Radiation induced lung damage

(1)

UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)

Seppenwoolde, Y.

Publication date

2002

Link to publication

Citation for published version (APA):

Seppenwoolde, Y. (2002). Radiation induced lung damage.

General rights

It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s)

and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open

content license (like Creative Commons).

Disclaimer/Complaints regulations

If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please

let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material

inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter

to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You

will be contacted as soon as possible.

(2)

Chapterr 7

Portall imaging to assess setup errors,

tumorr motion and tumor shrinkage

duringg conformal radiotherapy of

non-smalll cell lung cancer

Sarahh C. Erridge, Yvette Seppenwoolde, Sara H. Muller, Marcel van Herk,

Katrienn De Jaeger, José S A Bekferbos, Liesbeth J. Boersma, Joos V. Lebesque

(3)

(4)

Portall imaging to assess setup errors, tumor motion

andd tumor shrinkage during conformal radiotherapy of

non-smalll cell lung cancer

Inn 97 patients, electronic portal images (EPIs) were acquired to investigate patient setup, tumorr movement and shrinkage during 3D conformal radiotherapy for non-smafl oeK lung cancer.. For 25 selected patents, the orthogonal EPte (taken at random points in the breamingg cycle) throughout the six to seven week course of treatment were assessed to establishh the tumor position in each image using both an overlay and a delineation technique.. The range of movement in each direction was calculated. The position of the tumorr in the digitally reconstructed radiograph (DRR) was compared to the average positionn of the lesion in the EPte. tn addition, tumor shrinkage was assessed. The mean overalll setup errors after correction were 0 mm, 0.6 mm and 0.2 mm in the x (left-right), y (cranial-caudal)) and z (anterior-posterior) direction, respectively. After correction, the standardd deviations (SD) of systematicc errors were 1.4 mm. 1.5 mm and 1.3 mm and the SDD of random errors were 2.9 mm, 3.1 mm and 2.0 mm in the x-, y- and z-direction, respectively.. Without correction, 41% of patients had a setup error of more than 5 mm vectorr length, but with the setup correction protocol this percentage reduced to 1%. The meann amplitude of tumor motion was 7.3 mm (SD 2.7), 12.5 mm (SO 7.3) and 9.4 mm (SOO 5.2) in the x-, and z-direction, respectively. Tumor motion was greatest in the y-directionn and in particular for tower lobe turner. In 40% ofthe patients, the projected area off the tumor regressed by more than 20% during treatment in at least one projection. In 166 patients it was possible to define the position of the center of the tumor in the DRR. Theree was a mean difference of 6 mm vector length between the tumor position in the DRRR and the average position in the portal images. The application of the correction protocoll resulted in a significant improvement in Sue setup accuracy. There was wide variationn in the observed tumor motion with more movement of tower lobe lesions. Tumor shrinkagee was observed which could potentially be used to reduce margins in the final weekss of treatment. The position of the tumor on the planning CT scan did not always coincidee with the average position as measured during treatment.

Introduction n

Non-smalll cell lung cancer (NSCLC) is one of the most common malignant diseases. For a largee number of patients surgery is inappropriate either because of locally advanced disease orr because of comorbidity. For these patients the only potentially curative treatment modality iss radical radiotherapy. However, for conventional doses the survival remains poor and many patientss die with local failure (Perez 1986, 1987, Arriagada 1991). Several studies have suggestedd that a dose response exists (Cox 1990, Byhardt 1995, Schaafsma 1998), with higherr radiation doses resulting in a higher probability of local control and hence prolonged survival.. The development of 3D conformal radiotherapy techniques has allowed for dose escalationn with acceptable levels of morbidity (Armstrong 1997, BeWerbos 2000, Hayman 2001).. The reduction of treatment volumes facilitates dose-escalation, but implies a potential riskk of a geographical miss.

Thee ICRU Report 50 and its recently published supplement 62 define the volumes to be placed aroundaround the gross tumor volume (GTV). The clinical target volume (CTV) includes the GTV and aa margin for subclinical malignant disease. Around this is placed an additional volume to

(5)

createe the planning target volume (PTV). The margin between the CTV and the PTV has in

generall two components; a small component to account for random errors and a large

componentt to account for systematic deviations (van Herk 2000).

Thee introduction of electronic portal imaging devices has enabled acquisition of frequent and

clearr images (EPIs) of the patient in the treatment position. The bony anatomy, visible on these

imagess can be compared to that at the time of simulation to correct for errors in setup. On n

thesee EPIs also the tumor is sometimes visible. Therefore, these images were used in this

studyy to assess i) the effect on setup accuracy of an off-line correction protocol based on a

shrinkingg action level, H) tumor movement and iii) tumor regression during a six to seven week

coursee of radical radiotherapy, in addition, the difference between the average tumor position

duringg treatment compared to the tumor position in the planning CT scan was evaluated.

Materialss and methods

Inn August 1996, 3D conformal radiotherapy (3D-CRT) was introduced at the Netherlands

Cancerr Institute (NKI) for the radical treatment of norvsmall cell lung cancer. By the end of

1999,, 122 patients were treated in this way. The patients were positioned using a forearm

supportt and a knee-roll and instructed to breathe gently.

setupp verification

Duringg radiotherapy, in a pre-defined schedule, two or three orthogonal sets (anterior-posterior

(AP)) and ipsi-Eateral) of electronic portal images (Variant PortaJVision (t liquid ionisation

chamber)) per fraction were acquired to identify and correct patient setup inaccuracies. The

positionn of the bony structures in these images was determined by calculating the average of

thee two or three consecutive images to reduce breaming motion of tile organs and improving

thee signal-to-noise ratio. This average image was then compared with the digitally

reconstructedd radiograph (DRR) derived from the planning CT scan in an off-line setup

verificationn protocol (Bel 1993). The decision rule for setup corrections was based upon a

shrinkingg action leve! (OWN, with a = initial action level and a fixed number of initial

measurementss (N

max

)). In the protocol the initial action level was 9 mm vector length. N ^ was

22 at the introduction of the protocol. After the first 25 patients this number was increased to 3

becausee the random errors appeared to be somewhat larger than expected. Using the revised

values,, in the first fraction an error of 9 mm vector length was acceptable, in the second

fractionn 6.3 mm and in the third 5.2 mm. if the setup positions within the first three days were

withinn these limits, the images were repeated weekly. If a correction was performed, a further

threee sets of daily images were acquired. The effect of the off-line correction protocol on the

setupp accuracy was analysed for the 97 patients treated with the revised N ^ = 3.

Breathingg motion

Thee portal images were reviewed for those patients in whom the lesion was clearly visible on

thee majority of the portal images. Twenty-five patiënte (13 upper lobe and 12 lower/middle lobe

tumors)) with visible tumor were selected for further evaluation. The characteristics of these

patientss are shown in Table 1. In some patients the tumor was less easily visible on the lateral

image.. In five of the 25 patients it was not possible to evaluate the tumor in this projection, so

forr these patients tumor motion was calculated in the x- and y-direction only. A total of 574

anterior-posteriorr (AP) and 407 lateral images were analyzed, with an average of 23 AP

imagess and 15 lateral images per patient Because all portal images were matched on bony

anatomy,, the tumor position was measured relative to tile bony anatomy of each patient.

(6)

PortalPortal imaging to assess set-up errors and tumor motion

Tablee 1 . Patient characteristics.

Numberr of patients 25 Radiation n

Male e Female e

Meann age (range)

177 (68%) 88 (32%) 733 (48 - 87) years

Prescribedd dose 54-81 Gy

Fractionss 27-36 Overalll treatment time 40-50 days

Pathology y Stage e Squamouss cell 9 (36%) Adenocarcinomaa 9 (36%) Largee cell 6 (24%) Unknownn 1 (4%) II I lila a 1Mb b 100 (40%) 44 (16%) 66 (24%) 55 (20%)

Inn each treatment session the first portal image was taken at a random point within the breathingg cycle, with the second and third images made approximately one and three seconds thereafter,, resulting in 'snap-shots' of the tumor position during respiration (Figure 1).

T i m ee (s) Breathingg curve Exhale e

00 1 3

D R RR Fraction 1

00 1 3

Fractionn n fyfy = average X

FigureFigure 1. The first portal image of each day is taken at a random point during the breathing cycle. The second and third images areare taken one and three seconds after the first. The measured range in tumor position can vary each day, depending on the point inin the breathing cycle where the first image is taken. To obtain the measured tumor motion, the encompassing range of all days (Ax„J(Ax„J is determined. However, due to the limited amount of measurements, this encompassing range may underestimate the true rangerange (Ax,) of tumor motion. For the same reason, the true average tumor position can differ from the measured average tumor position.position. Because the tumor position on the DRR is measured from a free breathing CTscan, this position can also differ from thethe measured (and true) tumor position during treatment. Differences in breathing level will increase both Ax, and Axm and will

influenceinfluence the average tumor position.

Thee measured range in tumor position can vary each day, depending on the point in the breathingg cycle where the first image is acquired. To obtain the overall range of tumor motion

(7)

thee encompassing range of all days (Axm) is determined. However, due to a limited number of

sampledd measurements, this encompassing range will underestimate the true range (Axt) of tumorr motion. Because the tumor position in the DRR is measured from a free breathing CT scan,, this position can also differ from the average tumor position during treatment. Differencess in breathing level will increase Axm and this also influences the position of the

measuredd average tumor position.

Thee AP and lateral images are taken sequentially. To determine the position of the tumor two methodss were used:

Methodd 1

Inn this method, the color scale of the two images to be compared were set in complementary colorss (green and fuchsia) and overlaid, so that a perfect match would result on the image in gray-valuess (van Herk 1993). A mismatch is visible by green and fuchsia edges around the structures.. The tumor projection was matched and the translation between the images relative too the bony anatomy represented the tumor motion. The advantage of this method is that the tumorr shape can be matched, so that variations in the opacity of adjacent structures can be ignored.. However, when tumor shrinkage occurs during treatment inter-fraction comparison becomess difficult.

Methodd 2

Usingg a delineation tool, the visible gross tumor projection was outlined on all portal images (Figuree 2). The center of gravity of this two-dimensional tumor image relative to the bony anatomy,, and the area of the tumor projection were calculated. The advantage of this method wass its ability to assess tumor regression. However, overlapping structures influenced the targett delineation and hence the measurement of tumor position. In addition, alterations in the exposuree of the images can lead to discrepancies in the delineation.

FigureFigure 2. Tumor motion during treatment as measured with method 2. The tumor delineated with the solid white line on thethe first portal image, moved to the tumor delineated with a dashed black line on the second portal image of the first fraction.fraction. Note the corresponding diaphragm motion (white arrows). The dark contours (black arrow) are caused by an isocenterisocenter maricer that is only used during the first fractbn.

(8)

Thee correlation between the two methods was calculated to ensure that both methods gave reliablee results. The tumor motion was recorded using both methods and averaged to reduce thee influence of the disadvantages of each method. Due to concerns regarding the accuracy off the interpretation of the lateral images, the tumor motion in the y-direction was recorded only fromm the AP images. Due to poor visibility of the tumor it was not possible to match or delineate somee of the tumors properly. These unreliable measurements were excluded from this analysis.. The results were compared with the fluoroscopic assessment of tumor movement performedd at the time of simulation. This movement was graded as follows: grade 1 is no movement,, grade 2 is movement less than 10 mm and grade 3 is movement larger than 10 mm,, in any of the three directions.

Tumorr shrinkage

Methodd 2 gives information on the size of the 'Beams-Eye-View' projection of the tumor onto thee coronal and sagittal planes. The mean of the measured area of the projected surface duringg the first week of treatment (when shrinkage is unlikely to occur), was calculated to give thee base-line tumor area. The measurements in the other fractions were compared to the baselinee tumor area to determine the time course and magnitude of tumor shrinkage during treatment. .

Comparisonn of planning CT position to treatment position

Inn order to assess whether or not the position of the GTV delineated on the planning CT scan wass representative for the tumor position during treatment, the difference between the center off gravity of the tumor seen on the DRRs and the average tumor position on the portal images wass calculated. For 16 patients the tumor was visible on the DRR. For some patients, in order too visualize the tumor, the bony structures had to be eliminated from the CT by an in-house developedd clipping tool scan before calculating the DRR. A region around the tumor (without bone)) was selected to calculate a 'clipped' DRR (Figure 3). When the tumor could be delineatedd on the DRR, the center of gravity was defined as the (0,0) position of the tumor.

FigureFigure 3. An example of the construction of a clipped DRR (right lower image): instead of using all the CT information asas is done in a normal DRR (right upper image), only the CT information between the two white lines is used. The mediastinummediastinum and vertebrae that obscure the visibility in the normal DRR are not visible in the clipped DRR.

(9)

Results s

setupp verification

Thee mean overall setup errors and the standard deviations of the systematic and the random errorss of the patients' setup were calculated. The use of the correction protocol resulted in a considerablee reduction of systematic setup errors (Table 2). The impact of the setup correction protocoll was further evaluated by plotting the length of the setup deviation vector, with and withoutt correction, in a cumulative histogram (Figure 4). Of the 97 patients, 4 1 % had a setup errorr with a vector length larger than 5 mm without corrections. Using the correction protocol, thiss percentage was reduced to 1%.

Tablee 2. Patient setup errors (n=97), x = left-right, y = cranial-caudal, z = anterior-posterior.

Direction n X X y y z z Meann overall errorr (u.) (mm) -0.2 2 1.6 6 0.5 5 Standardd Deviation of systematicc error ( I )) (mm) 3.3 3 4.4 4 2.2 2 Standardd deviation of randomm error (a)) (mm) 2.4 4 2.6 6 1.8 8 A.. without setup corrections.

X X y y z z 0.0 0 0.6 6 0.2 2 1.4 4 1.5 5 1.3 3 2.9 9 3.1 1 2.0 0

B.. with setup corrections.

FigureFigure 4. The cumulative histogram of the distribution of the vector length of the systematic setup errorwith and without corrections.corrections. Of the 97 patients, 41% had a setup error with a vector length of more than 5 mm without corrections. Using thethe correction protocol, this was reduced to 1%.

(10)

Breathingg motion

Respiration-inducedd tumor motion was determined in twenty-five patients (Table 3). The agreementt between the two methods was good (correlation coefficient r2 = 0.8). Figure 5 demonstratess the tumor positions measured over eight fractions. The effects illustrated in Figuree 1 are clearly visible: the number of measurements is limited each day and the range of motionn differs from day-to-day, due to random sampling of trie breathing phase or due to changess in breathing level. The tumor movement averaged over all patients was 7.3 mm (SD 2.7,, range 3.2 to 12.8 mm) in the x (left-right) direction, 12.5 mm (SD 7.3, range 4.7 to 33.8 mm)) in the y (caudal-cranial) direction and 9.4 mm (SD 5.2, range 4.8 to 21.4 mm) in the z (posterior-anterior)) direction. No systematic shift in tumor position was observed during the coursee of the treatment in either direction.

Tablee 3. Tumor movement

Patient t 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 0 11 1 12 2 13 3 14 4 15 5 16 6 17 7 18 8 19 9 20 0 21 1 22 2 23 3 24 4 25 5 Lobe e LLL L LLL L LLL L LLL L LLL L RLL L RLL L RLL L RLL L RML L RML L RML L RMURUL L LUL L LUL L LUL L LUL L LUL L LUL L RUL L RUL L RUL L RUL L RUL L RUL L Location n central l central l central l peripheral l peripheral l central l peripheral l peripheral l peripheral l central l central l central l peripheral l central l central l central l peripheral l peripheral l peripheral l central l central l central l peripheral l peripheral l peripheral l Invasion n no o no o no o no o no o mediastinum m no o no o pleura a no o mediastinum m no o no o no o no o no o no o No o chestt wall chestt wall mediastinum m mediastinum m no o no o no o Tumorr movement. encompassingg range (mm) X X 12.2 2 5.8 8 5.4 4 4.4 4 5.9 9 4.3 3 5.8 8 9.9 9 7.4 4 12.3 3 4.4 4 9.4 4 7.8 8 12.3 3 6.4 4 9.2 2 4.5 5 7.3 3 9.1 1 7.9 9 10.4 4 5.6 6 7.2 2 5.3 3 3.2 2 y y 15.5 5 10.9 9 11.5 5 33.8 8 27.5 5 9.9 9 14.6 6 13.6 6 8.4 4 15.4 4 11.2 2 28.0 0 5.5 5 16.1 1 6.5 5 11.0 0 4.7 7 5.6 6 10.0 0 8.7 7 10.2 2 8.3 3 7.2 2 11.1 1 6.4 4 z z 6.8 8 21.4 4 13.6 6 5.6 6 9.2 2 7.0 0 7.4* * 6.7 7 5.0* * 9.1* * 6.1 1 16.5 5 9.4 4 21.5 5 6.6 6 5.6 6 5.0 0 4.0 0 13.0* * 7.3 3 LLLL = left lower lobe, RLL = right lower lobe, RML - right middle lobe, LUL = left upper lobe, RUL == right upper lobe. Central = lesion with no lung interposed between it and the mediastinum. Mediastinumm = invasion of mediastinum, pleura = invasion of pleura, chest - invasion of chest wall.. Tumor movement = average of methods 1 and 2, x = left-right, y - caudal-cranial, z = posterior-anterior. .

(11)

Theree was significantly more movement in the y-direction (p<0.001 Wilcoxian t-test) than in the x-direction,, but not when compared to the z-direction (p = 0.08). The range of movement in the x-- and z-direction was not significantly different (p = 0.14). There was significantly (p = 0.003 Mannn Whitney U-test) more movement seen in the lower/middle lobe tumors (mean range of movementt 16.7 mm) than in the upper lobes (mean range of movement 8.8 mm) in the y-direction.. Patient 13 with a lesion spanning the upper and middle lobes was excluded. In this series,, neither tumor position (central versus peripheral) nor the presence of local invasion, for examplee into the chest wall, had a significant effect on the range of tumor movement (Mann Whitneyy U-test). 1.00 -i AXm m

3 3

0.5 5 -0.33 -L -x(cm) )

FigureFigure 5. The tumor positions of one patient, measured with portal imaging during eight fractions of the treatment. The centercenter of gravity of the tumor for two to three portal images per fraction is indicated by each dot. The Tirst tumor position ofof each day is numbered. The star represents the measured average tumor position during the treatment. The origin is thethe tumor position on the DRR.

Inn order to compare the movement assessed by fluoroscopy to that demonstrated during treatment,, the latter results were graded similar to the pre-treatment grading of tumor movementt during the simulation phase (grade 1: movement < 5 mm, grade 2: movement < 10 mmm and grade 3: movement > 10 mm). In 29 of 70 (41%) fluoroscopic assessments the grade wass identical, in 35 (50%) the grade differed by one and in 6 cases (9%) the disagreement was 22 grades. This might be due to the fact that assessment of tumor motion during fluoroscopy doess not take into account day-to-day (and inter-fractional) changes in breathing level.

Tumorr shrinkage

Thee mean of the measured projected surface during the first week of treatment, in which time framee shrinkage was unlikely to occur, was calculated as the base-line tumor area. To assess thee accuracy of this technique, the variation in the measurements during the first week was calculated,, yielding a standard deviation of 2.4% in the AP projection and a standard deviation off 4.3% in the lateral images. These figures suggest an acceptable degree of consistency. The tumorr area during the treatment was then compared to the base-line measurement (Figures 6 andd 7 show some examples). The average ratio between the baseline size and that in the last

(12)

seriess of portal images (usually taken in the last week of treatment) was 0.83 (SD 0.11, range 0.555 to 0.99) in the AP projection and 0.84 (SD 0.17, range 0.44 to 1.09) in the lateral projection.. Ten of the lesions regressed 20% or more in at least one of the projections. Of the tumorss that regressed during treatment, it took on average 4.5 weeks (2 SD = 1 week) from thee start of the treatment for the size to drop below 80% of the original size. To analyze effect off tumor shrinkage on tumor movement, the results from the intra- and inter-fraction of those patientss with a lesion that regressed, were compared. No systematic time-dependent change inn the range of tumor motion and/or shifts in tumor location were observed in these patients. Onee tumor showed asymmetric shrinkage.

Firstt fraction Lastt fraction

FigureFigure 6. An example of tumor regression during treatment. The initial tumor area is delineated in white and the residual tumortumor area in the last fraction of the treatment is delineated in black tumor. (Note that tumor motion is also visible).

J5 5

€% %

AP

*** * < & * ***

OO Lateral

% %

..

B 100 20 Fractionn number 100 20 Fractionn number 30 0

FigureFigure 7. The ratio of the projected tumor size through the course of treatment for a lesion that regressed more than 40%40% in both AP and lateral projections (left panel) and for a lesion that varied little in size (right panel). The average amountamount of tumor regression was in-between.

(13)

Comparisonn of planning CT position to treatment position

Forr 16 of the 25 patients it was possible to measure the center of the lesion on the AP DRRs derivedd from the planning CT scan in the x- and y-direction. For 10 of the 16 patients the center couldd also be determined on the lateral DRR. The average distance between the center of the lesionn on the CT scan and the average tumor position during treatment as measured with the portall images was 6 mm in vector length (for more details, see Table 4). A histogram of the differencee between the planned and average measured tumor positions was plotted (Figure 8). Forr all patients the difference between the planned and measured average tumor positions duringg treatment was less than 1 cm. The average difference over all patients did not differ fromm zero.

Tablee 4. Difference between planned (DRR)

treatmentt with electronic portal imaging.

xx (mm), n = 16 Averagee -2

SDD 4

andd average tumor position measured during

yy (mm), n = 16 z (mm), n = 10

22 1 44 3

xx = left-right, y = cranial-caudal, z = anterior-posterior nn = number of evaluated patients

-22 -1.5 -1 -0.5 0 0.5 1 1.5 2 Differencee in tumor position between DRR and EPID (cm)

FigureFigure 8. The difference between the tumor position on the DRR and the average tumor position as measured with portalportal imaging during treatment, in 3 directions for 16 patients in the x- and y-direction and for 10 patients in the z-direction. z-direction.

Discussion n

Overr the last decade 3D-CRT has become the standard of care in the treatment of inoperable butt localized NSCLC. Several series have reported promising results (Graham 1995,

(14)

Porta]Porta] imaging to assess set-up errors and tumor motion

Armstrongg 1997) with acceptable levels of toxicity. However, the tocal control rate with

conventionall doses remains poor (Martel 1999), which has fuelled the interest in the possibility

off dose escalation (Robertson 1997, Belderbos 2000, Mehta 2001, Hayman 2001), Martel et

al.. (1999) examined the radtobtology of NSCLC and suggested that a 50% tumor control

probabilityy will require prescribed doses in excess of 85 Gy when treating with standard 2 Gy

perr day fractionation. In order to deliver these doses safely, with the minimum of long-term

sequelae,, particular attention must be paid to the volume of normal lung mat is irradiated. To

thiss end, attempts are being made to minimize the margin added to the GTV to produce the

PTV,, however, without careful assessment this strategy risks geographical misses.

Firstly,, a margin to account for microscopic disease is added to the GTV to construct the CTV.

Inn a recent study, Gtraud et al. (2000) examined 70 NSCLC surgical resection specimens for

microscopicc extension. They concluded that in order to treat 95% of microscopic extensions,

aa margin of 6 mm is required for squamous cell carcinomas and 8 mm for adenocarcinomas.

Oncee the CTV has been created the PTV must be defined.. This should take into account the

nett effect of aU possible geometrical variations and inaccuracies in order to obtain a clinically

acceptablee and specified probability that tine prescribed dose is actually received by the CTV

(vann Herk 2000). The geometrical variations consist of systematic and random setup errors

andd organ movement.

Setupp verification

Systematicc errors are due to discrepancies that occur during the preparation of the treatment

plan.. These can be due several factors such as, target delineation errors, position of the

patientt and the tumor when the planning CT scan is acquired compared to that during

treatmentt or the use of simulator images rather than DRRs as setup reference images (Bel

1994,, de Boer 2001). Random errors result from day-to-day variations in patient position and

organn motion.

Too identify setup inaccuracies using thoracic portal images, comparison must be made to

areass of relative stability. Samson et al. (1999) demonstrated that the thoracic wall, trachea

andd clavicles are the most stable structures and the correlation of the position of two of these

landmarkss on the reference image and the portal image, produces the most reliable

comparison. .

Severall studies investigated setup inaccuracies in thoracic radiotherapy but not all studies

analyzedd systematic and random errors separately. Ekberg et al. (1998) quoted an overall

meann error of 3.1 mm (SD 4.0), 3.6 mm (SD 4.6) and 2.9 mm (SD 3.8) in the x-, y- and

z-direction,, respectively. The studies that define systematic errors and random errors separately

aree summarized in Table 5. De Boer et al. (2001) compared portal images to simulator images

andd DRRs. They demonstrated that there could be a large systematic error if simulator images

aree used rattier than DRRs fttx = 0.4 mm and Xx = 4.0 mm, \iy = 0.6 mm and £y = 2.8 mm,

jizz = 0.3 mm 2z = 2.5 mm). In our study, setup errors similar to that in other series were

observedd (Table 5), with standard deviation tor the systematic errors (without correction

protocol)) in the x-, y- and z-directions of 3.3 mm, 4.4 mm and 2.2 mm. respectively. The

standardd deviations of the random errors were 2.9 mm, 3.1 mm and 2.0 mm. The correction

protocoll for systematic setup errors was quite efficient (Figure 3); the systematic errors were

reducedd to 1.4 mm, 1.5 mm, and 1.3 mm, respectively. This reduction is important, as the

impactt of systematic errors on the dose to the CTV is far greater titan the impact of random

errorss (van Herk 2000).

(15)

Breathingg motion

AA possible source of error in the treatment of thoracic tumors is the extreme movement exhibitedd by some lesions. The development of high-resolution portal imaging devices has enabledd us to study the movement of bronchogenic neoplasms during a six to seven week coursee of irradiation. The use of portal images in mis study could urvder-estimate the range of movementt (Figure 1) as only three images were acquired per fraction covering about half of thee breathing cycle. Seven to nine images taken at one-second intervals would be more accuratee for the evaluation the entire breathing cycle but would result in a higher radiation exposuree of the patient Inter-fraction changes in breathing level might introduce an overestimationn of the tumor motion.

Thee two methods developed for this study to establish the tumor position produced similar results.. Method 1 could be biased by tumor shrinkage while Method 2 was less accurate if overlayingg structures obscured the edge of the tumor. However, the agreement between the methodss was good. Both methods have demonstrated tumor movement in all directions, which wass most marked in the cranial-caudal direction. It is important to note that there was a wide variationn between individual patients, with the lesions in the lower iobes demonstrating the largestt movement. In this series we were unable to demonstrate reduced motion of lesions withh local invasion but this is probably due to the small number of these patients in this series. Thee fluoroscopic assessment of tumor movement performed at time of simulation appears to havee a reasonable correlation with that observed during treatment, suggesting that mis is a validd method for assessing organ movement on a coarse scale for each individual patient.

Tablee 5. Studies examining setup errors without corrections in thoracic irradiation.

Authors s

Patientss and technique

Dee Boer et at. 40 patients (No specific immobilization). (2001)) Orthogonal portal images compared

too simulator images.

Samsonn etat. 8 patients (Cross bar or none). (1999)) AP portal images compared to

simulatorr images.

Vann de 16 patients (Two hand immobilizer). Steenee et at, AP portal images compared to (1998)) simulator images.

Yann ét al. 27 patients (No specific immobilization). (1997)) AP portal images compared to DRRs.

Currentt study 25 patiënte (forearm support and kneee rolt). Orthogonal portal images comparedd to DRRs.

x x

y y

z z

X X

y y

X X

y y

X X

y y

X X

y y

z z

1.6 6

0.0 0

-0.7 7

1.3 3

-2.1 1

-1.8 8

0.4 4

0.3 3

-0.6 6

-0.2 2

1.6 6

0.5 5

3.2 2

3.6 6

1.7 7

2.5 5

2.0 0

5.1 1

2.8 8

2.5 5

3.5 5

3.3 3

4.4 4

2.2 2

2.0 0

2.1 1

1.8 8

2.0 0

2.8 8

2.7 7

3.5 5

2.3 3

2.7 7

2.4 4

2.6 6

1.8 8

\i\i - mean overall error (mm)

xx = left-right

££ = SD of systematic error (mm) o = SD of random error (mm) yy = cranial-caudal z = anterior-posterior

Rosss et at. (1990) examined the tumor movement in 20 patients using ultra-fast CT scans. The lesionss were scanned at 1 cm intervals with 10 scans performed at each level over 7 seconds

(16)

PortelPortel imaging to assess set-up errors and tumor motion

(2-33 breathing cycles). They demonstrated an average x-movement of 6.1 mm (range 0-22

mm)) and a z-movement of 2.7 mm {range 0-15 mm). They were unable to measure y

(cranial-caudal)) movement with this technique. They also noted that lesions in the lower lobes and

thosee adjacent to the aorta or heart demonstrated the greatest movement, white lesions

attachedd to the chest wall showed very little movement. Shimizu et al. (2000) analyzed the

movementt Of 16 lung tumors in 13 patients. Twenty sequential CT images were acquired

throughh the central slice of the tumor. They demonstrated an average distance of 6.4 mm

(rangee 2.1 to 24.4 mm) between the treatment couch and the posterior border of the tumor and

aa mean distance of 5.1 mm (range 0 to 6.0 mm) between the anterior chest wall and the

anteriorr tumor surface. In addition they estimated that the y-movement was 6.2 mm (range

2.4-11.3)) for the upper/middle lobe lesions and 9.1 mm (range 3.4 to 24.0 mm) in the lower

lobee tumors. Wrth the real-time tumor tracking system (Seppenwookfe 2001b), the measured

averagee peaMo-peak distance of the tumor motion was largest (12 2 mm (SO)) in the

y-directionn for tumors that were situated in the lower lobes and are not attached to rigid

structuress tike the chest wall or vertebrae. In the x- and z-directions the tumor motion was

small,, for both upper and lower lobe tumors (2 1 mm). The much larger motion in these

directionss that was observed in the study, presented in this paper, can be caused by

inter-fractionall changes in breathing level of the patient Ekberg et al. (1998) used fluoroscopy at

timee of simulation to assess tumor movement They demonstrated an average movement of

2.44 mm (SD 1.4, maximum 5.0 mm) in the x-direction, 3.9 mm (SD 2.6, maximum 12 mm) in

thee y-directton and 2.4 mm (SD 1.3, maximum 5.0 mm) in the z-direction. In a recently

publishedd paper, Stevens et al. (2001). used orthogonal radiographs taken at extremes of

movementt to investigate movement in the y-direction. They demonstrated an average

movementt of 4.5 mm (range 0-22 mm) with 10 of the 22 tumors having no movement. The

averagee of the range of movement demonstrated in this study is somewhat greater than that

shownn in previous studies that have used CT scanning (Ross 1990, Shimizu 2000) or

fluoroscopyy (Ekberg 1998). This is probably due to a number of tumors in this series which

exhibitedd a much greater range of movement than has previously been documented (12% of

thee lesions moved more than 20 mm In the y-direction and 10% more than 20 mm in the

z-dlrection).. Furthermore, overestimation of the motion tumor motion can be caused by changes

inn breathing level between different fractions (Figure 1) or partly by delineation and matching

errorss of the technique, which is estimated to be in the order of 3 mm.

Tumorr shrinkage

Thee development of a tool for outlining structures on portal images has enabled us to

demonstratee that tumor shrinkage of at least 20% occurred in 40% of the patients. We are not

awaree mat thus has been previously documented. The observation of tumor shrinkage may

offerr new challenges to dose escalation studies. By repeating the planning CT scan during the

penultimatee week of the treatment for the regressed lesions, the GTV (and PTV) volume can

bee reduced for the last fractions. This will reduce the normal tissue complication probability.

However,, it should be noted that though the size of the visible lesion on a CT scan is smaller,

itt is unknown what has occurred exactly on a microscopic level.

Comparisonn of planning CT position to treatment position

Thee effect of breathing and random setup errors on the dose distribution in the CTV is small

comparedd to that of systematic errors, even for the largest breathing amplitude (Engelsman

2001a).. The current study demonstrates that for a number of patients the technique of

performingg planning CT scans during free breathing produces a substantial systematic error

(17)