pressure sensor could be used in 4DCT scanning. In this paper, we used such a pressure-based respiratory monitoring system (PRMS) to measure the respiratory motion and investigated its application for 4DCT scanning.

Methods and materials

PRMS description

Figure 2 is a photo of the respiration measurement system. The device could be divided into three parts: (1) a film type pressure sensor, (2) a signal processing box, and (3) a personal computer (PC) with measurement software.

Figure 2.

Photo of the pressure based respiratory monitoring system. PVDF, polyvinylidene fluoride.

A 50 cm ×1 cm polyvinylidene fluoride (PVDF) film (Jinzhouxinke Co., Beijing, China) sealed by foamed rubber was adopted as the sensor to measure the pressure change on the back of body. The sensor was attached to the vacuum mat and adjusted to the best position longitudinally where tight contact with the patient’s back could be achieved. The signal processing box could magnify, filter, convert the pressure signal from the PVDF sensor and could also monitor an I/O port for synchronizing with CT scanner. The signal was sampled at frequency of 500 Hz and sent to a PC via USB interface. A software was developed to receive, display and save the respiratory data. This system was assessed by acquiring BCG signal from healthy subjects and heart failure patients in a previous study.²²

RPM description

The RPM system is widely used in 4DCT scanning and respiratory gating. It consists of a plastic box with infrared reflecting markers, an infrared camera and a PC. The box is placed on the patient’s chest or abdomen, and the position of the infrared reflecting markers is captured by the camera. The displacement of the marker is regarded as the signal of respiration. A software on the PC can display and record the signal in real time. The sample rate of RPM system is 30 Hz.

Synchronization of respiratory signals

The “X-ray on” signal from the CT scanner was used to synchronize two breathing signals, by connecting it to both RPM system and PRMS through a “T” junction cable and a optocoupler. Thus, the “X-ray on” signal was recorded by the RPM system and the PRMS, respectively. Figure 3 depicts the experiment setups.

Figure 3.

Block diagram for the synchronization of two respiratory signals. The “X-ray on” signal is recorded by two systems and used for synchronization. 4DCT, four-dimensional CT; PC, personal computer; RPM, real-time position management.

Data acquisition

This experiment was conducted in four steps:

1. The pressure sensor is attached to the surface of the vacuum mat or localization plate.

2. A patient lies down supine on the vacuum mat or localization plate as in normal procedure. The sensor’s position could be adjusted to ensure qualified pressure signal.

3. The RPM marker box is placed between navel and xiphoid of the patient and the RPM signal should be verified before 4DCT scanning.

4. 4DCT scanning is started. Both systems record the breathing signal and the “X-ray on” signal.

In this experiment, CT images were acquired with Siemens CT scanner (SOMATOM Definition AS, Siemens, Berlin, Germany) operated in cine mode. The slice thickness was 3 mm. There is no risk of extra radiation exposure to the patients.

4DCT reconstruction

RPM signal and pressure signal were used to sort the CT slices respectively. Phase-based sorting technique was adopted to reconstruct 4DCT data set at 10 respiratory phases. As pressure signal on patients’ back might contain the interference of the heartbeat and other noises, a low-pass filter was applied to extract the respiratory waves. Phase angle was then generated from the filtered signal. All these amplitude and phase data were downsampled and reformatted to the RPM file format (.vxp file) for reconstructing the 4DCT images through pressure signals. The pressure data were processed by Matlab (v. 2014b).

Subjects’ information

The experiment is conducted at Chinese PLA general hospital under informed consent and approved by the IRB of Chinese PLA General Hospital. 28 subjects receiving 4DCT scanning participated in this experiment. For 11 patients with content to using their images, two 4DCT series were reconstructed by RPM and pressure data respectively. Among the 11 patients, 4 had liver cancer; 2 had lung cancer; 3 had cervical cancer; 2 had abdominal lymph node metastases. Among the rest 17 patients, 6 had liver cancer; 3 had lung cancer; 2 had cervical cancer; 3 had pancreatic cancer; 1 had abdominal lymph node metastases; 2 had kidney cancer.

Method of data analysis

Correlation and time shift

The Pearson’s correlation coefficient was computed between the RPM and PRMS signals as described in equation (1).

$${\displaystyle GCC=\frac{\sum _{i=1}^n(S_{rpm}(i)-\overline{S}rpm)(Sprms(i)-\overline{S}prms)}{(n-1){\sigma }_{rpm}{\sigma }_{prms}}}$$ (1)

where s _rpm and s _prms were the amplitudes of RPM and PRMS signals respectively and σ was the standard deviation. As the data set used to compute the correlation coefficient was from the time of starting to the time of ending of CT scan, the correlation coefficient could be referred as global correlation coefficient (GCC).

Methodology in Ionascu et al³² were adopted to define the time shift. The Pearson’s correlation analysis was carried out dynamically by computing the GCC as a function of a variable time shift induced on the pressure signal. The RPM signal was set as reference for maximizing the correlation coefficient. The positive time shift value indicated that the pressure signal lagged behind the RPM signal and vice versa for negative values. As shown in Figure 4, the GCC was maximized by shifting pressure signal v.s RPM signal. The time shift between RPM and pressure signals was consequently obtained as the time value where the GCC was maximized.

Figure 4.

Illustration of the method to obtain the time shift and maximum of GCC. By moving pressure signal back and forth and computing GCC at each step, the time shift is the time value where the GCC was maximized. GCC, global correlation coefficient; PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Respiratory phase comparisons

A further analysis of the time difference in each respiratory phase was conducted. Each breathing cycle was divided into 10 phase from 0 to 90%. To determine the phase angle, first the peak and valley in each breathing cycle were identified from the pressure trace manually. We designated t ₁, t ₂ and t ₃ be the times at the first peak, the following valley and the second peak respectively. The phases at these three points were defined as 0, π and 2π. The phase angle φ at time t_i was linearly interpolated as in equation (2).

$$\begin{gathered} {\displaystyle \phi =\frac{t_i-t_1}{t_2-t_1}\times \mathrm{\Pi },if{t}_1\le t_i\le t_2 \\ \phi =\mathrm{\Pi }+\frac{t_i-t_2}{t_3-t_2}\times \mathrm{\Pi },if{t}_2<t_i\le t_3} \end{gathered}$$ (2)

This process was repeated in each breathing cycle. 10 phase angles were selected with equal interval, which means 0% corresponds to 0; 10% corresponds to π/5; 20% corresponds to 2π/5, etc. The time difference at each phase angle between RPM signal and pressure signal were then computed as shown in Figure 5.

Figure 5.

Illustration of the time shift at different phase angle in each breath cycle. PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Evaluation of 4DCT images

4DCT images were reconstructed by using the RPM and pressure signals respectively and artifacts could be caused by mismatch between CT slices and respiratory signal. Inspired by the method used by Yamamoto et al,³³ internal artifacts such as mushrooms on diaphragm were evaluated in coronal view and the artifacts on abdominal skin were inspected in sagittal view and were referred as skin artifacts. Phases with artifacts and magnitudes of artifacts were recorded.

The evaluation process was shown in Figure 6. First, all the 10 phases of 4DCT images were loaded in video mode where the images were displayed periodically. Then, we scrolled through all the coronal positions while the video was playing and carefully reviewed the artifact in each coronal level. If artifacts were observed, we searched for the coronal position where the artifact is most prominent and then checked each phase manually to record the phase(s) where the artifacts exist. The magnitude of the artifact was measured in the phase where the artifact is most obvious. If no artifact was observed, then this case was labeled as none artifact. The procedures were same for the skin artifact evaluation in sagittal view. Three experienced clinical physicist conducted this evaluation without knowing how these images were reconstructed. All the evaluation was done in MIM system (v. 6.0, MIM Software Inc., ).

Figure 6.

(a) A flowchart illustrates the method to evaluate artifacts in 4DCT images. The same procedure was used for evaluation of skin artifacts in sagittal views. (b, c) Illustration of measurement of skin and internal artifacts. 4DCT, four-dimensional CT.

The mean square gray value difference (MSD)³⁴ was also adopted to measure the reconstruction quality. Artifacts usually occurred when two adjacent slices did not correspond to the same respiratory phase. Thus, these two slices showed dissimilarity which was not found in the normal anatomical change between two adjacent slices. On the contrary, high quality reconstruction was reached when two adjacent slices were as similar as anatomical change. The similarity can be measured by MSD which is defined by the equation (3).

$${\displaystyle MSD=\frac{1}{N_s}\sum _{z=1}^{N_s-1}\frac{1}{\mathrm{\Omega }}\sum _{x,y\in \mathrm{\Omega }}{\left[I\left(x,y,z\right)-I\left(x,y,z+1\right)\right]}^2}$$ (3)

Where I is the hounsfield unit, z is the couch position, Ns is the total number of couch position, Ω = [0, 511]² is the size of image. The MSD of each phase was computed first and then the mean and standard deviation of MSD over 10 phases were computed. Therefore, a larger MSD value indicated lower quality of 4DCT images.

Results

Respiratory signals

Both the pressure signal and the RPM signal could reflect the periodicity of respiration. For all 28 patients, the breathing patterns of RPM and pressure signal matched well whenever the patient breathed regularly or not as shown in (a) and (b) of Figure 7. Large excursions were observed in pressure signals of seven patients, which might be caused by muscle contraction or tiny body movement. The RPM remained normal waveform when excursions occurred as shown in (c) of Figure 7. For one patient, unqualified RPM waveforms was observed while the pressure signal was normal as shown in (d) of Figure 7.

Figure 7.

Comparison of respiratory signals. (a) regular breathing; (b) irregular breathing with two breathing cycle longer than others; (c) the example of large excursion recorded by pressure sensor; (d) the situation where RPM signal is noisy but pressure signal is normal. PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Correlation of signals

Figure 8 (a) showed the GCCs and the time shifts between RPM and PRMS signals. Across all the 28 patients, the best correlation reached 0.92 while the most irregular RPM (as shown in (d) of Figure 7) gave the worst correlation of 0.19. The largest time shift was 0.92 s and the minimum value is 0.2 s. By considering the influence of time shift, the original GCC was optimized to best GCC (Figure 8a) with average GCC improved from 0.43 ± 0.25 to 0.68 ± 0.19.

Figure 8.

(a) Plot illustrates the value and trend of GCC and time shift of each patient. (b) Plot compares the respiratory cycle length computed from RPM and pressure signals. The bar represents standard deviation over all breathing cycles. The gray background indicates the 11 patients included in the evaluation of 4DCT images. 4DCT, four-dimensional CT; GCC, global correlation coefficient; PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

The respiratory period of each patient was calculated by two breathing signals and the difference of breathing cycle length was 0.02 ± 0.17 s averagely. There was no significant difference between the periods defined by RPM and pressure signals, with the p-value of 0.41. Figure 8 (b) shows the breathing cycle length computed from these two signals.

Time shift of respiratory phase angle

The time shifts of all the 10 phase angles in each breathing cycle were calculated between the RPM signals and pressure signals. Figure 9 shows the mean and standard deviation of time shift in Phase 1, Phase 4 and Phase 8. The average value and standard deviation of time shift were close among different phase angles for a certain patient. For some patients (e.g. Patient no. 10 and 19), large standard deviations were found, which indicated that there were large differences of the time shift among different breathing cycles.

Figure 9.

Plot of the time shift at different phase angles across all patients. The bar represents tandard deviation over all breathing cycles. The gray background indicates the 11 patients included in the evaluation of 4DCT images. 4DCT, four-dimensional CT.

Evaluation of 4DCT images

Figure 10 are examples of the images reconstructed by using RPM and pressure signals. For Patient No. 1, mushroom artifacts were observed in Phase 8 for RPM reconstructed images but no artifact was found in images reconstructed by pressure signal. For the Patient No. 6 in sagittal view, the abdominal skin in images reconstructed by pressure signal was not as smooth as it was in images reconstructed by RPM signal.

Figure 10.

Examples of 4DCT images reconstructed by RPM and pressure signals. Internal mushroom artifact can be observed in the diaphragm of images reconstructed by RPM for Patient No. 1. Skin artifacts can be observed in the upper abdomen of images reconstructed by PRMS for patient No. 6. PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Figure 11 showed the average number of artifacts and counts of phases with artifacts. Four patients were found with skin artifacts and seven patients with internal artifacts for the images reconstructed by RPM signals while the numbers were five with skin artifacts and eight with internal artifacts for PRMS signals. 8 of 11 4DCT series were observed with artifacts both for RPM and PRMS signals. More phases with artifacts were observed for the images sorted by PRMS signal except Patient No.1, 3, 4 and 5. The number of artifacts and phases with artifacts showed no significant difference between RPM and PRMS.

Figure 11.

Bar charts illustrate the average number of artifacts and total phases with artifacts evaluated by three physicists. PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Figure 12 showed the magnitude of artifacts that were measured by visual inspection. The maximum internal artefact was around 20 mm for both signals, and the averages of the artifacts were 8.8 and 8.2 mm across all the patients for PRMS and RPM signals, respectively. It was similar for skin artifacts where the maximum was around 10 mm and the average was around 5 mm for both reconstruction methods. The average magnitude evaluated by three physicists showed no significant difference between RPM and PRMS with the p values of 0.54 (internal) and 0.58 (skin).

Figure 12.

Box plots illustrate the magnitudes of artifacts assessed by three physicists. PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Table 1 showed the MSD results of 10 phases for all the 11 patients. There was no significant difference between the MSD values computed from two reconstruction methods with the p-value of 0.52.

Table 1.

Table of MSD values for 11 4DCT series

Patient no.	MSDPRMS	MSDRPM
1	1191 ± 14	1239 ± 5
2	1049 ± 10	1054 ± 10
3	1551 ± 7	1556 ± 9
4	1345 ± 21	1156 ± 7
5	1424 ± 7	1450 ± 9
6	1139 ± 25	1120 ± 17
7	1552 ± 11	1556 ± 11
8	1617 ± 14	1619 ± 13
9	1266 ± 15	1264 ± 14
10	1460 ± 37	1460 ± 20
11	1271 ± 10	1254 ± 6

4DCT, four-dimensional CT; MSD, mean square gray value difference; PRMS, pressure-based respiratory monitoring system; RPM, real-time position management.

Discussion

In this paper, we investigated the possibilities whether the back pressure of patients could be a surrogate of respiration and used for 4DCT scanning. The results showed that the pressure signal was close to RPM signal and 4DCT images reconstructed by these two signals were comparable in terms the amount and severity of artifacts.

PRMS can be substitute or supplement for existing method such as RPM. PRMS is easy to install and can generate qualified respiratory signal in some particular cases such as patients with abdominal retraction or much abdominal fat when the RPM signal is usually unqualified. Other than RPM or spirometer, PRMS measures the respiratory motion imperceptibly, which can reduce the stress to patients, avoid extra body movement and improve treatment comfort. Additionally, PRMS could be used for RGRT and body position and motion monitoring during radiotherapy in the future.

In pressure signals, large excursion was observed in some patients. It was found that tiny body movement and contraction of pectoralis or dorsal muscle could generate larger pressure change than the respiration and thus brought large excursions in the signal. Compared with RPM, the PRMS is much more sensitive to tiny body movement and can offer more information about body motion during CT scanning or radiation treatment.

The correlation between RPM and pressure signals affected by the time shift obviously and the influence to the GCC was individually different. Factors such as mode of respiration (abdominal/thoracic), motion of internal organ, weight, position of sensors, etc. can be the source the time shift. The time shifts could be understood much better if the two systems could be used on a standard motion phantom with specified motion characteristics, that is, the ground truth. However, we could not yet find such a phantom that can produce similar pressure changes as a human together with the required motions for the RPM system.

It was found that 8 of 11 patients had skin or internal artifacts for images sorted by both RPM and PRMS signals and the magnitude of the artifacts by both methods are close. It is noteworthy that two patients (No. 1 and 5) had artifacts by using RPM but none by using PRMS. The MSD was also calculated to be another indicator of the artifacts. MSD was calculated by considering the whole image as shown in equation (3), but it was dependent on the body contour instead of the background as the HU value of background was zeros and made no contribution to the final result. The results from 11 patients could confirm the quality and feasibility of the PRMS method in 4DCT scanning.

There are some limitations in this study. First, the position of the pressure sensor need to be adjust to measure qualified signals and it relies on the experience of physicists, which might bring uncertainties to the signal. Second, internal motion such as diaphragm motion is better than external motion including abdominal wall motion and pressure on patient's back. Due to the artifact one the diaphragm, the relationship between internal motion and external motion was hard to investigated.

A multi sensor system could be developed in the future. By analyzing the multi channel signals, body movements, respiratory motion and heartbeats might be distinguished and provide more information for 4DCT scanning and radiotherapy. The signal processing method of PRMS signal was based on filtering and phase calculation which was post analysis in this paper. For future application in RGRT, real time processing of the PRMS signal will be necessary to provide gating signals.

Conclusion

In this paper, a pressure sensor based respiratory monitoring system was developed to provide a new approach to monitoring the respiration in 4DCT scanning. To validate this method, an experiment was designed by measuring the pressure signal and RPM signal simultaneously when patients were undergoing 4DCT scanning. The respiratory waves measured by PRMS were correlated well with waves measured by RPM. Artifacts in the 4DCT images reconstructed by both signals were evaluated and the results showed that the quality of images were similar. It can be proposed that the pressure measured on the patient’s back is a qualified surrogate of respiration. This method can be used in 4DCT scanning and potentially used as an auxiliary monitor to offer more information about respiration and body motion during RT in the future.