1、Journal of the Korean Physical Society, Vol. 66, No. 2, January 2015, pp. 301 313Evaluation of Breathing Patterns for Respiratory-gated Radiation Therapy Using the Respiration Regularity IndexKwang-Ho Cheong, MeYeon Lee, Sei-Kwon Kang, Jai-Woong Yoon, SoAh Park, Taejin Hwang, Haeyoung Kim, KyoungJu
2、Kim, Tae Jin Han and Hoonsik BaeDepartment of Radiation Oncology, Hallym University College of Medicine, Anyang 431-070, Korea(Received 3 November 2014)Despite the considerable importance of accurately estimating the respiration regularity of a patient in motion compensation treatment, not to mentio
3、n the necessity of maintaining that regularity through the following sessions, an eective and simply applicable method by which those goals can be accomplished has rarely been reported. The authors herein propose a simple respiration regularity index based on parameters derived from a correspondingl
4、y simplified respiration model. In order to simplify a patients breathing pattern while preserving the datas intrinsic properties, we defined a respiration model as a cos4(t) t) wave form with a baseline drift. According to this respiration formula, breathing-pattern fluctuation could be explained u
5、sing four factors: the sample standard deviation of respiration period (sf ), the sample standard deviation of amplitude (sa ) and the results of a simple regression of the baseline drift (slope as , and standard deviation of residuals as r ) , where is a variableof a respiration signal. The overall
6、 irregularity () was defined as | | newly-derived by using principal component analysis (PCA) for the four fluctuation parameters and has two principal components (1, 2). The proposed respiration regularity index was defined as = ln(1 + (1/)/2, a higher indicating a more regular breathing pattern. W
7、e investigated its clinical relevance by comparing it with other known parameters. Subsequently, we applied it to 110 respiration signals acquired from five liver and five lung cancer patients by using real-time position management (RPM; Varian Medical Systems, Palo Alto, CA). Correlations between t
8、he regularity of the first session and the remaining fractions were investigated using Pearsons cor-relation coecient. Additionally, the respiration regularity was compared between the liver and lung cancer patient groups. The respiration regularity was determined based on ; patients with 0.7 was su
9、itable for respiratory-gated radiation therapy (RGRT). Fluctuations in the breathing cycle and the amplitude were especially determinative of . If the respiration regularity of a patients first session was known, it could be estimated through subsequent sessions. Notably, the breathing patterns of t
10、he lung cancer patients were more irregular than those of the liver cancer patients. Respiration regularity could be objec-tively determined by using a composite index, . Such a single-index testing of respiration regularity can facilitate determination of RGRT availability in clinical settings, esp
11、ecially for free-breathing cases.PACS numbers: 87.19.St, 87.52.-g, 87.53.Vb, 87.56.-vKeywords: Breathing pattern, Respiration regularity index, RPM system, Lung, LiverDOI: 10.3938/jkps.66.301I. INTRODUCTIONIn radiation therapy, delivery of an appropriate dose to targets in the lung or the upper-abdo
12、minal regions is complicated by respiratory motion 1, 2. In one inhale-exhale cycle for example, a lung tumor can move as much as 30 mm 2. Imaging and beam-delivery tech-niques have been developed to minimize the eects of or-gan motion, among which modalities, respiratory-gated radiation therapy (RG
13、RT) and motion tracking tech-E-mail: b8510hallym.or.kr; Fax : +82-31-380-3913niques have become popular in the radiation oncology field 35. In RGRT, the beam is on only during a pre-determined gating window interval. In the case of motion tracking techniques, multi-leaf collimator (MLC) leaves follo
14、w the targets motion while the beam is on and radiation is delivered. The newly-developed Vero tracking system is also appropriate for motion tracking 6, and a motion compensation technique is applied in CyberKnife, as well. Recent studies on the Synchrony of CyberKnife showed the possibility of mar
15、gin reduc-tion without sacrificing target coverage 7,8. Moreover, Ernst introduced motion prediction and compensation methods to tackle in detail the problem of latency for-301-302- Journal of the Korean Physical Society, Vol. 66, No. 2, January 2015quasi-periodic motion in robotic radiosurgery 9. A
16、l-though compensation for the targets motion is roughly possible by using several known algorithms, prediction still relies on breathing regularity.Most breathing motion control techniques are based on the assumption that during a treatment session, not only does the breathing pattern not change 1,2
17、 but the ini-tial breathing pattern in the remaining fractions is main-tained. However, if a patients breathing is irregular, a dosimetric error more serious than that in conventional treatment can be incurred. This is an especially critical issue in stereotactic body radiation therapy (SBRT), in th
18、at the treatment time is longer than it is in conven-tional therapy and very high precision is required 10, 11. Thus, the respiration regularity of a patient as well as the probability of maintaining that regularity through the following sessions must be estimated. However, this issue has not yet re
19、ceived adequate attention, despite the important information that is provided for treatment of patients with tumor motion. Most studies reported since the early 2000s, rather, have focused on overcoming the transient changes in irregular breathing patterns 12.In this study, we investigated the facto
20、rs that can de-termine the regularity of respiration, and then formu-lated a respiration regularity index based on parameters derived from a simplified respiration model. We then evaluated the respiration regularity and validity of the index by application to clinical cases. We determined that the i
21、nitial regularity was maintained to the end of treatment by confirming the correlation between the reg-ularity of the first session and the remaining fractions.II. MATERIALS AND METHODS1. Patient Selection and Acquisition of Respi-ration SignalsWe randomly selected the respiration signals of liver c
22、ancer (Nos. 1 5) and lung cancer patients (Nos. 6 10) that had been treated in our facility from 2007 to 2010 (Table 1). We separated the lung cancer pa-tients from the liver cancer patients owing to the possi-bility that the former had inferior respiration regularity relative to that of the latter
23、13. Respiration signals were obtained using a real-time position management (RPM; Varian Medical Systems, Palo Alto, USA) sys-tem 14. Although respiration signals can be acquired during either four-dimensional computerized tomogra-phy (4DCT) scanning or RGRT, we used only signals from treatment sess
24、ions because the 4DCT scanning time recorded in the exported RPM data was not long enough. The patients breathed freely, without any visual or audio guidance. A total of 11 respiration signals were extracted per patient (initial session: 1 signal, early session: 4 sig-nals, middle session: 3 signals
25、, final session: 3 signals); we set the first session as the reference for predicting theFig. 1. Simplified breathing model using Eq. 1. The breathing pattern can be represented using a cosine4 wave form with fluctuations in the respiration frequency, ampli-tude, and baseline.regularity variation th
26、ereafter. The patient information, including planning target volume PTV, tumor length in the cranial-caudal (CC) direction, uncontrolled and con-trolled tumor motion in CC direction, and RGRT gating window size, is displayed. The parameters were similar, except that the PTV volume of the liver cance
27、r patients was more than twice that of the lung cancer patients.2. Respiration-pattern ModelingClinically, multiple physiological factors, such as breathing period, lung volume change, inhalation time and exhalation time, and their ratio (exhalation-inhalation time ratio: EI ratio), are used to expl
28、ain the breathing pattern. However, with respect to RGRT, tu-mor motion and baseline stability are additional critical factors that demand consideration. A respiration pat-tern is time-series data represented as a periodic function (i.e., in a sine or a cosine wave form), though an actual breathing
29、signal is more irregular and, thus, more di-cult to standardize. A quasi-periodic function account-ing for variations observed in real respiration is useful for a relatively precise simulation of human respiration mo-tion 9, though sometimes a more simplified respiration form is sucient for model-ba
30、sed analysis. In order to simplify a patients breathing pattern while preserving the datas intrinsic properties, we defined the respiration formula asY (t) = A(t) cos4(t) t) + B(t), (1)where A, , and B indicate the amplitude, period and baseline drift, respectively. Because each factor is a func-tio
31、n of time, the regularity is considered to be adequate if each parameters fluctuation is moderate. At the ini-tial stage of this study, we investigated the correlationsEvaluation of Breathing Patterns for Respiratory-gated Radiation Kwang-Ho Cheong et al. -303-Table 1. Patient information used in th
32、is study: PTV, tumor length in the cranial-caudal (CC) direction, uncontrolled and controlled tumor motion in the CC direction, and RGRT gating window size.ControlledPTV volume Tumor length Tumor motion GatingStudy No. Diagnosis (cm3) (cm) (cm) tumor motion window (%)(cm)1 Liver 651.6 9.1 0.9 0.3 38
33、 622 Liver 2134.8 20.4 1.2 0.5 30 703 Liver 1004.8 15.6 1.3 0.4 38 634 Liver 748.5 11.1 1 0.3 40 705 Liver 189.4 8.1 1.1 0.5 30 70945.8 727.1 12.86 5.1 1.10 0.15 0.40 0.106 Lung (L) 85.2 6.3 1.1 0.4 35 707 Lung (L) 597.2 17.8 1.4 0.5 30 708 Lung (R) 191.8 13.1 1 0.4 35 759 Lung (L) 1007.5 16.7 1.2 0
34、.4 38 6310 Lung (R) 284.2 10.4 0.6 0.2 40 60433.2 373.6 12.86 4.7 1.06 0.30 0.38 0.11among the variations in the period frequency, inhalation time, exhalation time, EI ratio, amplitude, and marker positions at the end of exhalation and at the end of in-halation, then, we defined, by using a principa
35、l compo-nent analysis (PCA), the major factors aecting respi-ration. From the preliminary result, we found that (1) strong correlations existed among the period, inhalation time, exhalation time and EI ratio; (2) strong correla-tions existed among the amplitude and marker positions at the ends of ex
36、halation and inhalation; (3) explaining the respiration pattern by using variations in the period frequency and the amplitude is reasonable. Additionally, baseline changes in the sum had to be considered. There-fore, the respiration model in Eq. 1 is feasible. Ruan et al. followed a similar decompos
37、ition approach, including additional noise, in their work 15; they did not focus on respiratory regularity. Fig. 1 provides a simple illustra-tion of the model. Next, we will define the fluctuation parameters for each component in detail.A. Regularity of respiration cycleFigure 2 is an actual lung c
38、ancer patient (No. 10)s respiration signal acquired using the RPM system and used in this study. The amplitude, breathing frequency and baseline are unstable. Full exhalation is represented by the upper peaks, and full inhalation by their oppo-sites. Hereafter, upper peaks represent the end of ex-ha
39、lation (EE) in all figures unless otherwise indicated. The frequency of the time-series data is commonly ana-lyzed on the basis of the frequency domain spectra 16; however, due to the wide distribution of periods, the characteristics of the respiration period are not clearlyFig. 2. (Color online) RP
40、M systems 200-second respira-tion signal for a lung cancer patient (No. 10). The peaks at the end of exhalation (EE: o marks) and of inhalation (EI: x marks) are displayed.seen. Therefore, we defined “peak-to-peak”, one EE to the next EE, as one respiration period; thus, correctly detecting peaks in
41、 a noisy signal is important. To deter-mine whether each peak is significantly larger or smaller than the data around it, we used a peak detection al-gorithm that finds, according to a predefined threshold, local peaks and valleys in a noisy respiration signal. The distribution of respiration period
42、s can then be modeled as a Gaussian distribution 17; in this way, the char-acteristics of the period can be represented as sample mean and sample standard deviation. We confirmed the normality of this distribution by using a Q-Q plot. If there is no variation, the sample mean is the same as the nomi
43、nal period but if the respiration cycle is not regu-304- Journal of the Korean Physical Society, Vol. 66, No. 2, January 2015Fig. 3. RPM systems 11 respiration signals for a lung cancer patient (No. 10). The signals were from the 1 initial, 4 early, 3 middle, and 3 final sessions.lar, the sample sta
44、ndard deviation will be increased. In other words, the sample standard deviation is the proper factor for explaining fluctuations in the respiration pe-riod. We defined the sample standard deviation of the respiration period as sf. The marker positions at the end of inhalation (EI) and at the EE, ca
45、lculated automati-cally using in-house analysis tools with MATLAB 2011b (MathWorks Inc., Natick, USA), are also displayed in Fig. 2. SPSS 21 (IBM SPSS Inc., Armonk, USA) was used for further statistical analysis.B. Regularity of amplitudeThe amplitude of the respiration pattern at the i-th period wa
46、s calculated using Eq. 2, where x is a marker position at the EE:1Ai = 2(|xEE,i xEI ,i| + |xEI ,i xEE,i+1|), (2)he amplitude is the average of the left- (ampL) and right-side (ampR) values for any inhalation peaks, as indicated in Fig. 1. Points that could not be located on a complete cycle were rem
47、oved from the calculation. Figure 3 shows all of the 11 respiration signals for a lung cancer patient (No. 10). As indicated, the range of the amplitude varied among sessions due to marker-position and camera-angle variations. However, the overall magnitude of the ampli-tude generally did not change
48、 due to the fact that themagnitudes are absolute values normalized to the known intervals between two reflectors on a marker block. The distribution of amplitude ranges can be explained by us-ing sample means and sample standard deviations of the breathing-cycle frequency 18. We take the sample stan
49、-dard deviation of the amplitude (sa) as the amplitude irregularity parameter.C. Evaluation of baseline changesBaseline changes can be considered, separately, ac-cording to either of two components: random changes and long-term drift 9. Baseline long-term drift, also known as “baseline drift”, is a common phenomenon in breathing patterns 19. Generally, the baseline is de-fined as the position at the EE because it is more sta-ble than it is at the EI 20
