- Email: [email protected]
Evaluation of internal lung motion for respiratory-gated radiotherapy using MRI: Part II—margin reduction of internal target volume
Int. J. Radiation Oncology Biol. Phys., Vol. 60, No. 5, pp. 1473–1483, 2004 Copyright © 2004 Elsevier Inc. Printed in the USA. All rights reserved 0360-3016/04/$–see front matter
doi:10.1016/j.ijrobp.2004.05.054
CLINICAL INVESTIGATION
Lung
EVALUATION OF INTERNAL LUNG MOTION FOR RESPIRATORY-GATED RADIOTHERAPY USING MRI: PART II—MARGIN REDUCTION OF INTERNAL TARGET VOLUME H. HELEN LIU, PH.D.,* NICHOLAS KOCH, M.S.,* GEORGE STARKSCHALL, PH.D.,* MARC JACOBSON, M.D.,† KENNETH FORSTER, PH.D.,* ZHONGXING LIAO, M.D.,* RITSUKO KOMAKI, M.D.,* AND CRAIG W. STEVENS, M.D., PH.D.* *Division of Radiation Oncology and †Division of Diagnostic Imaging, The University of Texas M. D. Anderson Cancer Center, Houston, TX Purpose: To analyze the relationship between lung motion and skin surface motion during respiration, determine the uncertainties and variability of such a relationship, and assess the potential of reducing internal target margin for gated radiotherapy. Methods and Materials: Three healthy volunteers and four lung cancer patients were recruited in a prospective imaging study using MRI to track the internal lung and external skin motion during breathing. The relationship between the lung and skin motion was modeled using linear regression analysis. The slope of the linear fit and its confidence interval were analyzed for different lung locations, skin surface locations, and breathing patterns from separate imaging sessions. The margins of the internal target volume were calculated based on the residual lung motion during gating and its uncertainties from multiple treatment fractions for the gated treatment. Results: The slope and confidence interval of the linear regression from the motion analysis were uniquely defined by the locations of the lung, skin surface, and breathing patterns. Statistically significant differences were observed among individuals and between different times of measurement. The normal free-breathing motion averaged from all volunteer and patient data was 13.4 ⴞ 7.4 mm along the superior–inferior (SI) direction and 6.9 ⴞ 2.6 mm along the anterior-posterior (AP) direction. With simulated respiratory gating, the average margin reduction was 5.5 ⴞ 4.8 mm and 1.6 ⴞ 1.0 mm, respectively, along the SI and AP directions (or 36% ⴞ 15% and 25% ⴞ 14%, respectively, relative to free-breathing motion). Conclusion: Because respiratory movement is rather complex, the relationship between the lung and skin surface motion is affected by many anatomic and physiologic factors. The reduction of internal target margin and efficacy of the free-breathing gating technique should be assessed for individual cases. © 2004 Elsevier Inc. Lung motion, MRI, Respiratory gating, Internal target volume, Margin reduction.
because respiration-induced tumor movement causes a large amount of normal lung tissue to be irradiated (2–5). To reduce such a margin, several techniques have recently been developed for radiation beam delivery. Examples include breathing control methods, such as active breathing control (6) or deep inspiration breath-hold (7), and gating techniques, both passive (8, 9) and active (6), some of which involve using implanted internal fiducials (10). The purpose of these techniques is to reduce effectively the degree of tumor motion during RT by either stopping the breathing or synchronizing the radiation with respiration. The respiratory signal can be monitored using surrogate markers acquired from internal or external fiducials or by measuring air volume and/or flow. The effectiveness of the gated treatments depends on a
INTRODUCTION In typical radiotherapy (RT) regimens, the clinical and physical uncertainties encountered in delineating and localizing tumors are currently accounted for by enlarging the target volume with sufficient margins. Report 62 of the International Commission on Radiation Units and Measurements (ICRU) (1) defined four volumes related to the treatment margin: gross target volume, clinical target volume, internal target volume (ITV), and planning targe volume. The concept of the ITV was formally introduced in this ICRU report to describe specifically the geometric uncertainty caused by tumor mobility, namely intra- and interfractional tumor motion. In RT for lung cancer, reducing the ITV margin is important Reprint requests to: H. Helen Liu, Ph.D., Division of Radiation Oncology, Unit 94, The University of Texas M. D. Anderson Cancer Center, 1515 Holcombe Blvd., Houston, TX 77030. Tel: 713-563-2546; Fax: 713-563-2545; E-mail: [email protected] Supported in part by a sponsored research agreement with
Varian Oncology Systems and an internal research grant from M. D. Anderson Cancer Center. Received Dec 11, 2003, and in revised form May 14, 2004. Accepted for publication May 26, 2004. 1473
1474
I. J. Radiation Oncology
● Biology ● Physics
number of factors. The position of the tumor and its surrogates, such as the internal or external fiducials, must have a one-to-one correspondence. It is also critical that such a correspondence be consistent and reproducible during the entire treatment course. Even with gated treatment, an ITV margin would still be required to compensate for the uncertainties of tumor position that primarily come from two sources: the residual tumor motion within the gating window, which normally occupies a fraction of the breathing cycle (e.g., 10 –30%); and possible variations of the gating window because of changes in tumor position among different fractions. Because the respiratory motion could potentially be very complex, as has been demonstrated by some of the existing research (4, 11), the efficacy of the gated treatment depends strongly on whether the lung motion can be monitored and predicted accurately and reproducibly and whether a clinically meaningful reduction of the ITV margin can be accomplished with respiratory gating. The aims of our research were to investigate the motion of internal lung structures and how such motion is related to the motion of an external fiducial and to determine whether the external fiducial can be used reliably to predict the lung motion during a treatment course. We have developed a technique to track the lung and external skin surface movement during breathing using MRI. The details of the imaging and tracking technique have been described in a companion report (12). In the present study, we focused on quantitative analysis of the relationship between the lung and skin movement, specifically how such a relationship may be affected by variables associated with different locations of the lung, skin marker, and breathing patterns. A more important question we attempted to address was whether the relationship is reproducible over time, and consequently, whether the degree of ITV margin reduction was achievable in gated treatment according to the residual tumor motion and interfractional variability. METHODS AND MATERIALS Image acquisition and data analysis An MRI-based imaging technique was developed to acquire dynamic cine images through the sagittal and coronal planes of the studied subjects and to assess the relationship between the internal lung motion and external skin surface motion. Details of the imaging protocol and methods of processing the MRI to measure the lung and skin motion have been previously reported (12). In brief, 7 subjects, 3 healthy young volunteers and 4 lung cancer patients, were recruited into the study. Sagittal and coronal cine MRI scans with a temporal resolution of 0.45 s/frame were acquired through the thoracoabdominal region while subjects performed normal breathing and other altered breathing maneuvers (e.g., chest, abdominal, and deep breathing). The motion of the internal lung vessels, as well as the motion of the exterior skin surface at designated locations (skin 1 to skin 4) were tracked from the cine MRI scans using in-house– developed image processing algorithms.
Volume 60, Number 5, 2004
Thus, the positions of the lung vessels as a function of time, V(t), can be described as a function of that of the skin surface, S(t). This allowed us to model the relationship between the vessel positions and the skin surface positions during breathing directly. On the basis of the results obtained from the breathing traces, simple linear regression analysis was adequate in quantifying such a relationship: V共t兲 ⫽ a ⫹ b · S共t兲
(1)
where V(t) and S(t) are the displacements of the vessel and skin surface from their mean positions at time, t. The slope, b, of the regression line was an indicator of the relationship between the marker and vessel motion, describing the displacement of the vessel position relative to that of the skin over the same period. The uncertainty of the relationship between the skin and lung motion can be described by the 95% confidence intervals of the slope values computed from the regression analysis. An example of such a linear regression analysis is given in Fig. 5 in Koch et al. (12). Because a typical RT course consists of 30 – 40 treatment fractions, the relationship between the skin and lung motion could potentially change over time. To assess such a possibility, we repeated the image acquisition with an approximately 2-week interval for all studied subjects. For the patients, the two imaging sessions usually occurred near the middle and end of the treatment course. The same image processing procedures were applied for the images acquired from both sessions. The slope value b and its confidence interval of the two imaging sessions gave us additional estimates of the uncertainty and possible variation of the skin–lung relationship during the 2–3-week interval between the image sessions. Assessment of margin reduction for gated treatment As mentioned earlier, if a gating technique were to be used for treating lung on the basis of the skin surface motion, the internal margin required for treatment must include two components. The first component is the intrafractional motion, which is contributed by residual tumor motion within the gating window occupying a fraction of the breathing cycle. The second component is the interfractional motion, which comes from the position change of the gating window from fraction to fraction. At our clinic, the gating beam-on window is selected near the end-expiration phase with a duty cycle of approximately 25% (i.e., the duration of such a gating window is approximately onequarter of a breathing cycle). The internal margin is defined as the component of the margin due to the internal lung motion only, excluding setup error uncertainties. Thus, the internal margin, or the range of lung motion within this gating window from two separate sessions, m1 for session 1 and m2 for session 2, can be calculated as m 1 ⫽ 共V max,1 ⫹ 1兲 ⫺ 共V min,1 ⫺ 1兲 m 2 ⫽ 共V max,2 ⫹ 2兲 ⫺ 共V min,2 ⫺ 2兲
(2)
in which Vmin and Vmax are the minimal and maximal vessel
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1475
Fig. 1. (a) Relationship between lung vessel movement and skin 4 movement during normal breathing from two imaging sessions (a) session 1; (b) session 2. (c) Margin calculation for two imaging session (session 1 and 2, solid and dashed lines, respectively). Minimal and maximal vessel excursion from two imaging sessions (Vmax,1 ⫹ 1, Vmax,2 ⫹ 2, and Vmin,1 ⫺ 1, Vmin,2 ⫺ 2) within gating window also shown.
positions within the gating window, respectively, for each session; is the 95% confidence interval of the position due to the uncertainty of the vessel positions. Here the subscript 1 or 2 denote the results from the two imaging sessions because the extent of the vessel motion may change over time during the treatment course. The above margin estimation attempts to enclose the range of lung motion within the 95% confidence interval by including the terms in the equation. The total margin accounting for the results from the two imaging sessions can then be calculated as m 1,2 ⫽ V max,1,2 ⫺ V min,1,2 in which
(3)
V max,1,2 ⫽ max关共V max,1 ⫹ 1兲, 共V max,2 ⫹ 2兲兴 V min,1,2 ⫽ min关共V min,1 ⫺ 1兲, 共V min,2 ⫺ 2兲兴
(4)
where Vmin,1,2 and Vmax,1,2 are the minimal and maximal excursion of the lung motion from the two sessions, respectively. To illustrate the margin calculation, Fig. 1 shows an example of the superior–inferior (SI) displacement of a vessel versus the displacement of the skin surface at the abdominal marker skin 4. Also shown are the lines of best fit (solid) and the 95% confidence interval lines (dashed) of the residuals from the fit. Note difference in the slope indicating the relationship between vessel and skin surface
1476
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Fig. 2. Effect of vessel location on regression slope of vessel–skin relationship from Volunteer 1. Results of tracking five vessels at skin 4 location during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
motion may be slightly different between the two scanning sessions. The two regression lines and their confidence lines in Fig. 1a and Fig. 1b are superimposed in Fig. 1c, assuming the mean positions of the skin marker or lung vessels were identical in both imaging sessions. This assumption was used because the mean position of the skin marker or lung vessels caused by the setup error can be excluded, because the setup error is not included in the ITV margin estimation. Figure 1c illustrates how the internal margin was estimated from the data of the two imaging sessions. Vmin ⫺ and Vmax ⫹ , as defined by the above equations, are illustrated by the arrows for each session. Although each session had its own margin window defined by Eq. 2, the largest difference between Vmin ⫺ and Vmax ⫹ from the two sessions gives the margin size, defined by Eq. 3, necessary to account for the displacement of the lung vessels from both sessions according to the skin marker location. Ideally, the relationship between the vessel and skin movements should be assessed for each fraction of the treatment to obtain the statistical distribution of such a relationship. Here, we used data from only two imaging sessions as a starting point to illustrate the principle of the margin estimation and the effect of motion variability over time. For each vessel tracked from the volunteers and patients, we computed the normal free-breathing motion, as well as the motion within the gating window, from two imaging sessions. Using the above approach, we, thus, estimated the margin reduction in both absolute (millime-
ters) and proportional (percentages) terms relative to the free-breathing motion.
RESULTS See Figures 2–10 for summary of the observations made on the relationship between lung vessel motion and skin motion, and more importantly, the dependency of such a relationship on the vessel and skin locations and the variation of the relationship with the studied subjects and time of measurement. We used the results taken from Volunteer 1 as a representative case for all studied subjects (see Figs. 2– 4). In Fig. 2, the results of linear regression analysis (Eq. 1), namely, the slope value b and its confidence interval, are displayed for the five vessels tracked for this volunteer from two separate imaging sessions. The skin surface motion at marker 4 and normal breathing data were used in this analysis. Figure 2 shows that the slope values are quite different among the five vessels along either the SI or AP directions. The difference was because, for the same magnitude of skin 4 movement, the degree of motion of the five vessels was distinctively different owing to their locations. Along the SI direction, the V1-R82b and V1-L98a, which are located in the lower lung lobe, had greater motion, and thus, greater slope values, compared with other vessels, particularly in contrast to V1-R82a, located more superiorly. Along the AP direction, the degree of vessel motion, in general, was less than that of the SI direction, leading to smaller slope values.
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1477
Fig. 3. Effect of skin surface location on regression slope of vessel–skin relationship from Volunteer 1. Results of tracking two vessels at skin 1 to skin 4 locations during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) vessel motion along AP direction.
Vessel V1-R82b and V1-L98a, again, showed larger slope values than the other vessels owing to the greater mobility of these two vessels. More noteworthy was the statistically significant difference in the slope values, which exceeded their confidence intervals from the two imaging sessions, indicating that the relationship between the skin 4 and vessel movement had changed between the two sessions, particularly along the SI direction for the first four vessels. Thus, for the same amount of skin movement, the degree of vessel movement could be different, possibly depending on how exactly the studied subject was performing the regular breathing during the different imaging sessions. Figure 3 further illustrates the results from all four skin locations for the first two vessels only, V1-R82a and V1R82b. Again, the data of the normal breathing and from two imaging sessions are included. For the same vessel, the slope values were also quite different at the four skin locations. This was because, with the same amount of vessel movement, the degree of skin movement was not uniform throughout the anterior skin surface. The slope values at skin 4 tended to be the smallest along either the SI or AP direction, because skin 4 had the greatest amount of movement. The slope values for skin 2 tended to be the greatest because it had the least amount of movement. In addition, the error bars of the slope values were associated with the uncertainty and degree of correlation between the vessel and skin movement (i.e., the larger the error bar, usually the poorer the correlation). The results suggested that for gated treatment using external markers, skin 4 may be a better
location owing to its greater range of movement and, thus, a greater sensitivity for detecting the vessel movement. In contrast, skin 2 may not be an ideal location because it had the smallest degree of movement during breathing and a poor correlation along the SI direction. Figure 4 demonstrates the comparison of four different breathing patterns for vessels V1-R82a and V1-R82b. Only the data from skin 4 movement were included. The results showed that the relationship between the vessel and skin 4 movement was also unique for an individual breathing pattern along both SI and AP directions. During altered breathing, the lung expansion and use of muscles could be quite different compared with normal breathing, leading to more complex movement and, thus, a more complex relationship between the lung and skin surface movement. For example, comparing abdominal and chest breathing, the slope values tended to decline along the SI direction but increase along the AP direction. This difference in slope values was possibly due to reduced SI lung motion and magnified AP lung expansion during chest breathing. Because the relationship between the lung and skin movement can be quite different depending on the breathing pattern, even for regular normal breathing, the slope values from the two imaging sessions could also change, particularly for vessel V1-R82b, which was more inferiorly located than V1-R82a. This means that it might be very difficult to reproduce exactly the same lung and skin movement during each imaging session at different times, even for healthy volunteers.
1478
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Fig. 4. Effect of breathing pattern on regression slope of vessel–skin relationship from Volunteer 1. Results of tracking two vessels at skin 4 location during normal, abdominal, chest, and deep breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
Figure 5 summarizes the lung movement during normal breathing from three different vessels, at skin 2, 3, and 4, as an example of a typical lung cancer patient
(Patient 3). The uncertainty of the slope values was greater, particularly at the skin 2 location for this patient because of a poor correlation between the vessel and skin
Fig. 5. Regression slope of vessel–skin relationship from Patient 3. Results of tracking three vessels at skin 2, 3, and 4 locations during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1479
Fig. 6. Intersubject variation of regression slope of vessel–skin relationship from three individuals (Volunteers 1 and 3 and Patient 3). Results of tracking three vessels at skin 4 location during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
movement. Similar to that observed for the volunteers, the relationship between the lung and skin movement depended on the location of the vessel, as well as of the skin marker. Again, the patient data demonstrated statistically significant differences in slope values between the two imaging sessions that were greater than the confidence intervals. Comparing the results of skin 2, 3, and 4,
in general, we noted that the confidence intervals of the slope values tended to be smaller at skin 4. The reproducibility of the slope values from two imaging sessions was also slightly improved at this location from other volunteer and patient data. Thus, skin 4 seemed to be the preferred location if a single external fiducial has to be chosen for monitoring the respiratory cycle because of
Fig. 7. Distribution of vessel motion during normal free breathing from all vessels tracked of volunteers. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
1480
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Fig. 8. Distribution of vessel motion during normal free breathing from all vessels tracked of patients. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
the sensitivity of skin 4 in tracking lung motion and the consistency of the correlation from different imaging sessions. Figure 6 compares the results from tracking three vessels located at approximately same location of the lung from three different studied subjects: Volunteers 1 and 3 and Patient 3. The slope values using skin 4 during normal breathing from two imaging sessions are demonstrated. Even for the same region of the lung and skin marker location, the lung and skin movement and their relationship were unique to the individual and exhibited a large degree of intersubject dependency. Thus, the degree of lung movement as monitored by the skin movement should be assessed carefully for an individual, possibly because of the differences in the breathing pattern among the population, especially for lung cancer patients who have impaired lung function. Figures 7 and 8 summarize the magnitude of the lung motion based on tracking all the vessels in volunteers and patients, respectively, during normal breathing. The results are also included in Table 1. For volunteers, the SI and AP
motion was 11.0 ⫾ 5.3 mm and 6.5 ⫾ 2.2 mm, respectively. For patients, the distribution of free-breathing motion was not statistically different from that of the volunteers, but the sample size was small (p ⬎0.2, Kolmogorov-Smirnov test), although the average magnitude of the motion was slightly greater (18.1 ⫾ 9.5 mm and 7.8 ⫾ 3.5 mm along the SI and AP directions, respectively). For both the volunteers and patients, the primary axis of motion was in the SI direction. For comparison purposes, the results of the vessel motion during deep breathing were also included in Table 1. Deep breathing caused similar changes in SI motion for both patients and volunteers, but volunteers seemed to be able to increase their AP motion more than the patients. Figures 9 and 10 summarize the degree of margin reduction achievable according to the data from two imaging session using Eqs. 2– 4. Skin 4 was used as the location of the external fiducial in tracking the vessel movement; for the reasons discussed above. The data in Fig. 9 suggest that, for volunteers, along the SI direction, the possible margin reduction varied significantly from 1 to 11 mm or 20 –50%, depending on the individual vessels. Along the AP direc-
Table 1. Possible margin reduction with gating using skin 4 to track vessel movement during breathing Volunteers
Patients
SI
Normal breathing (mm) Margin reduction (mm) Margin reduction (%) Deep breathing (mm)
AP
SI
AP
Mean
SD
Mean
SD
Mean
SD
Mean
SD
11.0 4.2 35.0 28.4
5.3 3.1 15.3 13.8
6.5 2.0 29.6 21.9
2.2 1.0 12.3 6.3
18.1 8.0 38.8 25.4
9.5 6.4 16.1 12.5
7.8 1.0 15.4 12.9
3.5 0.9 11.8 5.1
Abbreviations: SI ⫽ superior–inferior; SD ⫽ standard deviation.
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1481
Fig. 9. Distribution of margin reduction with gating at skin 4 location from all vessels tracked of volunteers. (a) Absolute margin reduction and (b) relative margin reduction compared with normal free breathing.
tion, the degree of absolute margin reduction was quite reduced owing to a smaller degree of AP lung motion and uncertainties associated with tracking the motion. For patients, Fig. 10 shows that the degree of AP margin reduction was curtailed further compared with that of the volunteers, because the correlation between the lung and skin movement was quite poor along this direction (see Fig. 13 of Koch et al. [12]). Along the SI direction, a more significant margin reduction was possible for one patient (Patient 3) that showed excellent correlation between the SI and skin 4 motion (see Fig. 7 of Koch et al. [12]). This suggests that certain patients might be more suitable for, and will realize a greater benefit from, gated treatment. Even for this particular group of patients, margin reduction might be more feasible along the SI direction than the AP direction. The results from Figs. 9 and 10 indicated that the distributions of the margin reduction were not statistically different between the volunteers and patients (p ⬎0.05, Kolmogorov-Smirnov test). The mean and standard deviation of the absolute and
relative margin reduction are also given in Table 1. The average margin reduction was 5.5 mm ⫾ 4.8 mm and 1.6 mm ⫾ 1.0 mm along the SI and AP directions, respectively, corresponding to 36% ⫾ 15% and 25% ⫾ 14%, respectively, relative to free-breathing motion. Compared with the SI direction, the margin reduction along the AP direction was much smaller and might not lead to meaningful clinical reductions of normal tissue volume in treating lung tumors. DISCUSSION In this research, we analyzed the relationship between the lung and skin movement during breathing for healthy volunteers and patients recruited in a prospective imaging study to evaluate the efficacy and improvement of the respiratory gated RT for lung cancer. In our companion report (12), the details of the MRI technique and image processing procedure were given, along with the results from analyzing the correlation between the lung and skin
Fig. 10. Distribution of margin reduction with gating at skin 4 location from all vessels tracked of patients. (a) Absolute margin reduction and (b) relative margin reduction compared with normal free breathing.
1482
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Table 2. Motion of skin markers during normal and deep breathing from volunteers and patients Volunteers
Patients
Normal
Deep
Normal
Deep
Skin marker (mm)
Mean
SD
Mean
SD
Mean
SD
Mean
SD
1 2 3 4
3.3 3.2 4.1 6.8
1.1 1.0 1.1 3.1
21.5 17.6 17.6 23.7
5.6 4.1 3.9 7.8
— 3.4 5.3 9.7
— 0.9 2.4 4.0
— 8.6 13.7 21.4
— 1.7 6.7 6.2
Skin markers were tracked at sagittal planes at which MRI scans were taken. Abbreviation: SD ⫽ standard deviation.
movement. We found that the SI motion of the lung, particularly in the lower lobe, can be better described by abdominal skin surface movement, but the AP lung expansion correlated better with chest marker movement. Complimentary to the findings in our companion report, in the current paper, we focused on using the linear regression technique to quantify the relationship between the lung and skin movement and its dependency on several variables, including the location of the lung, skin surface, breathing pattern, subject, and time of measurement. Our results demonstrated that along the SI direction the relationship between the lung and skin movement could be described well by a simple linear model. However, along the AP direction, the lung movement was more complex, with a possibility of hysteresis (see Fig. 7 of Koch et al. [12]). This resulted in a larger degree of uncertainty, reflected by a greater confidence interval of the slope value along the AP direction. Although more sophisticated mathematic models, other than linear regression, may be used to describe such a relationship, their necessity was not fully warranted because we were more interested in quantifying the variability of such a relationship and its effect on margin reduction. The results from healthy volunteers and patients demonstrated that, apparently, various regions of the lung move quite differently during breathing. The upper lobe of the lung exhibited expansion mainly along the AP direction, and the lower lobe of the lung moved along the SI direction driven by the diaphragm. Consequently, the relationship between the internal lung parenchyma and external skin movement at different regions will be uniquely dependent on the relative locations of the lung and skin surface. As indicated by Figs. 2 and 3, using data of Volunteer 1 as an example, the slope values reflected the degree of lung motion relative to the skin surface motion during breathing. A greater degree of mobility was observed for lower lobe vessels along the SI direction and upper lobe vessels along the AP direction. In patients, lung motion may be complicated further by abnormal breathing patterns caused by patient-specific obstructive or restrictive effects of pulmonary disease. Additionally, the body anterior surface also moves with a different magnitude with lung expansion because of regional muscle involvement. Table 2 summarizes the results
of the fiducial movement at different locations during normal and deep breathing. The upper abdominal surface such as skin 4 had the greatest movement (6.8 ⫾ 3.1 mm, and 9.7 ⫾ 4.0 mm, respectively, for volunteers and patients). Thus, skin 4 was a more sensitive location to track the lung motion, particularly along the SI direction. Because the lung and skin surface movement are also greatly affected by the breathing pattern, their relationship is strongly dependent on the breathing maneuver and exactly how the lung expands and contracts during breathing. Therefore, a significant degree of variability of the lung– skin relationship is due to changes in the breathing patterns (Fig. 4), which is also the primary cause for differences in such a relationship observed from separate imaging sessions. Certain vessels also exhibited a greater variability than others (e.g., V1-R82a vs. V1-R82b [Fig. 4]), indicating that certain portion of the lung may be more reproducible to track using external fiducials than others. Similarly, certain skin markers tended to produce more consistent results than others. For example, skin 4 seemed to be a more preferred location than skin 2 (Fig. 5), if a single fiducial had to be chosen for gating purposes. This conclusion also compliments the previous observations made in our companion report (12); in a sense, the SI motion, which is the primary source of lung movement, correlated better with the upper abdominal skin movement. Owing to the uncertainties and variability associated with the lung–skin movement, the degree of margin reduction may not be as great as expected even if skin 4 were used to track the lung motion. The data presented in Figs. 9 and 10 suggest that the margin reduction can be more easily accomplished along the SI direction for both volunteers and patients, again because of the stronger and more consistent correlation between skin 4 and SI motion of the lower lobe lung. However, along the AP direction, the degree of the correlation and its uncertainty make the margin reduction virtually impossible for a quite a few studied subjects. Currently, most of the radiation beams in treating lung cancer are coplanar beams involving the AP direction. Because the SI movement of the lung is the primary source of motion with a greater magnitude (as indicated by the data shown in Figs. 7 and 8), gated treatment could be still very beneficial for a fraction of the patient population. The key
MRI evaluation of internal lung motion for respiratory gating: II
issue is that we need to identify such individuals with careful imaging and apply appropriate margin reduction if gated treatment is chosen. Reliable means of monitoring the lung motion will be critical for gated treatments, particularly for longer treatment courses during which the possibility of changes in respiratory and other physiologic functions will be greater. The margin calculations presented here were the estimated internal margins described in ICRU Report 62 (1). The method of computing this margin used the image data from two imaging sessions only, which represents very limited sampling during a treatment course. Ideally, multiple sessions would be needed to access fully the range of lung motion and margin reduction. However, for practical reasons, our purpose was to demonstrate the principle of margin calculation and the effect of inter- and intrafractional motion. The actual margin reduction that could be achieved during a full treatment course is likely to be even less than what we estimated from only two fractions separated by about 2–3 weeks. Using the cine MRI scans, the internal lung motion was measured using pulmonary vessels as landmarks and surrogates for lung structure. The limitations of the MRI technique have been addressed in our companion paper (12). Mainly, the sagittal MRI scans we used to analyze the lung and skin positions could only provide motion along the SI and AP directions. The lateral movement of the lung could be acquired using coronal images, although they did not contain information on the simultaneous skin motion. However, because respiration mainly involves chest expansion
● H. H. LIU et al.
1483
and diaphragmatic movement, lateral lung movement could be a relatively minor component compared with the SI and AP lung movement, the results of which have been also shown by existing imaging studies (4, 13). Thus, we expect that the degree of motion and associated margin reduction could be less in the lateral dimension than the results we report for the SI and AP dimensions. Our results were also confined to those regions of the lung in which the vessels could be identified. Thus, our observations were near the lung hila. Additionally, because lung cancer will cause significant anatomic and physiologic changes to the lung parenchyma, the motion of the lung tumor may well be different from that of the vessels. However, the above results may be applicable to small, welllocalized, and free-floating nodules similar to pulmonary vessels. The motion of large lung tumors, especially along their edges, may be more complex, making it even more difficult to track using exterior markers and reduce the treatment margin. The determination of the motion of large tumors will be more challenging than that of the vessels because the borders of a lung tumor are usually difficult to distinguish from the surrounding tissues such as pericardium. Also, different regions of large tumors may move relatively independently (i.e., mediastinal nodes may move mostly along an AP axis, hilum nodes may move in an arc, and the primary tumor may move along an SI axis). However, we have developed a technique of tracking of lung tumor motion from cine MRI that will facilitate the study of such complex motion in the future.
REFERENCES 1. International Commission on Radiation Units and Measurements. ICRU report 62: Prescribing, recording and reporting photon beam therapy (supplement to ICRU report 50). Bethesda: International Commission on Radiation Units and Measurements; 1999. 2. Ross CS, Hussey DH, Pennington EC, et al. Analysis of movement of intrathoracic neoplasms using ultrafast computerized tomography. Int J Radiat Oncol Biol Phys 1990;18: 671–677. 3. Chen Q, Weinhous MS, Deibel C, et al. Fluoroscopic study of tumor motion due to breathing: Facilitating precise radiation therapy for lung cancer patients. Med Phys 2001;28:1850– 1856. 4. Seppenwoolde Y, Shirato H, Kitamura K, et al. Precise and real-time measurement of 3D tumor motion in lung due to breathing and heartbeat, measured during radiotherapy. Int J Radiat Oncol Biol Phys 2002;53:822–834. 5. Stevens C, Munden R, Forster K, et al. Respiratory-driven lung tumor motion is independent of tumor size, tumor location, and pulmonary function. Int J Radiat Oncol Biol Phys 2001;51:62–68. 6. Wong JW, Sharpe MB, Jaffray DA, et al. The use of active breathing control (ABC) to reduce margin for breathing motion. Int J Radiat Oncol Biol Phys 1999;44:911–919. 7. Rosenzweig K, Hanley J, Mah D, et al. The deep inspiration
8.
9.
10.
11.
12.
13.
breath-hold technique in the treatment of inoperable nonsmall-cell lung cancer. Int J Radiat Biol Phys 2000;48:81–87. Kubo HD, Len PM, Minohara S, et al. Breathing-synchronized radiotherapy program at the University of California Davis Cancer Center. Med Phys 2000;27:346–353. Mageras GS, Yorke E, Rosenzweig K, et al. Fluorscopic evaluation of diaphragmatic motion reduction with a respiratory gated radiotherapy system. J Appl Clin Med Phys 2001; 2:191–200. Shimizu S, Shirato H, Ogura S, et al. Detection of lung tumor movement in real-time tumor-tracking radiotherapy. Int J Radiat Oncol Biol Phys 2001;51:304–310. Ozhasoglu C, Murphy MJ. Issues in respiratory motion compensation during external-beam radiotherapy. Int J Radiat Oncol Biol Phys 2002;52:1389–1399. Koch N, Liu HH, Starkschall G, et al. Evaluation of internal lung motion for respiratory-gated radiation therapy using magnetic resonance imaging (MRI): Part I—Correlation of motion of the internal anatomy with external fiducial. Int J Radiat Oncol Biol Phys 2004;60:1459 –1472. Giraud P, De Rycke Y, Bernard D, et al. Conformal radiotherapy (CRT) planning for lung cancer: Analysis of intrathoracic organ motion during extreme phases of breathing. Int J Radiat Oncol Biol Phys 2001;51:1081–1092.
doi:10.1016/j.ijrobp.2004.05.054
CLINICAL INVESTIGATION
Lung
EVALUATION OF INTERNAL LUNG MOTION FOR RESPIRATORY-GATED RADIOTHERAPY USING MRI: PART II—MARGIN REDUCTION OF INTERNAL TARGET VOLUME H. HELEN LIU, PH.D.,* NICHOLAS KOCH, M.S.,* GEORGE STARKSCHALL, PH.D.,* MARC JACOBSON, M.D.,† KENNETH FORSTER, PH.D.,* ZHONGXING LIAO, M.D.,* RITSUKO KOMAKI, M.D.,* AND CRAIG W. STEVENS, M.D., PH.D.* *Division of Radiation Oncology and †Division of Diagnostic Imaging, The University of Texas M. D. Anderson Cancer Center, Houston, TX Purpose: To analyze the relationship between lung motion and skin surface motion during respiration, determine the uncertainties and variability of such a relationship, and assess the potential of reducing internal target margin for gated radiotherapy. Methods and Materials: Three healthy volunteers and four lung cancer patients were recruited in a prospective imaging study using MRI to track the internal lung and external skin motion during breathing. The relationship between the lung and skin motion was modeled using linear regression analysis. The slope of the linear fit and its confidence interval were analyzed for different lung locations, skin surface locations, and breathing patterns from separate imaging sessions. The margins of the internal target volume were calculated based on the residual lung motion during gating and its uncertainties from multiple treatment fractions for the gated treatment. Results: The slope and confidence interval of the linear regression from the motion analysis were uniquely defined by the locations of the lung, skin surface, and breathing patterns. Statistically significant differences were observed among individuals and between different times of measurement. The normal free-breathing motion averaged from all volunteer and patient data was 13.4 ⴞ 7.4 mm along the superior–inferior (SI) direction and 6.9 ⴞ 2.6 mm along the anterior-posterior (AP) direction. With simulated respiratory gating, the average margin reduction was 5.5 ⴞ 4.8 mm and 1.6 ⴞ 1.0 mm, respectively, along the SI and AP directions (or 36% ⴞ 15% and 25% ⴞ 14%, respectively, relative to free-breathing motion). Conclusion: Because respiratory movement is rather complex, the relationship between the lung and skin surface motion is affected by many anatomic and physiologic factors. The reduction of internal target margin and efficacy of the free-breathing gating technique should be assessed for individual cases. © 2004 Elsevier Inc. Lung motion, MRI, Respiratory gating, Internal target volume, Margin reduction.
because respiration-induced tumor movement causes a large amount of normal lung tissue to be irradiated (2–5). To reduce such a margin, several techniques have recently been developed for radiation beam delivery. Examples include breathing control methods, such as active breathing control (6) or deep inspiration breath-hold (7), and gating techniques, both passive (8, 9) and active (6), some of which involve using implanted internal fiducials (10). The purpose of these techniques is to reduce effectively the degree of tumor motion during RT by either stopping the breathing or synchronizing the radiation with respiration. The respiratory signal can be monitored using surrogate markers acquired from internal or external fiducials or by measuring air volume and/or flow. The effectiveness of the gated treatments depends on a
INTRODUCTION In typical radiotherapy (RT) regimens, the clinical and physical uncertainties encountered in delineating and localizing tumors are currently accounted for by enlarging the target volume with sufficient margins. Report 62 of the International Commission on Radiation Units and Measurements (ICRU) (1) defined four volumes related to the treatment margin: gross target volume, clinical target volume, internal target volume (ITV), and planning targe volume. The concept of the ITV was formally introduced in this ICRU report to describe specifically the geometric uncertainty caused by tumor mobility, namely intra- and interfractional tumor motion. In RT for lung cancer, reducing the ITV margin is important Reprint requests to: H. Helen Liu, Ph.D., Division of Radiation Oncology, Unit 94, The University of Texas M. D. Anderson Cancer Center, 1515 Holcombe Blvd., Houston, TX 77030. Tel: 713-563-2546; Fax: 713-563-2545; E-mail: [email protected] Supported in part by a sponsored research agreement with
Varian Oncology Systems and an internal research grant from M. D. Anderson Cancer Center. Received Dec 11, 2003, and in revised form May 14, 2004. Accepted for publication May 26, 2004. 1473
1474
I. J. Radiation Oncology
● Biology ● Physics
number of factors. The position of the tumor and its surrogates, such as the internal or external fiducials, must have a one-to-one correspondence. It is also critical that such a correspondence be consistent and reproducible during the entire treatment course. Even with gated treatment, an ITV margin would still be required to compensate for the uncertainties of tumor position that primarily come from two sources: the residual tumor motion within the gating window, which normally occupies a fraction of the breathing cycle (e.g., 10 –30%); and possible variations of the gating window because of changes in tumor position among different fractions. Because the respiratory motion could potentially be very complex, as has been demonstrated by some of the existing research (4, 11), the efficacy of the gated treatment depends strongly on whether the lung motion can be monitored and predicted accurately and reproducibly and whether a clinically meaningful reduction of the ITV margin can be accomplished with respiratory gating. The aims of our research were to investigate the motion of internal lung structures and how such motion is related to the motion of an external fiducial and to determine whether the external fiducial can be used reliably to predict the lung motion during a treatment course. We have developed a technique to track the lung and external skin surface movement during breathing using MRI. The details of the imaging and tracking technique have been described in a companion report (12). In the present study, we focused on quantitative analysis of the relationship between the lung and skin movement, specifically how such a relationship may be affected by variables associated with different locations of the lung, skin marker, and breathing patterns. A more important question we attempted to address was whether the relationship is reproducible over time, and consequently, whether the degree of ITV margin reduction was achievable in gated treatment according to the residual tumor motion and interfractional variability. METHODS AND MATERIALS Image acquisition and data analysis An MRI-based imaging technique was developed to acquire dynamic cine images through the sagittal and coronal planes of the studied subjects and to assess the relationship between the internal lung motion and external skin surface motion. Details of the imaging protocol and methods of processing the MRI to measure the lung and skin motion have been previously reported (12). In brief, 7 subjects, 3 healthy young volunteers and 4 lung cancer patients, were recruited into the study. Sagittal and coronal cine MRI scans with a temporal resolution of 0.45 s/frame were acquired through the thoracoabdominal region while subjects performed normal breathing and other altered breathing maneuvers (e.g., chest, abdominal, and deep breathing). The motion of the internal lung vessels, as well as the motion of the exterior skin surface at designated locations (skin 1 to skin 4) were tracked from the cine MRI scans using in-house– developed image processing algorithms.
Volume 60, Number 5, 2004
Thus, the positions of the lung vessels as a function of time, V(t), can be described as a function of that of the skin surface, S(t). This allowed us to model the relationship between the vessel positions and the skin surface positions during breathing directly. On the basis of the results obtained from the breathing traces, simple linear regression analysis was adequate in quantifying such a relationship: V共t兲 ⫽ a ⫹ b · S共t兲
(1)
where V(t) and S(t) are the displacements of the vessel and skin surface from their mean positions at time, t. The slope, b, of the regression line was an indicator of the relationship between the marker and vessel motion, describing the displacement of the vessel position relative to that of the skin over the same period. The uncertainty of the relationship between the skin and lung motion can be described by the 95% confidence intervals of the slope values computed from the regression analysis. An example of such a linear regression analysis is given in Fig. 5 in Koch et al. (12). Because a typical RT course consists of 30 – 40 treatment fractions, the relationship between the skin and lung motion could potentially change over time. To assess such a possibility, we repeated the image acquisition with an approximately 2-week interval for all studied subjects. For the patients, the two imaging sessions usually occurred near the middle and end of the treatment course. The same image processing procedures were applied for the images acquired from both sessions. The slope value b and its confidence interval of the two imaging sessions gave us additional estimates of the uncertainty and possible variation of the skin–lung relationship during the 2–3-week interval between the image sessions. Assessment of margin reduction for gated treatment As mentioned earlier, if a gating technique were to be used for treating lung on the basis of the skin surface motion, the internal margin required for treatment must include two components. The first component is the intrafractional motion, which is contributed by residual tumor motion within the gating window occupying a fraction of the breathing cycle. The second component is the interfractional motion, which comes from the position change of the gating window from fraction to fraction. At our clinic, the gating beam-on window is selected near the end-expiration phase with a duty cycle of approximately 25% (i.e., the duration of such a gating window is approximately onequarter of a breathing cycle). The internal margin is defined as the component of the margin due to the internal lung motion only, excluding setup error uncertainties. Thus, the internal margin, or the range of lung motion within this gating window from two separate sessions, m1 for session 1 and m2 for session 2, can be calculated as m 1 ⫽ 共V max,1 ⫹ 1兲 ⫺ 共V min,1 ⫺ 1兲 m 2 ⫽ 共V max,2 ⫹ 2兲 ⫺ 共V min,2 ⫺ 2兲
(2)
in which Vmin and Vmax are the minimal and maximal vessel
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1475
Fig. 1. (a) Relationship between lung vessel movement and skin 4 movement during normal breathing from two imaging sessions (a) session 1; (b) session 2. (c) Margin calculation for two imaging session (session 1 and 2, solid and dashed lines, respectively). Minimal and maximal vessel excursion from two imaging sessions (Vmax,1 ⫹ 1, Vmax,2 ⫹ 2, and Vmin,1 ⫺ 1, Vmin,2 ⫺ 2) within gating window also shown.
positions within the gating window, respectively, for each session; is the 95% confidence interval of the position due to the uncertainty of the vessel positions. Here the subscript 1 or 2 denote the results from the two imaging sessions because the extent of the vessel motion may change over time during the treatment course. The above margin estimation attempts to enclose the range of lung motion within the 95% confidence interval by including the terms in the equation. The total margin accounting for the results from the two imaging sessions can then be calculated as m 1,2 ⫽ V max,1,2 ⫺ V min,1,2 in which
(3)
V max,1,2 ⫽ max关共V max,1 ⫹ 1兲, 共V max,2 ⫹ 2兲兴 V min,1,2 ⫽ min关共V min,1 ⫺ 1兲, 共V min,2 ⫺ 2兲兴
(4)
where Vmin,1,2 and Vmax,1,2 are the minimal and maximal excursion of the lung motion from the two sessions, respectively. To illustrate the margin calculation, Fig. 1 shows an example of the superior–inferior (SI) displacement of a vessel versus the displacement of the skin surface at the abdominal marker skin 4. Also shown are the lines of best fit (solid) and the 95% confidence interval lines (dashed) of the residuals from the fit. Note difference in the slope indicating the relationship between vessel and skin surface
1476
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Fig. 2. Effect of vessel location on regression slope of vessel–skin relationship from Volunteer 1. Results of tracking five vessels at skin 4 location during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
motion may be slightly different between the two scanning sessions. The two regression lines and their confidence lines in Fig. 1a and Fig. 1b are superimposed in Fig. 1c, assuming the mean positions of the skin marker or lung vessels were identical in both imaging sessions. This assumption was used because the mean position of the skin marker or lung vessels caused by the setup error can be excluded, because the setup error is not included in the ITV margin estimation. Figure 1c illustrates how the internal margin was estimated from the data of the two imaging sessions. Vmin ⫺ and Vmax ⫹ , as defined by the above equations, are illustrated by the arrows for each session. Although each session had its own margin window defined by Eq. 2, the largest difference between Vmin ⫺ and Vmax ⫹ from the two sessions gives the margin size, defined by Eq. 3, necessary to account for the displacement of the lung vessels from both sessions according to the skin marker location. Ideally, the relationship between the vessel and skin movements should be assessed for each fraction of the treatment to obtain the statistical distribution of such a relationship. Here, we used data from only two imaging sessions as a starting point to illustrate the principle of the margin estimation and the effect of motion variability over time. For each vessel tracked from the volunteers and patients, we computed the normal free-breathing motion, as well as the motion within the gating window, from two imaging sessions. Using the above approach, we, thus, estimated the margin reduction in both absolute (millime-
ters) and proportional (percentages) terms relative to the free-breathing motion.
RESULTS See Figures 2–10 for summary of the observations made on the relationship between lung vessel motion and skin motion, and more importantly, the dependency of such a relationship on the vessel and skin locations and the variation of the relationship with the studied subjects and time of measurement. We used the results taken from Volunteer 1 as a representative case for all studied subjects (see Figs. 2– 4). In Fig. 2, the results of linear regression analysis (Eq. 1), namely, the slope value b and its confidence interval, are displayed for the five vessels tracked for this volunteer from two separate imaging sessions. The skin surface motion at marker 4 and normal breathing data were used in this analysis. Figure 2 shows that the slope values are quite different among the five vessels along either the SI or AP directions. The difference was because, for the same magnitude of skin 4 movement, the degree of motion of the five vessels was distinctively different owing to their locations. Along the SI direction, the V1-R82b and V1-L98a, which are located in the lower lung lobe, had greater motion, and thus, greater slope values, compared with other vessels, particularly in contrast to V1-R82a, located more superiorly. Along the AP direction, the degree of vessel motion, in general, was less than that of the SI direction, leading to smaller slope values.
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1477
Fig. 3. Effect of skin surface location on regression slope of vessel–skin relationship from Volunteer 1. Results of tracking two vessels at skin 1 to skin 4 locations during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) vessel motion along AP direction.
Vessel V1-R82b and V1-L98a, again, showed larger slope values than the other vessels owing to the greater mobility of these two vessels. More noteworthy was the statistically significant difference in the slope values, which exceeded their confidence intervals from the two imaging sessions, indicating that the relationship between the skin 4 and vessel movement had changed between the two sessions, particularly along the SI direction for the first four vessels. Thus, for the same amount of skin movement, the degree of vessel movement could be different, possibly depending on how exactly the studied subject was performing the regular breathing during the different imaging sessions. Figure 3 further illustrates the results from all four skin locations for the first two vessels only, V1-R82a and V1R82b. Again, the data of the normal breathing and from two imaging sessions are included. For the same vessel, the slope values were also quite different at the four skin locations. This was because, with the same amount of vessel movement, the degree of skin movement was not uniform throughout the anterior skin surface. The slope values at skin 4 tended to be the smallest along either the SI or AP direction, because skin 4 had the greatest amount of movement. The slope values for skin 2 tended to be the greatest because it had the least amount of movement. In addition, the error bars of the slope values were associated with the uncertainty and degree of correlation between the vessel and skin movement (i.e., the larger the error bar, usually the poorer the correlation). The results suggested that for gated treatment using external markers, skin 4 may be a better
location owing to its greater range of movement and, thus, a greater sensitivity for detecting the vessel movement. In contrast, skin 2 may not be an ideal location because it had the smallest degree of movement during breathing and a poor correlation along the SI direction. Figure 4 demonstrates the comparison of four different breathing patterns for vessels V1-R82a and V1-R82b. Only the data from skin 4 movement were included. The results showed that the relationship between the vessel and skin 4 movement was also unique for an individual breathing pattern along both SI and AP directions. During altered breathing, the lung expansion and use of muscles could be quite different compared with normal breathing, leading to more complex movement and, thus, a more complex relationship between the lung and skin surface movement. For example, comparing abdominal and chest breathing, the slope values tended to decline along the SI direction but increase along the AP direction. This difference in slope values was possibly due to reduced SI lung motion and magnified AP lung expansion during chest breathing. Because the relationship between the lung and skin movement can be quite different depending on the breathing pattern, even for regular normal breathing, the slope values from the two imaging sessions could also change, particularly for vessel V1-R82b, which was more inferiorly located than V1-R82a. This means that it might be very difficult to reproduce exactly the same lung and skin movement during each imaging session at different times, even for healthy volunteers.
1478
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Fig. 4. Effect of breathing pattern on regression slope of vessel–skin relationship from Volunteer 1. Results of tracking two vessels at skin 4 location during normal, abdominal, chest, and deep breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
Figure 5 summarizes the lung movement during normal breathing from three different vessels, at skin 2, 3, and 4, as an example of a typical lung cancer patient
(Patient 3). The uncertainty of the slope values was greater, particularly at the skin 2 location for this patient because of a poor correlation between the vessel and skin
Fig. 5. Regression slope of vessel–skin relationship from Patient 3. Results of tracking three vessels at skin 2, 3, and 4 locations during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1479
Fig. 6. Intersubject variation of regression slope of vessel–skin relationship from three individuals (Volunteers 1 and 3 and Patient 3). Results of tracking three vessels at skin 4 location during normal breathing from two imaging sessions included. Error bars show 95% confidence intervals of regression slope. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
movement. Similar to that observed for the volunteers, the relationship between the lung and skin movement depended on the location of the vessel, as well as of the skin marker. Again, the patient data demonstrated statistically significant differences in slope values between the two imaging sessions that were greater than the confidence intervals. Comparing the results of skin 2, 3, and 4,
in general, we noted that the confidence intervals of the slope values tended to be smaller at skin 4. The reproducibility of the slope values from two imaging sessions was also slightly improved at this location from other volunteer and patient data. Thus, skin 4 seemed to be the preferred location if a single external fiducial has to be chosen for monitoring the respiratory cycle because of
Fig. 7. Distribution of vessel motion during normal free breathing from all vessels tracked of volunteers. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
1480
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Fig. 8. Distribution of vessel motion during normal free breathing from all vessels tracked of patients. (a) Vessel motion along SI direction. (b) Vessel motion along AP direction.
the sensitivity of skin 4 in tracking lung motion and the consistency of the correlation from different imaging sessions. Figure 6 compares the results from tracking three vessels located at approximately same location of the lung from three different studied subjects: Volunteers 1 and 3 and Patient 3. The slope values using skin 4 during normal breathing from two imaging sessions are demonstrated. Even for the same region of the lung and skin marker location, the lung and skin movement and their relationship were unique to the individual and exhibited a large degree of intersubject dependency. Thus, the degree of lung movement as monitored by the skin movement should be assessed carefully for an individual, possibly because of the differences in the breathing pattern among the population, especially for lung cancer patients who have impaired lung function. Figures 7 and 8 summarize the magnitude of the lung motion based on tracking all the vessels in volunteers and patients, respectively, during normal breathing. The results are also included in Table 1. For volunteers, the SI and AP
motion was 11.0 ⫾ 5.3 mm and 6.5 ⫾ 2.2 mm, respectively. For patients, the distribution of free-breathing motion was not statistically different from that of the volunteers, but the sample size was small (p ⬎0.2, Kolmogorov-Smirnov test), although the average magnitude of the motion was slightly greater (18.1 ⫾ 9.5 mm and 7.8 ⫾ 3.5 mm along the SI and AP directions, respectively). For both the volunteers and patients, the primary axis of motion was in the SI direction. For comparison purposes, the results of the vessel motion during deep breathing were also included in Table 1. Deep breathing caused similar changes in SI motion for both patients and volunteers, but volunteers seemed to be able to increase their AP motion more than the patients. Figures 9 and 10 summarize the degree of margin reduction achievable according to the data from two imaging session using Eqs. 2– 4. Skin 4 was used as the location of the external fiducial in tracking the vessel movement; for the reasons discussed above. The data in Fig. 9 suggest that, for volunteers, along the SI direction, the possible margin reduction varied significantly from 1 to 11 mm or 20 –50%, depending on the individual vessels. Along the AP direc-
Table 1. Possible margin reduction with gating using skin 4 to track vessel movement during breathing Volunteers
Patients
SI
Normal breathing (mm) Margin reduction (mm) Margin reduction (%) Deep breathing (mm)
AP
SI
AP
Mean
SD
Mean
SD
Mean
SD
Mean
SD
11.0 4.2 35.0 28.4
5.3 3.1 15.3 13.8
6.5 2.0 29.6 21.9
2.2 1.0 12.3 6.3
18.1 8.0 38.8 25.4
9.5 6.4 16.1 12.5
7.8 1.0 15.4 12.9
3.5 0.9 11.8 5.1
Abbreviations: SI ⫽ superior–inferior; SD ⫽ standard deviation.
MRI evaluation of internal lung motion for respiratory gating: II
● H. H. LIU et al.
1481
Fig. 9. Distribution of margin reduction with gating at skin 4 location from all vessels tracked of volunteers. (a) Absolute margin reduction and (b) relative margin reduction compared with normal free breathing.
tion, the degree of absolute margin reduction was quite reduced owing to a smaller degree of AP lung motion and uncertainties associated with tracking the motion. For patients, Fig. 10 shows that the degree of AP margin reduction was curtailed further compared with that of the volunteers, because the correlation between the lung and skin movement was quite poor along this direction (see Fig. 13 of Koch et al. [12]). Along the SI direction, a more significant margin reduction was possible for one patient (Patient 3) that showed excellent correlation between the SI and skin 4 motion (see Fig. 7 of Koch et al. [12]). This suggests that certain patients might be more suitable for, and will realize a greater benefit from, gated treatment. Even for this particular group of patients, margin reduction might be more feasible along the SI direction than the AP direction. The results from Figs. 9 and 10 indicated that the distributions of the margin reduction were not statistically different between the volunteers and patients (p ⬎0.05, Kolmogorov-Smirnov test). The mean and standard deviation of the absolute and
relative margin reduction are also given in Table 1. The average margin reduction was 5.5 mm ⫾ 4.8 mm and 1.6 mm ⫾ 1.0 mm along the SI and AP directions, respectively, corresponding to 36% ⫾ 15% and 25% ⫾ 14%, respectively, relative to free-breathing motion. Compared with the SI direction, the margin reduction along the AP direction was much smaller and might not lead to meaningful clinical reductions of normal tissue volume in treating lung tumors. DISCUSSION In this research, we analyzed the relationship between the lung and skin movement during breathing for healthy volunteers and patients recruited in a prospective imaging study to evaluate the efficacy and improvement of the respiratory gated RT for lung cancer. In our companion report (12), the details of the MRI technique and image processing procedure were given, along with the results from analyzing the correlation between the lung and skin
Fig. 10. Distribution of margin reduction with gating at skin 4 location from all vessels tracked of patients. (a) Absolute margin reduction and (b) relative margin reduction compared with normal free breathing.
1482
I. J. Radiation Oncology
● Biology ● Physics
Volume 60, Number 5, 2004
Table 2. Motion of skin markers during normal and deep breathing from volunteers and patients Volunteers
Patients
Normal
Deep
Normal
Deep
Skin marker (mm)
Mean
SD
Mean
SD
Mean
SD
Mean
SD
1 2 3 4
3.3 3.2 4.1 6.8
1.1 1.0 1.1 3.1
21.5 17.6 17.6 23.7
5.6 4.1 3.9 7.8
— 3.4 5.3 9.7
— 0.9 2.4 4.0
— 8.6 13.7 21.4
— 1.7 6.7 6.2
Skin markers were tracked at sagittal planes at which MRI scans were taken. Abbreviation: SD ⫽ standard deviation.
movement. We found that the SI motion of the lung, particularly in the lower lobe, can be better described by abdominal skin surface movement, but the AP lung expansion correlated better with chest marker movement. Complimentary to the findings in our companion report, in the current paper, we focused on using the linear regression technique to quantify the relationship between the lung and skin movement and its dependency on several variables, including the location of the lung, skin surface, breathing pattern, subject, and time of measurement. Our results demonstrated that along the SI direction the relationship between the lung and skin movement could be described well by a simple linear model. However, along the AP direction, the lung movement was more complex, with a possibility of hysteresis (see Fig. 7 of Koch et al. [12]). This resulted in a larger degree of uncertainty, reflected by a greater confidence interval of the slope value along the AP direction. Although more sophisticated mathematic models, other than linear regression, may be used to describe such a relationship, their necessity was not fully warranted because we were more interested in quantifying the variability of such a relationship and its effect on margin reduction. The results from healthy volunteers and patients demonstrated that, apparently, various regions of the lung move quite differently during breathing. The upper lobe of the lung exhibited expansion mainly along the AP direction, and the lower lobe of the lung moved along the SI direction driven by the diaphragm. Consequently, the relationship between the internal lung parenchyma and external skin movement at different regions will be uniquely dependent on the relative locations of the lung and skin surface. As indicated by Figs. 2 and 3, using data of Volunteer 1 as an example, the slope values reflected the degree of lung motion relative to the skin surface motion during breathing. A greater degree of mobility was observed for lower lobe vessels along the SI direction and upper lobe vessels along the AP direction. In patients, lung motion may be complicated further by abnormal breathing patterns caused by patient-specific obstructive or restrictive effects of pulmonary disease. Additionally, the body anterior surface also moves with a different magnitude with lung expansion because of regional muscle involvement. Table 2 summarizes the results
of the fiducial movement at different locations during normal and deep breathing. The upper abdominal surface such as skin 4 had the greatest movement (6.8 ⫾ 3.1 mm, and 9.7 ⫾ 4.0 mm, respectively, for volunteers and patients). Thus, skin 4 was a more sensitive location to track the lung motion, particularly along the SI direction. Because the lung and skin surface movement are also greatly affected by the breathing pattern, their relationship is strongly dependent on the breathing maneuver and exactly how the lung expands and contracts during breathing. Therefore, a significant degree of variability of the lung– skin relationship is due to changes in the breathing patterns (Fig. 4), which is also the primary cause for differences in such a relationship observed from separate imaging sessions. Certain vessels also exhibited a greater variability than others (e.g., V1-R82a vs. V1-R82b [Fig. 4]), indicating that certain portion of the lung may be more reproducible to track using external fiducials than others. Similarly, certain skin markers tended to produce more consistent results than others. For example, skin 4 seemed to be a more preferred location than skin 2 (Fig. 5), if a single fiducial had to be chosen for gating purposes. This conclusion also compliments the previous observations made in our companion report (12); in a sense, the SI motion, which is the primary source of lung movement, correlated better with the upper abdominal skin movement. Owing to the uncertainties and variability associated with the lung–skin movement, the degree of margin reduction may not be as great as expected even if skin 4 were used to track the lung motion. The data presented in Figs. 9 and 10 suggest that the margin reduction can be more easily accomplished along the SI direction for both volunteers and patients, again because of the stronger and more consistent correlation between skin 4 and SI motion of the lower lobe lung. However, along the AP direction, the degree of the correlation and its uncertainty make the margin reduction virtually impossible for a quite a few studied subjects. Currently, most of the radiation beams in treating lung cancer are coplanar beams involving the AP direction. Because the SI movement of the lung is the primary source of motion with a greater magnitude (as indicated by the data shown in Figs. 7 and 8), gated treatment could be still very beneficial for a fraction of the patient population. The key
MRI evaluation of internal lung motion for respiratory gating: II
issue is that we need to identify such individuals with careful imaging and apply appropriate margin reduction if gated treatment is chosen. Reliable means of monitoring the lung motion will be critical for gated treatments, particularly for longer treatment courses during which the possibility of changes in respiratory and other physiologic functions will be greater. The margin calculations presented here were the estimated internal margins described in ICRU Report 62 (1). The method of computing this margin used the image data from two imaging sessions only, which represents very limited sampling during a treatment course. Ideally, multiple sessions would be needed to access fully the range of lung motion and margin reduction. However, for practical reasons, our purpose was to demonstrate the principle of margin calculation and the effect of inter- and intrafractional motion. The actual margin reduction that could be achieved during a full treatment course is likely to be even less than what we estimated from only two fractions separated by about 2–3 weeks. Using the cine MRI scans, the internal lung motion was measured using pulmonary vessels as landmarks and surrogates for lung structure. The limitations of the MRI technique have been addressed in our companion paper (12). Mainly, the sagittal MRI scans we used to analyze the lung and skin positions could only provide motion along the SI and AP directions. The lateral movement of the lung could be acquired using coronal images, although they did not contain information on the simultaneous skin motion. However, because respiration mainly involves chest expansion
● H. H. LIU et al.
1483
and diaphragmatic movement, lateral lung movement could be a relatively minor component compared with the SI and AP lung movement, the results of which have been also shown by existing imaging studies (4, 13). Thus, we expect that the degree of motion and associated margin reduction could be less in the lateral dimension than the results we report for the SI and AP dimensions. Our results were also confined to those regions of the lung in which the vessels could be identified. Thus, our observations were near the lung hila. Additionally, because lung cancer will cause significant anatomic and physiologic changes to the lung parenchyma, the motion of the lung tumor may well be different from that of the vessels. However, the above results may be applicable to small, welllocalized, and free-floating nodules similar to pulmonary vessels. The motion of large lung tumors, especially along their edges, may be more complex, making it even more difficult to track using exterior markers and reduce the treatment margin. The determination of the motion of large tumors will be more challenging than that of the vessels because the borders of a lung tumor are usually difficult to distinguish from the surrounding tissues such as pericardium. Also, different regions of large tumors may move relatively independently (i.e., mediastinal nodes may move mostly along an AP axis, hilum nodes may move in an arc, and the primary tumor may move along an SI axis). However, we have developed a technique of tracking of lung tumor motion from cine MRI that will facilitate the study of such complex motion in the future.
REFERENCES 1. International Commission on Radiation Units and Measurements. ICRU report 62: Prescribing, recording and reporting photon beam therapy (supplement to ICRU report 50). Bethesda: International Commission on Radiation Units and Measurements; 1999. 2. Ross CS, Hussey DH, Pennington EC, et al. Analysis of movement of intrathoracic neoplasms using ultrafast computerized tomography. Int J Radiat Oncol Biol Phys 1990;18: 671–677. 3. Chen Q, Weinhous MS, Deibel C, et al. Fluoroscopic study of tumor motion due to breathing: Facilitating precise radiation therapy for lung cancer patients. Med Phys 2001;28:1850– 1856. 4. Seppenwoolde Y, Shirato H, Kitamura K, et al. Precise and real-time measurement of 3D tumor motion in lung due to breathing and heartbeat, measured during radiotherapy. Int J Radiat Oncol Biol Phys 2002;53:822–834. 5. Stevens C, Munden R, Forster K, et al. Respiratory-driven lung tumor motion is independent of tumor size, tumor location, and pulmonary function. Int J Radiat Oncol Biol Phys 2001;51:62–68. 6. Wong JW, Sharpe MB, Jaffray DA, et al. The use of active breathing control (ABC) to reduce margin for breathing motion. Int J Radiat Oncol Biol Phys 1999;44:911–919. 7. Rosenzweig K, Hanley J, Mah D, et al. The deep inspiration
8.
9.
10.
11.
12.
13.
breath-hold technique in the treatment of inoperable nonsmall-cell lung cancer. Int J Radiat Biol Phys 2000;48:81–87. Kubo HD, Len PM, Minohara S, et al. Breathing-synchronized radiotherapy program at the University of California Davis Cancer Center. Med Phys 2000;27:346–353. Mageras GS, Yorke E, Rosenzweig K, et al. Fluorscopic evaluation of diaphragmatic motion reduction with a respiratory gated radiotherapy system. J Appl Clin Med Phys 2001; 2:191–200. Shimizu S, Shirato H, Ogura S, et al. Detection of lung tumor movement in real-time tumor-tracking radiotherapy. Int J Radiat Oncol Biol Phys 2001;51:304–310. Ozhasoglu C, Murphy MJ. Issues in respiratory motion compensation during external-beam radiotherapy. Int J Radiat Oncol Biol Phys 2002;52:1389–1399. Koch N, Liu HH, Starkschall G, et al. Evaluation of internal lung motion for respiratory-gated radiation therapy using magnetic resonance imaging (MRI): Part I—Correlation of motion of the internal anatomy with external fiducial. Int J Radiat Oncol Biol Phys 2004;60:1459 –1472. Giraud P, De Rycke Y, Bernard D, et al. Conformal radiotherapy (CRT) planning for lung cancer: Analysis of intrathoracic organ motion during extreme phases of breathing. Int J Radiat Oncol Biol Phys 2001;51:1081–1092.