Int. J. Radiation Oncology Biol. Phys., Vol. 65, No. 5, pp.

1560 1571, 2006 Copyright 2006 Elsevier Inc. Printed in the USA. All rights reserved 0360-3016/06/$see front matter

doi:10.1016/j.ijrobp.2006.04.031

PHYSICS CONTRIBUTION

MID-VENTILATION CT SCAN CONSTRUCTION FROM FOUR-DIMENSIONAL RESPIRATION-CORRELATED CT SCANS FOR RADIOTHERAPY PLANNING OF LUNG CANCER PATIENTS JOCHEM W. H. WOLTHAUS, M.SC., CHRISTOPH SCHNEIDER, PH.D., JAN-JAKOB SONKE, PH.D., MARCEL VAN HERK, PH.D., JOS S. A. BELDERBOS, M.D., MADDALENA M. G. ROSSI, D.C.R.(R), R.T.T., JOOS V. LEBESQUE, M.D., PH.D., AND EUGNE M. F. DAMEN, PH.D.
Department of Radiation Oncology, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands Purpose: Four-dimensional (4D) respiration-correlated imaging techniques can be used to obtain (respiration) artifact-free computed tomography (CT) images of the thorax. Current radiotherapy planning systems, however, do not accommodate 4D-CT data. The purpose of this study was to develop a simple, new concept to incorporate patient-specic motion information, using 4D-CT scans, in the radiotherapy planning process of lung cancer patients to enable smaller error margins. Methods and Materials: A single CT scan was selected from the 4D-CT data set. This scan represented the tumor in its time-averaged position over the respiratory cycle (the mid-ventilation CT scan). To select the appropriate CT scan, two methods were used. First, the three-dimensional tumor motion was analyzed semiautomatically to calculate the mean tumor position and the corresponding respiration phase. An alternative automated method was developed to select the correct CT scan using the diaphragm motion. Results: Owing to hysteresis, mid-ventilation selection using the three-dimensional tumor motion had a tumor position accuracy (with respect to the mean tumor position) better than 1.1 1.1 mm for all three directions (inhalation and exhalation). The accuracy in the diaphragm motion method was better than 1.1 1.1 mm. Conventional free-breathing CT scanning had an accuracy better than 0 3.9 mm. The mid-ventilation concept can result in an average irradiated volume reduction of 20% for tumors with a diameter of 40 mm. Conclusion: Tumor motion and the diaphragm motion method can be used to select the (artifact-free) midventilation CT scan, enabling a signicant reduction of the irradiated volume. 2006 Elsevier Inc. Computed tomography, Four-dimensional, Respiration-correlated, Breathing, Treatment planning, Lung cancer, Mid-ventilation.

INTRODUCTION Treatment outcomes for nonsmall-cell lung cancer using conventional radiation doses are poor, and the local recurrence rate is high. Tumor control can probably be improved by increasing the dose (e.g., 1, 2). However, the surrounding healthy lung tissue and the esophagus are dose limiting (e.g., 3, 4). To enable dose escalation, the irradiated surrounding normal tissue volume should be minimized. One possible approach is to reconsider the margins for the conventional planning target volume (PTV). The PTV is generally dened as the clinical target volume (CTV) (visible tumor plus margin for microscopic extensions) plus a marReprint requests to: Eugne M. F. Damen, Ph.D., Department of Radiation Oncology, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, Amsterdam 1066 CX, The Netherlands. Tel: ( 31) 20-512-2205; Fax: ( 31) 20 669 1101; E-mail: e.damen@nki.nl Supported by Grant NKI 03-2943 from the Dutch Cancer Society.
1560

gin to account for geometric uncertainties (5). The most important interfractional geometric uncertainties are patient setup errors, tumor shrinkage or growth, and respiratory baseline shifts (i.e., shifts in respiration levels). Intrafractional geometric uncertainties are due to respiratory and cardiac motion. Eliminating geometric uncertainties allows for a reduction of the CTV to PTV margins, reducing the volume of irradiated normal tissue and normal tissue complication probability and enabling dose escalation. A single free-breathing computed tomography (CT) scan is often used for radiotherapy planning for lung tumors. However, respiration-induced tumor motion (TM) during acquisition causes artifacts in tumor shape and position (6,
AcknowledgmentsThe authors thank Jasper Nijkamp for his clinical software development and Harry Bartelink, Ben Mijnheer, and Leah McDermott for critical reading of the manuscript. Received Jan 6, 2006, and in revised form April 10, 2006. Accepted for publication April 10, 2006.

Clinical implementation of 4D-CT scans

J. W. H. WOLTHAUS et al.

1561

7). The cause of these artifacts is that the CT scanner acquires a stack of images without time information from the moving tumor, thus obtaining a set of arbitrary snapshots of moving structures. To overcome this problem, time-resolved four-dimensional (4D) scanning techniques were developed (8 13). Basically, all these methods are similar i.e., the acquired oversampled three-dimensional (3D) data are sorted by the patients respiratory phase using an external breathing signal, yielding a set of 3D reconstructions at different breathing phases. This set (i.e., 4D image data) provides temporal and spatial motion information that can be used to optimize treatment planning. Currently, however, available commercial treatment planning systems cannot handle a 4D-CT data set as input for treatment planning. Several strategies for the implementation of 4D data sets in the treatment planning process have been published: Slow CT scan: Srensen de Koste et al. (14) used a slow CT scan (which can be considered as the time average of the 4D data set) to delineate the target volume encompassing the tumor at any position during the breathing cycle (internal target volume). A 5 mm expansion was added to the internal target volume to cover the CTV and a 5-mm expansion was added from the CTV to the PTV. Maximal inhale and exhale composite delineation: Allen et al. (15) studied the delineation of the gross tumor volume (GTV) in a maximal inhale scan and a maximal exhale scan and created a composite of the two delineation sets for planning (GTV to CTV margin recipe was not given). All-phase composite segmented tumor: Rietzel et al. (16) and Underberg et al. (17) proposed a method (based on maximal intensity projection) that automatically creates a composite segmented CT scan of all respiratory phase scans (the full 4D set). From this composite CT scan, the composite GTV was delineated (GTV to CTV margin recipe was not given). Breath-hold CT scan (with active breathing control): Wong et al. (18) and Rosenzweig et al. (19) studied breath-hold CT acquisition using the active breathing control device and voluntary breath-hold, respectively, to obtain a single 3D scan without motion artifacts for treatment planning purposes (GTV to CTV margin recipe was not given). Deformation dose mapping: Keall et al. (20) proposed a 4D dose calculation method based on 4D deformation maps from the 4D-CT scan. The deformation maps were applied to the peak inhale delineations of the tumor and other structures. For each respiratory phase of the 4D set, an optimized treatment plan was calculated. The rst three methods can be summarized as composite target volume concepts, in which the target volumes are often enlarged. Moreover, delineation of moving structures in the slow CT method is difcult because of motion blurring. The second and third method need multiple reconstructions for tumor delineation and dose calculation. The breath-hold method is not applicable to all lung cancer patients because of their physical condition. Moreover, breathing control also

needs to be maintained during treatment. The effectiveness of the last method, by Keall et al. (20), was demonstrated in a feasibility study, but its clinical implementation is not yet possible because of hardware limitations. In this report, we have proposed the selection of a single well-chosen CT scan from a 4D set. The choice of the single CT scan was based on studies by Engelsman et al. (21) and Witte et al. (22). These studies showed that if the tumor is irradiated at its average position during the respiration cycle, because of the presence of the wide-beam penumbra in the lung, good dose coverage would still be obtained even if the tumor was not fully within the high-dose region for a small part of the breathing cycle (0.6% and 6% tumor control probability loss for 5 mm and 15 mm motion amplitude with 0 margin, respectively, and no setup errors). Such an approach makes margin reduction possible. The aim of this study was to eliminate the systematic errors in the imaging process induced by respiratory motion and to obtain a more representative scan for delineation of the target area, normal structures, and dose calculation. In the presented approach, motion was not taken into account when delineating the tumor (thus no internal target volume) but will be incorporated in the margin expansion from CTV to PTV. This margin included the dose blurring caused by respiratory motion during treatment. See Fig. 1 for the clinical protocol for 4D analysis and scan selection currently used in our hospital METHODS AND MATERIALS Patient group
Patients with a lung tumor who were scheduled for radiotherapy underwent uoroscopy to estimate the TM due to respiration. If the tumor was estimated to move 0.5 cm owing to respiration, the patients were eligible for this study. All patients (12 men and 3 women) included in the study gave written informed consent to participate. The local medical ethics committee approved the study.

Respiration-correlated 4D-CT imaging

4D-CT scanning. During 4D-CT scanning, the patient was instructed to breath freely and normally. Patient respiration was registered using a thermocouple (Type T, copper-constantan, S-CC-U-O7/1, Volenec, Hradec Krlov, Czech Republic) inserted into the entry of a regular oxygen mask, which measures temperature changes in the airow during inhalation (cold) and exhalation (warm). During CT scanning, patients were positioned supine with their arms raised above their head using an in-house developed forearm support. To obtain a position similar to that on the treatment couch, a at tabletop (Sinmed, Reeuwijk, The Netherlands) was placed on the couch of the CT scanner (20-slice Somatom Sensation Open, Siemens, Forchheim, Germany). The helical cardiac scanning mode of the CT scanner was used for the respiration-correlated imaging. However, the thermocouple respiratory signal was used for data sorting, instead of the cardiac input signal. The scans were made at 120 kV, 1000 eff.mAs. The complete thorax was scanned (30 35 cm) from 5 cm above the top of the lung down to 5 cm below the diaphragm in the inhale position. The scan time was 60 70 s. Scan reconstruction. The beam-on signal of the CT scanner was

1562

I. J. Radiation Oncology

Biology Physics

Volume 65, Number 5, 2006

at the CT scanner

after scanning

oxygen mask with thermocouple is placed on the patient

examine tumor motion (TM) of 4D scan visually

pre-acquisition of respiration signal

< ~ 2 cm

cranio-caudal TM amplitude?

> ~ 2 cm

compute mean & std.dev. cycle length

determine diaphragm motion

determine (cranio-caudal) tumor motion

std.dev. cycle length? <~2s <5s mean+std.dev. cycle length?

>~2s compute mid-ventilation time-percentage during exhale >5s reconstruct MidV CT scan

set CT gantry rotation to 0.5 s pitch = 0.10

set CT gantry rotation to 1.0 s pitch = 0.15

verify visually in green/purple composite viewer if scan is adequate

scan patient

reconstruct scan

Fig. 1. Considerations presented in this report can be summarized in the following clinical protocol currently in use at our institution. The protocol is divided into two parts: during and after scanning. 4D four-dimensional; MidV midventilation. used to synchronize the respiration signal with the CT data. The time delay (response time) between the movement of the internal structures of the lung and the external (thermocouple) breathing signal (0.4 s, see Appendix) was corrected by back shifting the respiratory signal with respect to the beam-on signal. After this time-delay correction, the respiratory signal should, in principle, be in phase with the CT data of the internal structures. The respiration-correlated protocol on the scanner, an adaptation of the cardiac protocol, uses the peaks (maximal inhalation) of the respiration signal to sort the raw CT data. Between the peaks, the respiratory cycle is divided into 10 equidistant time-percentage bins (0% at maximal inhalation to 90%). Within one cycle, the time bins are equidistant, but from cycle to cycle, the exact length of a certain bin may vary owing to variations in the breathing period. Using equidistant time bins, which is equal to linear data sampling, the asymmetry between inhalation and exhalation length is incorporated in the fourth dimension of the resulting 4D scan by having a different number of generally more3D-CT frames in the exhalation phase than in the inhalation phase. For each table position (every 3 mm) and time-percentage within the breathing cycle, the corresponding location in the CT sinogram (as a function of table position and gantry angle) was determined. A slice was reconstructed using data from 110 to 110 (180 plus fan angle) from the determined gantry angle. The number of slices within one time bin was approximately 100 for the chosen slice thickness of 3 mm. Choice of CT acquisition parameters. The quality of 4D scanning in the helical mode depends critically on the interplay between the breathing characteristics of the patient and the acquisition parameters used for scanning. According to Ford et al. (9), complete coverage of moving structures over a full breathing cycle is obtained when CLGantry-rotation pitch CLRespiration, where CLRespiration is the breathing cycle length of the patient and CLGantry-rotation is the gantry rotation time. If the CLGantry-rotation is shorter than the pitch CLRespiration, undersampling of the data will occur, leading to gaps in the 4D data for particular phases. If the CLGantry-rotation is longer, oversampling will occur and motion will be blurred in the slice. The Somatom 20-slice scanner allows gantry rotation times of 0.5 and 1.0 s. We therefore used two different sets of acquisition parameters. When the average CLRespiration 1 standard deviation was less than 5 s, a CLGantry-rotation of 0.5 s and pitch of 0.1 was used, otherwise a CLGantry-rotation of 1.0 s and pitch of 0.15 was used. This procedure prevented undersampling of a breathing period up to 6.7 s. The

Clinical implementation of 4D-CT scans

(a) Concept Maximum inspiration Tumor trajectory
0%

J. W. H. WOLTHAUS et al.
(c) Simplification

1563

(b) Simplification Tumor trajectory

0%

TM t%midv-exhale t%midv-inhale t%midv-exhale

TM

t%midv-inhale

Cranio-caudal projection

Maximum inspiration

mean position (TM)

Maximum expiration 0% t%midv-exhale t%midv-inhale 100% Maximum inspiration

Maximum expiration

Anterior-posterior projection

mean (TM) position Maximum expiration

0%

t%midv-exhale

t%midv-inhale

100%

Fig. 2. (a) Schematic overview of trajectory of tumor with its mean position TM and multiple mid-ventilation positions in exhalation (t%midv-exhale) and inhalation (t%midv-inhale) due to hysteresis (multiple closest distances between tumor and mean tumor position). Geometric error with respect to mean position denoted by h. (b) Simplication of concept by considering craniocaudal (CC) movements only. Error in determination of time-percentage midvt% , due to simplication of only using CC motion direction; (c) CC projection of tumor trajectory, which is used to estimate mid-ventilation time-percentages. Because of hysteresis, phase shift exists between motion curves of different directions (c, lower panel), resulting in geometric error h for other directions. degree of oversampling may vary in a single 4D scan, depending on the variability of the breathing cycle length of a patient during scan acquisition. AP movements. The CC motion curve does not give the real minimal distance between the tumor and mean position, but provides an unambiguous single t%midv for exhalation and inhalation. As a consequence of this approach, two types of errors are introduced: rst, a geometric error with respect to the mean position ( h), due to hysteresis in the TM; and second, an error in the determination of the time-percentage ( midvt%), owing to simplication from using the CC motion direction only. Because of hysteresis, phase shifts exist between the motion curves of the different directions (Fig. 2c, lower panel), resulting in a geometric error h for the other directions. To calculate the t%midv, the TM was calculated using the normalized (scaled between 0 and 1) CC motion curve (as inferred from Determination of TM by tumor tracking [see below]), indicated by the horizontal solid line in Fig. 2. Subsequently, the motion curve was interpolated using cubic spline interpolation. The t%midv in the exhalation phase (t%midv-exhale) and inhalation phase (t%midv-inhale) were determined. Finally, the corresponding MidV-CT scan was reconstructed (at t%midv-exhale or t%midv-inhale). Verication of MidV-CT scan selection. To check whether this reconstructed MidV-CT scan was acceptable, the scan was veried by visually inspecting the position of the structures of the MidV-CT scan with respect to the corresponding moving structures in the 4D-CT scan using an in-house developed composite viewer (Fig. 3). In this composite viewer (24), the MidV-CT is represented by the color purple (magenta) and the 4D-CT scan is represented by the color green. When overlapping, regions of corresponding density are then represented in gray scale; unmatched regions will preserve their color. Verifying the geometric

Concept and denition of mid-ventilation scans

To plan the radiation of the tumor in the mean tumor position, we reconstructed a single 3D CT scan from the 4D data set that represents the mobile structures close to their (time-weighted) mean position. This scan was called the mid-ventilation (MidV) CT scan and was used for delineation of the tumor and normal tissue, as well as for treatment planning dose calculation. First, we dened the mean position and mid-ventilation time-percentage (t%midv) (Fig. 2). Denition. The mean position (TM) of a moving tumor or structure is the time-weighted mean position of the center of gravity of that tumor or structure in all three dimensions (leftright [LR], cranio-caudal [CC], and anterior-posterior [AP]). If hysteresis occurs in the TM (23), this mean position is not necessarily on the trajectory of the object (Fig. 2a). Denition. Mid-ventilation time-percentage . . . the t%midv is that time/percentage in the respiration cycle at where the tumor is closest to the time-weighted mean position. In general, two solutions are possible: the mid-ventilation time-percentage can be dened during exhalation and inhalation (Fig. 2a). When the tumor moves in more than one direction (hysteresis), it might be possible that there is more than one closest distance between the tumor and the mean position TM (Fig. 2a). To overcome this ambiguity in the determination of the t%midv, only the CC motion curve was considered (Fig. 2b,c). The CC respiratory movements were usually dominant compared to the LR and

1564

I. J. Radiation Oncology

Biology Physics

Volume 65, Number 5, 2006

Fig. 3. Composite viewer method for visual verication of selected mid-ventilation (MidV) computed tomography (CT) scan. MidV-CT is represented by purple (magenta) and four-dimensional (4D)-CT scan by green. When mixing the two scans, structures that match become white and structures that do not match will preserve their color.

the left from the right lung (Fig. 5c). Because the patient in the example of Fig. 5 had a tumor in the left lung, we considered the lower left part of the 4D-CT scan (Fig. 5d). This selection was cropped to the lung region, subsequently cropped to the lower half of the lung, and taken as the VOI. An average CT VOI was obtained by averaging the 10 time-percentages of the 4D-VOI. The rst (single) CT frame VOI (0%) was then subtracted from the average VOI (Fig. 5e,f). Finally, all voxels of the subtracted VOIs were averaged over space (Fig. 5g,h). The steps e through g were repeated for all 10 CT frames, and the 10 averaged values resulted in the motion curve of the diaphragm (Fig. 5h). The DM curve was scaled between 1 and 0 for comparison with the TM. The t%midv and the corresponding mid-ventilation tumor position derived from the DM were compared with the mean tumor position as inferred from analysis of the TM (resulting in error DM).

average with the human eye is a good rst estimate of the timeweighted mean position.

Validation of external respiration signal

Previous studies (26, 27) have reported that breathing measurements using an external device correlate well with movements of the internal structures, but phase shifts and time delays may exist between the two signals. Therefore, this phase shift and time delay was determined for the equipment used in our study. The verication of the thermocouple system, postprocessing of the thermocouple signal, and correlation to the DM is described in the Appendix. The postprocessed thermocouple signal, which is in phase with the DM, was used during 4D-CT reconstructions.

Determination of motion
Determination of TM by tumor tracking. The TM in the 4D-CT scan was determined using an image registration procedure (12). A volume-of-interest (VOI) was dened around the tumor in a reference CT frame (0%) using a manually drawn mask, encompassing the tumor. The VOI was subsequently registered to the scans of the other time-percentages on the basis of the gray value using the correlation ratio (25) of all voxels in the VOI. This procedure was repeated three times for three different reference CT frames (0%, 30%, and 70%). From each TM curve, the mean tumor position was subtracted to obtain the relative motion curves. The three curves were averaged to reduce spurious results due to random drawing or matching errors, resulting in one (relative) motion curve that was used for further analysis (Fig. 4). The TM curves were considered to be the reference. Automated determination of diaphragm motion. Automated determination of the tumor trajectory can be complex, especially if segmentation of the tumor is not available. Moreover, if the tumor is not solitary but is attached to the mediastinum or thoracic wall, tumor segmentation becomes difcult. We therefore investigated whether the diaphragm motion (DM) derived from the image information in the 4D-CT scan could be used as a surrogate for CC TM. This method is fully automated after the tumor has been located in the right or left lung. First, the rst frame (0% time percentage) from the 4D scan is selected. This scan is projected on an axial plane (Fig. 5a) and subsequently projected on the LR axis obtaining a one-dimensional prole of the pixel intensities (Fig. 5b). The position of the maximum is used to distinguish

RESULTS Respiration cycle length and amplitude Fifteen patients were included in this study (six upper lobe, one middle lobe, and eight lower lobe tumors). The average breathing cycle length was 4.4 s for all patients and over all cycles during scanning (Table 1). However, the cycle length varied between 1.8 s and 10.1 s. Seven patients had a mean cycle length 1 standard deviation larger than 5 s, and 10 of 15 patients had at least one cycle that was longer than 5 s. Ten patients underwent scanning with the 0.5 s gantry rotation protocol and ve with the 1.0 s gantry rotation protocol. Three patients were breathing with a shorter cycle before the scanning than during the scanning; therefore, the choice of the scanning parameters was not optimal. For 2 patients, a too-short gantry-rotation time was accidentally chosen (0.5 instead of 1.0 s). Although this

Translations (mm)

10 5 0 -5 -10

Maximum inspiration (Left; Caudal; Anterior)

Maximum expiration (Right; Cranio; Posterior)

25

50

75

100

Time-percentage (%)

Fig. 4. Example of tumor motion determined by automated gray-value matching algorithms. Solid, dashed, and dotted line denotes cranio-caudal, leftright, and anterior-posterior movements, respectively.

Clinical implementation of 4D-CT scans

J. W. H. WOLTHAUS et al.

1565

Fig. 5. Schematic overview of determination of diaphragm motion. First frame (0% time-percentage) from fourdimensional (4D) scan was picked. This scan was projected to axial plane (a) and subsequently projected to leftright axis obtaining one-dimensional prole of pixel intensities (b). Position of maximum (max.) used to distinguish left from right lung (c). Because patient shown had tumor in left lung, we considered left part of 4D-CT scan (d). This selection was cropped to lower half of lung region. 4D cropped image was averaged in time (e). Then, a single CT frame was subtracted from the average image (f). Finally, all voxels in subtracted images were averaged over space (g). The steps e through g were repeated for all 10 CT frames, resulting in (h) motion curve.

affected the scan quality, it did not seriously hamper the quantitative measurements (e.g., artifacts were not present in all 4D-CT frames and the resulting motion curves were smooth). CC amplitudes (Table 1) of the TM, as inferred from analysis of TM (see the section Determination of TM by Tumor Tracking above) ranged from 2 to 26 mm (mean 12.3 7.1 mm).

Thermocouple time delay The mean thermocouple time delay (see Appendix), measured over approximately 6 80 breathing cycles, was 0.40 0.18 s. This delay was equipment dependent and must be determined for the equipment used. In addition, it appeared that a very weak correlation was present between the cycle length and time delay (correlation coefcient 0.22). Mid-ventilation Figure 6 shows a typical example of the DM curve with the corresponding CC TM curve. The DM correlated highly with the TM. The average t%midv according to the CC TM (reference standard) was 17.4% 5.5% for exhalation and 75.0% 6.5% for inhalation (Table 2). The difference in t%midv between exhalation and inhalation did not appear to correspond to the scheme presented in Fig. 2c, which was because of an unexpected phase shift (see the section MidVentilation Determination under Discussion). To make

Table 1. Patient breathing characteristics (n

15)

TM Amplitude (mm) Mean cycle length (s) Mean SD 4.4 1.2 0.7 LR 2.2 1.2 CC 12.3 7.1 AP 3.6 2.2

Abbreviations: AP anterior-posterior; LR leftright; CC cranio-caudal; TM tumor motion. See 4D-CT Scanning and Determination of TM by Tumor Tracking under Methods and Materials for details.

1566

I. J. Radiation Oncology

Biology Physics

Volume 65, Number 5, 2006

Table 3. Average mid-ventilation time-percentages (t%midv) based on automated diaphragm motion (DM) method
DM

(mm) Inhale

t%midv (%) Exhale Mean SD 19.9 4.1 Inhale 76.7 5.3 LR 0.1 0.4

Exhale CC 0.2 0.8 AP 0.8 0.9 LR 0.0 0.4

CC 0.6 1.5

AP 1.1 1.1

Abbreviations as in Table 1. Geometric error is denoted by TM between determined midventilation position and real mean position of tumor during breathing for diaphragm motion method.

Fig. 6. Example of high correlation between cranio-caudal tumor motion (solid curve) and diaphragm motion computed by image gradient of four-dimensional scan (dashed curve). Both motion curves were scaled between maximal inhalation (1) and maximal exhalation (0). Note, time scale runs from 0% to 200% to emphasize periodicity.

time-percentage difference between DM and TM did not result in a smaller geometric error with respect to the mean tumor position ( DM). DISCUSSION The results of this work have shown that the mid-ventilation concept is a useful solution to incorporate motion information from 4D-CT scans in the treatment planning process. The determination of the t%midv using TM and DM is accurate; therefore, the appropriate planning CT scan can be selected from the 4D data set. The use of a single CT scan (with the tumor in the mean position) allows for margin reduction, resulting in a smaller irradiated volume (see the section Potential of Reduced Margins Because of Improved Imaging). Thermocouple time delay The comparison of the thermocouple signal vs. DM showed that the delay time of the thermocouple is stable and equipment dependent (see Appendix). If this time delay is not corrected, a time-percentage difference t% will exist between the rst frame of the 4D-CT scan (0%) and the actual maximal inhalation. t% can be estimated (after reconstruction of the 4D-CT scan) using DM determination, as described in our report. However, correction for this time delay before CT reconstruction is important, because the wrong synchronization of the thermocouple signal to the raw CT data will result in incorrect sorting of data into respiratory phases. This can be claried in a worst-case scenario example: when a patient would be scanned during two subsequent breathing cycles, with a cycle length of 3 and 6 s, respectively, and if a thermocouple time delay of 3 s exists that has not been corrected for, the rst half of the CT data of the second breathing cycle would be sorted over a full cycle. This type of error in sorting can result in serious imaging artifacts (e.g., when using external thorax movement respiration signal, see the section Phase Shifts due to Imperfections of 4D-CT below). However, the delay time of our thermocouple equipment (0.4 s) and the standard deviation of the cycle length were small (Table 1) compared with the breathing cycle length, resulting in negligibly small

the geometric results of the MidV-CT concept comparable with other geometric error contributions in radiotherapy, h and DM were calculated as the mean (group mean) and the standard deviation ( ) over the patient group. The geometric error, h, with respect to the mean position (Table 2) was, for exhalation, 0.2 0.3 mm in the LR direction and 0.7 0.8 mm in the AP direction. For inhalation, it was 0.0 0.4 mm in the LR direction and 1.1 1.1 mm in the AP direction. By denition, h in the CC direction was 0 because of the one-dimensional CC determination of t%midv. Using the DM method, the average t%midv for exhalation was 19.9% 4.1% and for inhalation was 76.7% 5.3% (Table 3). The difference between the DM and TM methods was signicant (exhalation and inhalation p 0.01, pairwise t test), but very small relative to the bin size of 10% (exhalation 2.5% 3.5%). The maximal geometric error, DM , with respect to the mean position (Table 3) in the CC direction, was 0.2 0.8 mm for exhalation and 0.6 1.5 mm for inhalation. The corresponding values in the AP and LR directions for exhalation and inhalation were smaller than 1.1 1.1 mm, not signicantly different from the reference of TM (p 0.1). Correcting for the systematic
Table 2. Average mid-ventilation time-percentages (t%midv) based on craniocaudal tumor motion and corresponding errors in three directions due to hysteresis ( h)
h

(mm) Inhale

t%midv (%) Exhale Mean SD 17.4 5.5 Inhale 75.0 6.5 LR 0.2 0.3

Exhale CC 0.0 0.0 AP 0.7 0.8 LR 0.0 0.4

CC 0.0 0.0

AP 1.1 1.1

Abbreviations as in Table 1.

Clinical implementation of 4D-CT scans

J. W. H. WOLTHAUS et al.

1567

Table 4. Mean error, calculated for patient group of present study, when scanning mobile tumors with conventional free-breathing 3D method
conv

(mm) AP 0 1.5

LR Mean SD 0 0.8

CC 0 3.9

Abbreviations: 3D three-dimensional; other abbreviations as in Table 1. Error calculated in three directions; group mean was 0 by denition.

artifacts in the 4D image reconstruction, even if the delay was not corrected. Mid-ventilation determination Phase shifts due to imperfections in 4D-CT. Because the 4D-CT data were sorted between the two maximal inhalation peaks after correcting for the thermocouple time delay as described in the section Scan Reconstruction, the rst CT frame (0%) of the 4D set was expected to represent the maximal inhalation phase. However, in general, a small time-percentage difference ( t%) was still found between the rst frame of the 4D-CT scan (0%) and the actual maximal inhalation phase. The time-percentage difference ( t%) was 3.3% 5.1% using TM and 1.0% 4.1% using the DM method. A possible cause for t% could be the quantization or binning error, which was 5% for the low number of bins (10) used during reconstruction. A second possible cause could be the existence of a (small) phase variation between internal structures and the external respiration system (see section Thermocouple Time Delay). Finally, the use of a one-dimensional (CC) simplication could be a third cause of the determined phase shift, because it does not have to represent the maximal inhale position in three dimensions. Rietzel et al. (16) reported a signicant phase difference between external thorax movements (from the Varian RPM respiratory measurement system) and internal structures. If this phase difference is stable during the entire 4D scan acquisition, the phase difference can be found retrospectively by methods we have described. However, when patients switch from lower body (abdomen) to upper body (chest) respiration during scanning, a signicant phase change may occur, which can result in severe artifacts on the 4D images. Mid-ventilation time-percentage. The t%midv derived from the TM and DM curves appeared to be signicantly different (Tables 2 and 3). However, relative to the bin size (10%), the difference was small; therefore, the automated DM method can be used as a surrogate for the MidV estimation with the TM method. Moreover, after correcting the t%midv for patients individual time-percentage difference t% (thus, setting the maximal inhalation to 0%, see previous section), the t%midv was no longer very different. The t%midv was 20.7% 2.6% and 20.7% 2.2% and 78.7% 2.9% and 78.8% 2.9% in exhalation and inha-

lation for TM and the DM method, respectively. The standard deviation in t%midv was reduced, supporting the use of the DM method as a surrogate for TM. These results of the t%midv were in agreement with the asymmetric properties of respiration (t%midv not 25% or 75%), as described by Seppenwoolde et al. (23). It appeared that the TM and DM curves were in phase except for patients with tumor attached to the mediastinum or chest wall. In those cases, the curves were also slightly different in shape, resulting in a different t%midv for TM and DM. TM will be used to select the MidV-CT scan. Uncertainties of mid-ventilation with respect to mean tumor position. From the standard deviation in h and DM (Tables 2 and 3), it can be concluded that a MidV reconstruction in exhalation is more accurate than that with inhalation. Because the exhalation phase is longer than the inhalation phase, the excursion per time-percentage is greater during inhalation than during exhalation. A small uncertainty in t%midv will thus result in a larger geometric error at inhalation than at exhalation. For clinical practice, a cutoff value in the TM amplitude, below which the automated DM method can be safely used, might be benecial. First, for all patients, the mean and standard deviation of the geometric error relative to the amplitude was computed in the three directions for all patients. This relative error ( rel) was 0.14 0.04 mm/mm, 0.07 0.04 mm/mm, and 0.31 0.10 mm/mm for the LR, CC, and AP directions, respectively. Limiting the mean geometrical error 1 standard deviation to 2 mm for all directions can be described as a constraint: max
rel,LR

ALR ;

rel,CC

ACC ;

rel, AP

AAP

2 mm

Using the statistics of the amplitude for this patient group (Table 1), the proportions of the mean amplitudes in the LR and AP direction relative to the CC amplitude were ALR 0.18ACC and AAP 0.29ACC

Applying these amplitude relations in the constraint equation showed that the use of the automated DM method is limited to a maximal CC tumor excursion of 2 cm. Comparison of MidV-CT with conventional CT. To evaluate the methods for the construction of a treatment planning CT scan, we compared the results with those from conventional CT scanning (no respiration-correlated imaging). Because we had determined the TM for all directions and all patients, we could easily estimate the errors in conventional imaging ( conv), which is the standard deviation of the motion over the respiratory cycle (28). The mean is 0 by denition. The standard deviation for all patients was 0.8 mm for LR, 3.9 mm for CC, and 1.5 mm for AP (Table 4). This is in agreement with the conclusions of van Herk (28), that the geometric error with respect to the mean tumor position ( conv) was one-third of the peak-to-peak amplitude. Comparing the DM method to conventional scanning,

1568

I. J. Radiation Oncology

Biology Physics

Volume 65, Number 5, 2006

the geometric variation DM was four times smaller in the CC direction than conv (Table 3). Dose verication of the new MidV concept is beyond the scope of this report; however, we refer to Cho et al. (29) who showed that, with constant error margins, small geometric variations have a negligible inuence on the dose distribution. The application of the results of Cho et al. (29) to the relatively small geometric differences between conventional and MidV-CT scans showed that MidV-CT scans can safely replace conventional CT scans. The improvement gained by the use of MidV-CT will primarily be a reduction in the error margins. A clinical protocol based on the presented results concerning 4D scanning and mid-ventilation CT reconstruction is given in Fig. 1. Reproducibility of respiratory patterns. Reproducibility in respiratory patterns during 4D imaging and treatment delivery is an important issue. We have begun a study of the possible changes in breathing patterns and mean tumor position when using an oxygen mask. Additional 3D scans are made before treatment with a cone-beam system mounted on a linear accelerator (30). These scans can be reconstructed into 4D scans (31) and mean tumor position and amplitude determined. Although the preliminary results have shown no signicant differences in breathing patterns with or without a mask, the nal conclusions cannot yet be made. Inuence of breathing cycle length on 4D-CT reconstructions The small standard deviation in the average cycle length (Table 1) suggests that the patients were breathing regularly during scanning, resulting in stable and practically artifactfree 4D-CT reconstructions. This makes a choice for one of the two protocols (0.5 s or 1.0 s gantry rotation) justied. However, a lower pitch or more choices of gantry rotation speed would be desirable to overcome the remaining artifacts. Interpolation artifacts in the 4D image appeared at locations where undersampling occurred. If the number of these artifacts were small, tumor match and DM measurements would not be seriously hampered. In some cases, the patient was breathing faster during setup than during the actual 4D scanning. Consequently, the gantry rotation was too fast compared to the patients cycle length, causing multiple imaging artifacts and making delineation of structures or tumor more difcult. However, quantitative measurements were not noticeably inuenced. For patients with irregular breathing, one solution was to wait a couple of minutes until the respiration stabilized, before measuring the cycle length, which was included in the 4D scanning protocol. In addition, audiovisual breathing coaching can stabilize the breathing signal amplitude and period (32), which will improve the image quality of the MidV-CT scans. However, coaching is often not applicable to this patient group because of physical limitations. Moreover, the inuence of coaching on the baseline (mean respiration level) variation is still unknown. Amplitude-based CT data binning (33) during regular breathing would not result in

better image quality and has the drawback that time information is lost. Time information is necessary to calculate the time-weighted mean position. Furthermore, with irregular breathing, some amplitude bins can be sparsely lled, resulting in serious artifacts (gaps). A principle disadvantage of our proposed method is that the tumor has the greatest speed at mid-ventilation and can result in more residual imaging artifacts than at maximal inhalation or maximal exhalation. However, according to Rietzel et al. (27), these artifacts are smaller than the artifacts due to irregular breathing. Moreover, compared with conventional scanning, these speed-induced artifacts are negligible. Potential of reduced margins because of improved imaging Treatment margins (from GTV to PTV) can be calculated using the margin recipe of van Herk et al. (22, 34, 35): margin 2.5 0.7 , where and denote the standard deviations of the systematic and random errors, respectively. These errors consist of contributions from respiration (periodic motion and mean tumor position) and setup errors (errors due to cardiac motion were not considered). Delineation uncertainties also contribute to additional margins (36) and are expected to become smaller using 4D treatment planning because of improved visualization of the tumor shape. However, we did not consider detailed knowledge on delineation uncertainty using free-breathing scans (37, 38). Margins due to systematic and random errors were estimated for a patient with a spherical tumor of 20 mm and a 20 mm CC tumor movement in diameter in two cases (Table 5): without using 4D imaging (columns 2 and 3) and with using 4D imaging (column 4 and 5). The displaced tumor in the conventional planning CT (random snapshot of the tumor) with respect to the mean position of the moving

Table 5. Estimated magnitude of existing uncertainties in craniocaudal direction for two cases: conventional 3D-CT or 4D-CT scan for treatment planning Conventional 3D-CT (mm) Respiration (ACC 20 mm) Periodic motion Baseline variation Setup error Total Margin (CTV to PTV) (mm) 4D-RCCT (mm) (mm)

6.7 2 1.5 7.2 23.2

6.7 2 3.1 7.6

0 2 1.5 2.5 11.6

6.7 2 3.1 7.6

Abbreviations: 3D three-dimensional; 4D four-dimensional; ACC amplitude of tumor motion in cranio-caudal direction; CTV clinical target volume; PTV planning target volume. Required margin from CTV to PTV was calculated according to van Herk et al.: margin 2.5 0.7 , where and denote standard deviations of distribution of systematic and random error, respectively.

Clinical implementation of 4D-CT scans

J. W. H. WOLTHAUS et al.

1569

tumor is referred to as the systematic contribution of the periodic motion. The probability distribution of the tumor in a certain position during treatment, which has a standard deviation of one-third of the peak-to-peak amplitude (28), is referred to as the random contribution of the periodic motion. The systematic and random setup errors were set at 1.5 mm and 3.1 mm, respectively, in the conventional and 4D MidV situation using portal images (39). Using 4D imaging will theoretically reduce the systematic contribution due to breathing motion to 0. The random error distribution due to respiration will not change because respiration is still present during treatment. However, the margin for random errors might possibly be reduced, such as was shown in simulations due to the shallow penumbra in lung tissue (21, 22). The additional margins needed to compensate for the breathing motion are small if the irradiated volume is centered at the average position of the tumor using 4D imaging techniques. Variations in the average position of the tumor (baseline) give a contribution that still remains in 4D treatment planning. This contribution can only be reduced when using an off-line 4D verication protocol (using multiple 4D-CT scans), reducing the systematic average position variation, or an on-line 4D verication protocol (using multiple 4D-CT scans on the treatment machine [40, 41]), reducing the systematic and random average position variation. Using multiple 4D conebeam CT scans from 10 patients (42), this average position error was estimated to be 2 mm for both systematic and random errors. For the patient data in Table 5 with a CC TM of 20 mm, the use of 4D imaging would result in a reduction of the required margin in the CC direction with a factor of

2. With an amplitude in the LR direction of 4 mm and 6 mm in the AP direction (see the section Uncertainties of MidVentilation with Respect to Mean Tumor Position), the margins in these directions will be 10 mm and 11 mm for conventional free-breathing CT scans and 9 mm and 9 mm for MidV-CT scans, respectively. In this example (spherical tumor of 20 mm diameter), this margin reduction decreases the PTV volume from 14.4 cm3 to 8.3 cm3 (43% reduction). In addition, using the mean motion amplitude values of Table 1, the average PTV reductions for tumors with diameters between 10 and 80 mm would be between 33% and 12%. Finally, for the patient group used in this study, the composite target volume methods described in the Introduction (slow-CT and maximal intensity projection) would result in an increased irradiated volume of 11% 3% (setup error and baseline variation as above). Our proposed MidV-CT method would result in a signicant reduction of 12% 11% irradiated volume. CONCLUSION We have developed a new concept for using 4D-CT scans to incorporate patient-specic motion information in radiotherapy for lung cancer. On the basis of the tumor motion, a single 3D-CT scan is chosen from the 4D-CT data set, which represents all lung structures in the time-average position. An alternative automated method was also developed to select the single 3D-CT scan using the diaphragm motion. Using mid-ventilation CT scans for treatment planning instead of the conventional free-breathing CT scans, margin reduction is possible, which can reduce the treatment volume up to 50%.

REFERENCES
1. Belderbos JS, De Jaeger K, Heemsbergen WD, et al. First results of a phase I/II dose escalation trial in non-small cell lung cancer using three-dimensional conformal radiotherapy. Radiother Oncol 2003;66:119 126. 2. Kong FM, Ten Haken R, Eisbruch A, et al. Non-small cell lung cancer therapy-related pulmonary toxicity: An update on radiation pneumonitis and brosis. Semin Oncol 2005;32:S42S54. 3. Belderbos J, Heemsbergen W, Hoogeman M, et al. Acute esophageal toxicity in non-small cell lung cancer patients after high dose conformal radiotherapy. Radiother Oncol 2005;75:157164. 4. Schwarz M, Alber M, Lebesque JV, et al. Dose heterogeneity in the target volume and intensity-modulated radiotherapy to escalate the dose in the treatment of nonsmall-cell lung cancer. Int J Radiat Oncol Biol Phys 2005;62:561570. 5. International Commission on Radiation Units and Measurements. Prescribing, recording, and reporting photon beam therapy: ICRU report No. 50. Bethesda: ICRU; 1993. 6. Balter JM, Ten Haken RK, Lawrence TS, et al. Uncertainties in CT-based radiation therapy treatment planning associated with patient breathing. Int J Radiat Oncol Biol Phys 1996;36:167174. 7. Chen GT, Kung JH, and Beaudette KP. Artifacts in computed tomography scanning of moving objects. Semin Radiat Oncol 2004;14:19 26. 8. Vedam SS, Keall PJ, Kini VR, et al. Acquiring a fourdimensional computed tomography dataset using an external respiratory signal. Phys Med Biol 2003;48:45 62. 9. Ford EC, Mageras GS, Yorke E, et al. Respiration-correlated spiral CT: A method of measuring respiratory-induced anatomic motion for radiation treatment planning. Med Phys 2003;30:88 97. 10. Keall PJ, Starkschall G, Shukla H, et al. Acquiring 4D thoracic CT scans using a multislice helical method. Phys Med Biol 2004;49:20532067. 11. Pan T. Comparison of helical and cine acquisitions for 4D-CT imaging with multislice CT. Med Phys 2005;32:627 634. 12. Wolthaus JW, van Herk M, Muller SH, et al. Fusion of respiration-correlated PET and CT scans: Correlated lung tumour motion in anatomical and functional scans. Phys Med Biol 2005;50:1569 1583. 13. Pan T, Lee TY, Rietzel E, et al. 4D-CT imaging of a volume inuenced by respiratory motion on multi-slice CT. Med Phys 2004;31:333340. 14. Srensen de Koste J, Lagerwaard FJ, de Boer HCJ, et al. Are multiple CT scans required for planning curative radiotherapy in lung tumors of the lower lobe? Int J Radiat Oncol Biol Phys 2003;55:1394 1399. 15. Allen AM, Siracuse KM, Hayman JA, et al. Evaluation of the inuence of breathing on the movement and modeling of lung tumors. Int J Radiat Oncol Biol Phys 2004;58:1251 1257. 16. Rietzel E, Chen GT, Choi NC, et al. Four-dimensional imagebased treatment planning: Target volume segmentation and

1570

I. J. Radiation Oncology

Biology Physics

Volume 65, Number 5, 2006

17.

18. 19. 20. 21. 22. 23.

24. 25.

26. 27. 28. 29.

dose calculation in the presence of respiratory motion. Int J Radiat Oncol Biol Phys 2005;61:15351550. Underberg RW, Lagerwaard FJ, Slotman BJ, et al. Use of maximum intensity projections (MIP) for target volume generation in 4DCT scans for lung cancer. Int J Radiat Oncol Biol Phys 2005;63:253260. 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;44:911919. Rosenzweig KE, Hanley J, Mah D, et al. The deep inspiration breath-hold technique in the treatment of inoperable nonsmallcell lung cancer. Int J Radiat Oncol Biol Phys 2000;48:81 87. Keall PJ, Joshi S, Vedam SS, et al. Four-dimensional radiotherapy planning for DMLC-based respiratory motion tracking. Med Phys 2005;32:942951. Engelsman M, Damen EM, De Jaeger K, et al. The effect of breathing and set-up errors on the cumulative dose to a lung tumor. Radiother Oncol 2001;60:95105. Witte M, van der Geer J, Schneider C, et al. The effects of target size and tissue density on the minimum margin required for random errors. Med Phys 2004;31:3068 3079. 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. van Herk M, De Jaeger K, de Munck J, et al. A delineation system for N modalitiesSoftware aspects [Abstract]. Presented at the ICCR 13th Heidelberg, Germany, 2000. p. 7375. Roche A, Malandain G, Pennec X, et al. The correlation ratio as a new similarity measure for multimodal image registration. Proceedings of MICCAI98. Lecture Notes in Computer Science 1998; 1496:11151124. Lu W, Parikh PJ, El Naqa IM, et al. Quantitation of the reconstruction quality of a four-dimensional computed tomography process for lung cancer patients. Med Phys 2005;32:890 901. Rietzel E, Pan T, Chen GT. Four-dimensional computed tomography: Image formation and clinical protocol. Med Phys 2005;32:874 889. van Herk M. Errors and margins in radiotherapy. Semin Radiat Oncol 2004;14:52 64. Cho BC, van Herk M, Mijnheer BJ, et al. The effect of set-up uncertainties, contour changes, and tissue inhomogeneities on target dose-volume histograms. Phys Med Biol 2002;29:2305 2318.

30. Jaffray DA, Siewerdsen JH, Wong JW, et al. Flat-panel conebeam computed tomography for image-guided radiation therapy. Int J Radiat Oncol Biol Phys 2002;53:13371349. 31. Sonke JJ, Zijp L, Remeijer P, et al. Respiratory correlated cone beam CT. Med Phys 2005;32:1176 1186. 32. Neicu T, Berbeco R, Wolfgang J, et al. Synchronized moving aperture radiation therapy (SMART): Improvement of breathing pattern reproducibility using respiratory coaching. Phys Med Biol 2006;51:617 636. 33. Fitzpatrick MJ, Starkschall G, Antolak JA, et al. Displacement-based binning of time-dependent computed tomography image data sets. Med Phys 2006;33:235246. 34. van Herk M, Remeijer P, Lebesque JV. Inclusion of geometric uncertainties in treatment plan evaluation. Int J Radiat Oncol Biol Phys 2002;52:14071422. 35. van Herk M, Witte M, van der Geer J, et al. Biologic and physical fractionation effects of random geometric errors. Int J Radiat Oncol Biol Phys 2003;57:1460 1471. 36. van de Steene J, Linthout N, de Mey J, et al. Denition of gross tumor volume in lung cancer: Inter-observer variability. Radiother Oncol 2002;62:37 49. 37. Steenbakkers RJ, Duppen JC, Fitton I, et al. Reduction of observer variation using matched CT-PET for lung cancer delineation: A three-dimensional analysis. Int J Radiat Oncol Biol Phys 2006;64:435 448. 38. Steenbakkers RJ, Duppen JC, Fitton I, et al. Observer variation in target volume delineation of lung cancer related to radiation oncologist-computer interaction: A Big Brother evaluation. Radiother Oncol 2005;77:182190. 39. Erridge SC, Seppenwoolde Y, Muller SH, et al. Portal imaging to assess set-up errors, tumor motion and tumor shrinkage during conformal radiotherapy of non-small cell lung cancer. Radiother Oncol 2003;66:75 85. 40. Sonke JJ, Zijp L, Remeijer P, van Herk M, et al. Respirationcorrelated cone-beam CT. Med Phys 2005;32:1176 1186. 41. Zijp L, Sonke JJ, van Herk M. Extraction of respiratory signal from sequential thorax cone-beam X-ray images. Proceedings of the XIVth ICCR, Seoul, Korea, 2004. p. 507509. 42. Sonke JJ, van Herk M, Belderbos J, et al. An off-line 4D cone beam CT based correction protocol for lung tumor motion [Abstract]. Int J Radiat Oncol Biol Phys 2005;63(Suppl. 1): S389 S390.

APPENDIX Thermocouple response time To verify the functioning of the thermocouple system and to determine the correlation with diaphragm motion, we measured the diaphragm movements using projection images from a cone-beam CT scanner mounted on a linear accelerator (30) concurrent with the thermocouple respiration signal system for 6 patients during 7 min. The respiration signal was acquired using an AD-converter simultaneously with the X-ray exposure pulses that were generated by the cone-beam imager. The exposure peaks provided a time stamp for each projection image. These time stamps of the acquired projection images were synchronized to the thermocouple signal. Subsequently, the diaphragm height was determined automatically in each projection image (31, 41). These measurements do not require a cone-beam CT system, but can be performed with conventional uoroscopy. Because the thermocouple (temperature) and diaphragm position (distance) have different units, both signals were scaled between 0 (maximal exhalation) and 1 (maximal inhalation) as follows. An upper and lower envelope was created by drawing one line intersecting all peaks in the signal and a second line intersecting all valleys, respectively. The original signals were scaled between their upper and lower envelopes. The advantage of this method is that it is easy to calculate, preserves the asymmetric shape of the curve (different length of exhalation and inhalation), and removes a possible global trend. In Fig. 7, an example of the

Clinical implementation of 4D-CT scans

J. W. H. WOLTHAUS et al.

1571

Fig. 7. Example of thermocouple respiration signal (dashed curve) and diaphragm motion (solid curve) determined from cone-beam computed tomography projection images. Both signals were scaled between maximal inhalation (1) and maximal exhalation (0). denotes time delay between two signals.

two signals was plotted (dashed curve represents thermocouple signal and solid curve represents diaphragm signal). A time delay occurs in the thermocouple signal with respect to the diaphragm (Fig. 7). The thermocouple time delay is due to the warming up and cooling down of the thermocouple, the presence of stationary air (end of expiration, end of inspiration), and the response time of the signal amplier equipment. The mean thermocouple time delay, measured for approximately 6 80 breathing cycles, was 0.40 0.18 s. This delay is equipment dependent and should be determined for the equipment used. Considering the quantization error in the acquisition of a cone-beam projection image (standard deviation of a uniform distribu12 0.371 12 0.107 s) tion acquisition time/ and additional noise, the standard deviation is quite acceptable. From these data, it appeared that a very weak correlation was present between the cycle length and time delay (correlation coefcient 0.22), suggesting that little phase delay exists and therefore the response time is mainly equipment dependent. This delay (determined with the conebeam CT system) will be corrected during 4D-CT reconstructions on a standard CT scanner.