- Email: [email protected]
The delineation of target volumes for radiotherapy of lung cancer patients
Radiotherapy and Oncology 91 (2009) 455–460
Contents lists available at ScienceDirect
Radiotherapy and Oncology journal homepage: www.thegreenjournal.com
Treatment planning
The delineation of target volumes for radiotherapy of lung cancer patients Hilke Vorwerk a,*, Gabriele Beckmann b, Michael Bremer c, Maria Degen d, Barbara Dietl e, Rainer Fietkau f, Tammo Gsänger a, Robert Michael Hermann a, Markus Karl Alfred Herrmann a, Ulrike Höller g, Michael van Kampen h, Wolfgang Körber i, Burkhard Maier j, Thomas Martin k, Michael Metz l, Ronald Richter m, Birgit Siekmeyer a, Martin Steder n, Daniela Wagner a, Clemens Friedrich Hess a, Elisabeth Weiss a,o, Hans Christiansen a a
Department of Radiotherapy and Radiooncology, University Hospital Göttingen, Robert-Koch-Straße, Germany Department of Radiotherapy and Radiooncology, University Hospital Würzburg, Josef-Schneider-Straße, Germany Department of Radiotherapy and Radiooncology, Medical School Hannover, Carl-Neubergstraße, Germany d Department of Pneumology, Lungenfachklinik, Waldhof Elgershausen, Greifenstein, Germany e Department of Radiotherapy and Radiooncology, University Hospital Regensburg, Franz-Josef-Strauß-Allee, Germany f Department of Radiotherapy and Radiooncology, University Hospital Erlangen, Universitätsstraße, Germany g Department of Radiotherapy and Radiooncology, Vivantes MVZ Neukölln, Rudower Straße, Berlin, Germany h Department of Radiotherapy and Radiooncology, Krankenhaus Nordwest, Steinbacher Hohl, Frankfurt, Germany i Department of Pneumology, Ev. Krankenhaus Weende, Pappelweg, Bovenden-Lenglern, Germany j Department of Radiotherapy and Radiooncology, Klinikum Bayreuth, Preuschwitzer Straße, Germany k Department of Radiotherapy and Radiooncology, Klinikum Bremen Mitte, St. Jürgen-Straße, Germany l Department of Haematooncology, University Hospital Göttingen, Robert-Koch-Straße, Germany m Department of Radiotherapy and Radiooncology, Klinikum St. Georg, Delitzscher Straße, Leipzig, Germany n Department of Haematooncology, Pius-Hospital, Georgstraße, Oldenburg, Germany o Department of Radiation Oncology, Virginia Commonwealth University, Richmond, VA, USA b c
a r t i c l e
i n f o
Article history: Received 23 May 2008 Received in revised form 5 March 2009 Accepted 7 March 2009 Available online 30 March 2009 Keywords: Radiation therapy Lung cancer Interobserver variability Delineation Target volume
a b s t r a c t Purpose: Differences in the delineation of the gross target volume (GTV) and planning target volume (PTV) in patients with non-small-cell lung cancer are considerable. The focus of this work is on the analysis of observer-related reasons while controlling for other variables. Methods: In three consecutive patients, eighteen physicians from fourteen different departments delineated the GTV and PTV in CT-slices using a detailed instruction for target delineation. Differences in the volumes, the delineated anatomic lymph node compartments and differences in every delineated pixel of the contoured volumes in the CT-slices (pixel-by-pixel-analysis) were evaluated for different groups: ten radiation oncologists from ten departments (ROs), four haematologic oncologists and chest physicians from four departments (HOs) and five radiation oncologists from one department (RO1D). Results: Agreement (overlap P 70% of the contoured pixels) for the GTV and PTV delineation was found in 16.3% and 23.7% (ROs), 30.4% and 38.6% (HOs) and 32.8% and 35.9% (RO1D), respectively. Conclusion: A large interobserver variability in the PTV and much more in the GTV delineation were observed in spite of a detailed instruction for delineation. The variability was smallest for group ROID where due to repeated discussions and uniform teaching a better agreement was achieved. Ó 2009 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 91 (2009) 455–460
Patients with unresectable or medically inoperable non-smallcell lung cancer (NSCLC) are treated with primary radiochemotherapy [1,2]. Although this is a curative approach, these patients still have a poor prognosis with a curative potential of 10–20% for stage IIIA/B. Published literature suggests that tumor control is improved with dose escalation using conformal radiation therapy [3,4].
* Corresponding author. Address: Department of Radiotherapy and Radiooncology, University Hospital Göttingen, Robert-Koch-Straße 40, 37099 Göttingen, Germany. Tel.: +49 551 398866; fax: +49 551 396192. E-mail address: [email protected] (H. Vorwerk). 0167-8140/$ - see front matter Ó 2009 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.radonc.2009.03.014
Whether adjuvant chemotherapy is able to improve survival after primary radiochemotherapy for inoperable NSCLC is still unclear. To answer this question, the GILT-CRT-1-study (multicenter, open-label, randomised, phase III study of oral Vinorelbine in combination with Cisplatin concurrently with radiotherapy to a total dose of 66 Gy) randomises to either two more cycles of consolidation therapy with oral Vinorelbine and Cisplatin plus Best Supportive Care or Best Supportive Care alone in these patients with unresectable locally advanced NSCLC. The comparability between different participants of multicenter studies is compromised by contouring errors which may cause inaccurate radiation treatments
456
Interobserver variability for lung cancer
that result from different RT volumes. Contouring variations are caused by observer-, image- and instruction-related reasons. Within the GILT-study, we wanted to evaluate the agreement of target volume delineation by participants of the study. The target volume definition for conformal radiotherapy may be the most important factor for adequate tumor coverage and sufficient sparing of critical structures [5,6]. The aim of our analysis was to examine, whether we could minimise the variability in GTV and PTV delineation for primary nonsmall-cell lung cancer patients by using a strict presetting for GTV definition and for the determination of the PTV (observer-related reason) with simultaneous exclusion of image- and instruction-related reasons for testing quality assurance. Such a precise delineation protocol to homogenize delineation was demanded by different authors [5,7]. We analysed different groups of physicians: radiation oncologist from different departments, haematologic oncologists and chest physicians from different departments and radiation oncologist from one department.
Methods and materials Delineation was performed during a workshop of the participants of the GILT-CRT-1-study. The physicians, who agreed to participate in the workshop, could obtain all patient informations digitally in the forefront of the meeting. At the beginning of the meeting all participants received a folder with all patient data to make themselves familiar with the facts. Contouring was done in the planning rooms of one department under survey of two physicists. Interaction with any physician during contouring was not allowed. Eighteen physicians were asked to contour the GTV and PTV for three patients with non-small-cell lung cancer stage IIIB scheduled for primary radiation therapy. Contouring was done by experienced physicians from fourteen different departments, divided into three groups: RO ten senior radiation oncologists (G.B., M.B., B.D., R.F., U.H., M.v.K., B.M., T.M., R.R., H.V.) from ten different departments used to 3D planning. HO four senior haematologic oncologists/and chest physicians (M.D., W.K., M.S., M.M.) from four different departments (further noticed as haematologic oncologists). RO1D four senior and one experienced junior radiation oncologists (R.M.H., M.H., M.N., B.S., H.V.) from one department, all familiar with 3D planning. Patient data were acquired from the first three men, who were treated in the GILT-CRT-1-study (Pierre Fabre Study Code: PM 0259 CA 304 J1; EudraCT number: 2004-005135-26) in the University of Göttingen, Germany. Before contouring every physician was provided with written information on every patient, including history, clinical examination, general performance status, bronchoscopic findings, lung function tests and computed tomography (CT) scans with iodine contrast enhancement. The CT-scans were also available in digital form. All data were made anonymous. The median age of the patients was 60 years. All patients were planned for curative radiochemotherapy in combination with Cisplatin and Vinorelbine and a total radiation dose of 66.0 Gy (5 2.0 Gy per week) with reduced field according to the study protocol. The field was reduced after 50.0 Gy utilizing a new restaging CT according to the study protocol. For our analysis we used the first CT before beginning of the treatment and analysed the contouring planned for the first part of the treatment up to 50.0 Gy.
The stage and location of the tumors were as follows: 1. cT3 cN3 M0 (left central, partial atelectasis of the lower left lobe, central and upper mediastinal lymph nodes), 2. cT3 cN3 M0 (right central, central mediastinal lymph nodes), 3. cT2 cN3 M0 (right central, central mediastinal lymph nodes). Continuous 5 mm CT-scans (10 mm in one patient) of the treatment region in the thorax were obtained with a single-slice spiral CT scanner (Somatom Balance, Siemens Medical Systems, Forchheim, Germany) for all three patients without contrast enhancement. We used a resolution of 0.94 mm in axial plane, a couch shift of 7.5 mm and a field-of-view of 480 mm (130 kV, 100 mAs). Physicians entered their contours directly at the treatmentplanning console (Eclipse, Version 6.5, Varian Medical Systems, Palo Alto, CA, USA). Radiation physicians not familiar with this planning system and haematologic oncologists beforehand underwent introductions in the management of the contouring workspace by a physicist. Particularly, it was demonstrated how to change the window level between soft tissue and lung window and how to measure distances. Sagittal and coronar CT reconstruction were available to allow orientation in craniocaudal direction. All physicians had permanent technical assistance during contouring by two physicists (D.W., T.G.). All physicians worked on their own CT image-set. Contouring was performed independently without information of other participants. Two physicians from the group ROs contoured only the CT from one patient. All physicians obtained a detailed instruction, which is provided by the GILT-CRT-1-study protocol, for definition of the gross target volume (GTV) and planning target volume (PTV) as follows: Gross tumor volume: 1. extension of the primary tumor in the soft tissue and lung window, 2. lymph nodes larger than 1.5 cm in cross diameter in the transversal CT-scans, 3. lymph nodes with central hypodensity, 4. lymph nodes larger than 1.0 cm in cross diameter in the transversal CT-scans, if multiple nodes are seen in the same compartment (defined by Naruke et al. [8]). It was possible to contour different structures in one volume (for example the GTV). The limits for the different window levels could be chosen by the participants. The structures could be contoured by a paint brush or pixel-by-pixel. Every physician received a detailed anatomic sagittal picture of the lymph node compartments according to the Naruke Classification [8] and described in more detail by Kirikuta et al. [9]. For easier distinction of the compartments we enhanced a transversal CTmap created by Kirikuta et al. [9] with coloured flags for all lymph node compartments. Planning target volume: 1. PTV is defined as GTV plus a safety margin of 1.0 cm in lateral direction, 1.0 cm in ventrodorsal direction and 1.5 cm in craniocaudal direction via automatic expansion (This considers both oncological (subclinical extension [CTV]) and geometrical (setup and motion uncertainty) factors). 2. If infracarinal involvement is suspected, caudal rim of the PTV should be placed at least 3 cm below the carina. It was possible to modify the PTV contour after automatic expansion of the GTV. We performed a volume analysis, an anatomic analysis and a pixel-by-pixel analysis of all contoured parts of all CT-slices similar to that explained in more detail below for different groups of
H. Vorwerk et al. / Radiotherapy and Oncology 91 (2009) 455–460
physicians. In the first part a comparison of ten radiation oncologists (ROs) from ten different departments with four haematologic oncologists (HOs) from four different departments were performed. In the second part the ten radiation oncologists from the first part were compared with five radiation oncologists from one department (RO1D). 3D-volumes of the GTV and PTV were calculated by use of the planning system EclipseÒ. The medians, standard deviations and ratio of maximum to minimum volume (Vmax/Vmin) were calculated for every patient and physician. Additionally, we assessed the deviation from the mean volume, calculated from the fourteen different departments, i.e. from RO’s and HO’s. We analysed which anatomic sites of the lymph node compartments were considered to be involved. A compartment was considered as contoured by a participant when at least a part of the compartment was contoured. A checklist was created by us which contained the following items: upper and lower paratracheal lymph nodes left and right, preaortic lymph nodes, subaortic lymph nodes (A-P window), hilar lymph nodes left and right, subcarinal lymph nodes, bronchopulmonal lymph nodes left and right, paraesophageal lymph nodes (below carina). This checklist was completed for every outlined GTV and analysed for agreement between the physicians. Agreement was stated, when 70% or more of the observers included or excluded the lymph node compartment in their contour of the structure. In our clinic we developed a mapping tool in order to analyse the concordance between the observers pixel-by-pixel. All contoured structures were exported from EclipseÒ with Dicom format, were converted into ASCII format and imported in MicrosoftÒ Office Excel 2003. The structures were specified with different points on the circumference of the structure. We composed a graticule with one cubic centimetre equivalent to 100 pixel points. Every pixel point of the contoured circumference from the structure was plotted as a ‘‘1” into one table map of ExcelÒ per physician, patient and z-axis of the CT-scans. We developed a source code to provide a linear regression between the pixel points on the circumference from the contoured structure, like the planning system EclipseÒ, and filled the area between the lines with the number
Fig. 1. Picture of a map from the pixel-by-pixel analysis of the contoured structures by 13 physicians. All structures per physician were displayed by the number 1 for all pixels in the contoured area and afterwards added for each pixel point. For better visualization we coloured the different areas of numbers (The darker the colour the higher the number). Hundred pixel points are equivalent to one cubic centimetre.
457
‘‘1” using an in house source code. Along this way we received a map of the structures in all transversal CT-scans of all patients and physicians, pixel-by-pixel. Afterwards we added up all maps of each group, who should be investigated, and achieved a new map with numbers equivalent to the number of physician, who had contoured this pixel (Fig. 1). This analysis detected differences in the 2D-transversal CT-slices for all slices. On this way a statement for the 3D-volume was possible. Therewith we calculated the agreement between the physician groups. Agreement was stated, if 70% or more of the observers included the pixels on the transversal CT slice in their contour of the structure. To identify the physicians’ level of agreement, encompassing and common volumes were calculated for every patient. The encompassing volumes are the smallest volumes integrating all entered volumes of one patient. The common volumes are the largest volumes that are part of all GTVs or PTVs of one patient [10–12].
Results Firstly, we analysed the delineation of the fourteen different clinics. Especially, we compared the ROs against the HOs. Secondly, we analysed the differences between the ROs from ten different departments and the RO1D from one department. The mean volume and the standard deviation of the GTV and PTV volumes contoured by the haematologic oncologists (HO’s) were significantly smaller than the contours of the radiation oncologists (RO’s) (Table 1) with 50–68% of the HO volumes relative to the RO volumes. The mean volumes of both the groups of radiation oncologists (ROs and RO1D) are comparable for the GTV and PTV with a much smaller standard deviation by the RO1D. The volumes delineated by the RO1D ranged from 83% to 90% of the volumes delineated by the ROs. The deviation from the mean GTV of all patients calculated for all fourteen different physicians (ROs and HOs), were largest considering the ROs (30.5 ± 86.2 cm3) and comparable considering the HOs ( 34.0 ± 75.2 cm3) and RO1D ( 1.9 ± 39.0 cm3). The differences between the three groups get larger regarding the contoured PTV (ROs: 46.4 ± 126.4 cm3, HOs: 30.3 ± 171.88 cm3, RO1D: 0.7 ± 89.9 cm3). For ROs and RO1D, the deviation of the PTV was similar to the deviation of the GTV, but the standard deviation was larger. The standard deviation of the PTV contours is larger than that of the GTV contour, especially for the HOs. We found a good agreement (P70% of the observers were conform) in 81% (92%, 92%) of the investigated lymph node compartments for the ROs (HOs, RO1D). On average the ROs and RO1D included 6.3 of all 12 lymph node compartments as defined in methods and materials in their contours in contrast to only 4.4 of the 12 lymph node compartments by the HOs. We compared the agreement of the contoured parts of all CTslices for all patients and physicians. We divided all slices of the CT in small parts of 0.001 cm3 and compared all of them with one another. The accurate analysis of this pixel-by-pixel mapping procedure leads to less agreement than the volume and anatomic analysis. Only 70% of the ROs contoured 16.3% of the encompassing GTV volume (smallest contoured part of the slices integrating all entered contours), 70% of the HOs and RO1D contoured 30.4% and 32.8% of the encompassing GTV, respectively (Table 2). The analysis of the fourteen different departments showed even fewer agreements (12.4%). The agreement between the physicians is better concerning the encompassing PTV volume. 70% of the ROs contoured 23.7% of the encompassing PTV volume in comparison to 38.6% (35.9%) by the HOs (RO1D). The common GTV volumes were very small for the 14 departments but also for the ROs and HOs (6.2–14.9 cm3) and not much larger for the RO1D (36.5 cm3). The smallest encompassing GTV
458
Interobserver variability for lung cancer
Table 1 Magnitude of volumes for each observer group. Mean volume over all physicians (cm3) GTV
14 Departments ROs HOs RO1D
PTV
Patient 1
Patient 2
Patient 3
Patient 1
Patient 2
Patient 3
137.5 ± 98.8 162.6 ± 105.6 81.1 ± 45.1 132.8 ± 50.8
157.1 ± 84.5 183.7 ± 86.9 97.5 ± 33.1 151.6 ± .47.8
179.6 ± 110.5 213.1 ± 116.3 122.5 ± 52.8 180.9 ± 38.6
458.0 ± 219.1 531.8 ± 206.4 291.8 ± 142.7 438.2 ± 104.8
400.1 ± 118.1 450.0 ± 94.1 287.7 ± 84.7 393.8 ± 125.2
555.8 ± 165.8 590.1 ± 183.6 487.2 ± 89.7 582.2 ± 85.5
ROs, radiation oncologists from 10 departments; HOs, haematologic oncologist/s pneumologists; RO1D, radiation oncologists from one department; GTV, gross target volume; PTV, planning target volume.
Table 2 Agreement of the contoured parts of the transversal CT-slices of the physicians with a pixel-by-pixel analysis. Agreement in 70% of contoured pixels (%) GTV 14 1.2 10 ROs 4 HOs 5 RO1D PTV 14 3.3 10 ROs 4 HOs 5 RO1D
Departments
16.3 ± 4.1 30.4 ± 8.0 32.8 ± 3.7 Departments
23.7 ± 3.9 38.6 ± 6.1 35.9 ± 2.2
Common volume (cm3) 12.4 ± 4.3
Encompassing volume (cm3)
Common/ encompassing volume (%)
6.2
507.5
14.9 9.5 36.5
422.9 227.4 225.0
3.3 4.0 15.8
19.3 ± 4.3
52.9
1446.2
1124.0 402.7 643.5
7.5 15.1 17.4
83.1 70.8 121.1
Agreement as stated, when 70% of the physicians had contoured the part of the transversal CT slice. The encompassing volumes are the smallest volumes integrating all entered volumes of one patient. The common volumes are the largest volumes that are part of all GTVs or PTVs of one patient. The described volumes are the sum of all three patients. ROs, radiation oncologists from 10 departments; HOs, haematologic oncologist/s pneumologists; RO1D, radiation oncologists from one department; GTV, gross target volume; PTV, planning target volume.
volumes could be found contoured by the RO1D and the HOs, whereas the encompassing volume of the 14 departments showed more than twice as much cubic centimetre. The relation between the common and encompassing volumes was less than 5% except for the RO1D with 15.8%. The observations for the common and encompassing GTV volumes are nearly the same for the PTV volumes with altogether larger volume. But the relation between the common and encompassing volumes is larger for all groups of physicians, especially for the HOs. Discussion Within the GILT-CRT-1-study (multicenter study for patients with unresectable locally advanced NSCLC), we wanted to evaluate the agreement of target volume delineation by participants of this study. The target volume definition for conformal radiotherapy may be the most important factor for adequate tumor coverage and sufficient sparing of critical structures [5]. Within a study protocol every effort should be taken to ensure that the contours are standardized and accurate with high interobserver reproducibility. Contouring variations result in systematic errors of radiation treatment that are a confounding factor for the interpretation of treatment results. Besides, the comparability between different departments, who take part in multicenter studies, will be very difficult. More accurate delineation of the GTV and PTV may result in
better conformal target coverage, improved sparing of normal tissue, and ultimately enhanced local control. Several reports on the variability of GTV and CTV delineation have been published for a variety of malignancies, such as tumors of the lung, breast, cervix, bladder, pancreas and prostate and of the head and neck region [5,10,13–18]. The principle reasons for interobserver variability’s can be image-, instruction- and observer-related. For our analysis, we wanted to analyse only observer-related reasons for interobserver variability. One major factor for interobserver variability is the different assessments regarding the risk of spreading of tumor [5,7]. The purpose of our analysis was the difference of interpretations whether a lymph node was involved or not and the extension of tumor size. The first patient presented an atelectasis of the left lower lobe, which could increase the variations. Other reasons are training, differences in the attention to detail in the contouring process itself and subjective interpretation of image knowledge of topographical cross-sectional anatomy. The observed wide range in volume size for the GTV (22.7– 486.9 cm3) and PTV (119.0–927.1 cm3) is comparable to the results observed for lung cancer [5]. The mean volume of tumors, suffer from rather large neoplasms, may cause a problem regarding interobserver variation [5]. The maximum to minimum volume (Vmax/ Vmin) from the GTV is larger for the fourteen departments (10.1) and comparable for the different physician groups (2.3–7.9) then described in the published literature. Van de Steene et al. found a ratio between 1.1 and 7.6 for the Vmax/Vmin of the GTV in patients with lung tumor. For some other tumors a quite good agreement on target volumes was observed, especially for the prostate [10,13]. Obviously, tumors that have obvious borders like the prostate can be outlined more consistently than tumors without clear border to the neighbouring organs. The Vmax/Vmin of the GTV (7.9) delineated by the ROs is much larger than that of the PTV (2.4). The rigid instruction for delineation gave the order to enlarge the GTV volume (possible automatically) with a default offset, so that we could assume, that the Vmax/ Vmin of the GTV and PTV could be of the same order, similar to that observed in the groups of the HOs and RO1D. The ROs have varied their PTV after automatic enlargement, especially in lateral directions, although this was not part of the study protocol. Thereby the large differences in the GTV delineation were reduced for the PTV. This is reflected in the lesser differences between the ROs and the HOs/RO1D in the PTV contours, whereby the clinical impact of the differences in GTV delineation will be much reduced. The accurate definition of the GTV can be assed by pathologycorrelated imaging, if patients with small tumors underwent surgery after radiotherapy. Thereby most GTV definitions are too larger, but can also be too small [19]. We recommend to focus not only on the GTV delineation but also more on the PTV delineation, because subsequent steps use the initial GTV data. Dosimetric and field planning decisions are based primary on the PTV and secondarily on the GTV.
H. Vorwerk et al. / Radiotherapy and Oncology 91 (2009) 455–460
The definition of lymph nodes classified as pathologic was exactly defined with P1.5 cm or many lymph nodes with P1.0 cm in a lymph node compartment (definition by Naruke et al. [8] and Kirikuta et al. [9]). Every participant received a detailed CTmap of the lymph node compartments. Particularly it was demonstrated to all physicians how to measure diameters of the lymph nodes. This led to a good agreement in 100% and P80% of all analysed lymph node regions in 42% and 75% of all participants, respectively. We found only disagreement in the paratracheal medial left paraesophageal and praeaortal lymph node regions. Another error was due to confusion between lymph node, esophagus and vessels. Other authors found 63% of involved lymph node regions were delineated correctly by the clinicians and 22% of the drawn lymph node regions were accepted to be false positive [5]. This correlates exactly with 75% agreement of all physicians in our analysis. So that we must state, that strict measuring of lesions with a rigid instruction for delineation could not notably reduce interobserver variations in lymph node judgement in our analysis. The pixel-by-pixel analysis compares every part of the contours against the contours of the other physicians instead of only mathematical values and anatomical sites. This reflects the exact differences between the participants. The differences in the GTV contours of the ROs and HOs (16.3% and 30.4%) were respectable in the analysis group by group, but less compared with each other groups (12.4%), because the HOs delineated smaller parts of the CTslices than the ROs but are more consistent among each other (Table 2). The relation between these two groups was reflected in the PTV delineation with better agreement within the groups. The agreement between the RO1D is higher than that between the ROs even in the GTV (32.8%) but also in the PTV contours (35.9%) and comparable to the HOs. This was also reflected in the relation between the common and encompassing volumes. Our analysis showed large discrepancies regarding the delineation of the lymph node regions by the HOs. This may be caused by less knowledge regarding the large mediastinal vessels and the position of esophagus in CT-scans by the HOs in comparison to the ROs. On the other hand, we found small discrepancies between the HOs regarding the delineation of the tumor. The HOs had another point of view than the ROs regarding the tumor because they are used to define the tumor expansion based on combination of bronchoscopy and CT. This could be the reason for this good agreement. The contoured volumes by the HOs are significantly smaller than contoured by the ROs and RO1D. This could be a consequence of the RO’s fear of missing a part of the tumor tissue and therefore design larger contours, whereas the HOs are more afraid of making a false positive error. The agreement in the group of RO1D regarding the delineated lymph node regions was superior with 92% agreement compared to the other groups and published data with 75%. We found the same good agreement in the GTV and also the PTV contoured by the RO1D in all analysis in contrast to the ROs. This implicates that the RO1D used the automatic expansion software of the contouring program and did not correct the PTV free-hand correction afterwards. Remarkably we found the same agreement in the PTV delineation after free-handly correction by the ROs and only automatically enlarged GTV by the RO1D. We suggest, that we could detect better agreement because the oncologists from one department are used to compare and discuss their target volumes daily for better accordance concerning the delineation of the tumor as well as the lymph node regions. The interobserver differences would not be relevantly improved by use of a restrictive instruction, but intra-institutional agreement can diminish differences in target volume delineation. The interobserver variabilities vary by use of different imaging techniques. PET-CT is superior to CT in delineation of the tumor and the lymph nodes [3,20–22], but PET was not requested in
459
GILT-study. By use of PET the variability between the participants would probably be smaller, especially in regions of atelectasis [23]. Delineation accuracy can be improved by using CT window level information. The CT lung windows correlate better with the pathologic size of lung adenocarcinoma than do the mediastinal windows [24]. Grills et al. described an overestimation of the gross pathologic size by a mean of 5.8 mm and an underestimation of the gross tumor size plus microscopic extension by a mean of 1.2 mm by use of the mediastinal window. All participants were instructed to delineate the tumor expansion in the lung and mediastinal window and pathologic lymph nodes – not the whole lymph node region – to deal with this problem. Nevertheless this led to no diminishment of the variation. Difficulties in discrimination between tumor and atelectasis are widely known and a major factor in interobserver variations [12]. In our analysis we could not detect higher discrepancies of the contours of patient 1 (with atelectasis) in comparison to patient 2 or 3. Possibly a microinvasive insertion of radio-opaque markers may be helpful in diminishing interobserver variabilities [25]. Conclusions No ‘‘golden standard” in GTV or PTV delineation can be defined so far. We will not till then be able to make appraisals of the clinical impact, if all physicians define the PTV in the same way with a much better agreement, than is currently detected. A good training of radiation oncologists, a good collaboration between radiation oncologists and radiologist, dummy runs for multicenter studies, establishment of central reviewing boards or detailed instructions for delineation (tested in this analysis) are recommended. However, up to now these procedures have not been shown to decrease the interobserver variability substantially [6,12,17,18,26,27]. Nevertheless, we could not demonstrate an improvement in agreement between the physicians by use of a detailed instruction for delineation compared to existing literature with no detailed instructions, but we could demonstrate, that physicians from one department agree more than physicians from different departments. We recommend online available examples and the possibility of comparison of the own delineation in order to control oneself periodically for better agreement in contouring of target volumes. A ‘‘golden standard” for target delineation should be developed taking relapse patterns into account. References [1] Jeremic B, Machtay M. Concurrent radiochemotherapy in the treatment of locally advanced non-small cell lung cancer. Hematol Oncol Clin North Am 2004;18:91–100. [2] Curran Jr WJ. Therapy for stage III non-small-cell lung cancer: Bob Beamon or Chemo-Beamo? Clin Lung Cancer 2006;8:84. [3] Hong R, Halama J, Bova D, Sethi A, Emami B. Correlation of PET standard uptake value and CT window-level thresholds for target delineation in CT-based radiation treatment planning. Int J Radiat Oncol Biol Phys 2007;67:720–6. [4] Martel MK, Ten Haken RK, Hazuka MB, et al. Estimation of tumor control probability model parameters from 3-D dose distributions of non-small cell lung cancer patients. Lung Cancer 1999;24:31–7. [5] Van de Steene J, Linthout N, de Mey J, et al. Definition of gross tumor volume in lung cancer: inter-observer variability. Radiother Oncol 2002;62:37–49. [6] Weiss E, Hess CF. The impact of gross tumor volume (GTV) and clinical target volume (CTV) definition on the total accuracy in radiotherapy theoretical aspects and practical experiences. Strahlenther Onkol 2003;179:21–30. [7] Giraud P, Elles S, Helfre S, et al. Conformal radiotherapy for lung cancer: different delineation of the gross tumor volume (GTV) by radiologists and radiation oncologists. Radiother Oncol 2002;62:27–36. [8] Naruke T, Suemasu K, Ishikawa S. Lymph node mapping and curability at various levels of metastasis in resected lung cancer. J Thorac Cardiovasc Surg 1978;76:832–9. [9] Kiricuta IC. Selection and delineation of lymph node target volume for lung cancer conformal radiotherapy. Proposal for standardizing terminology based on surgical experience. Strahlenther Onkol 2001;177:410–23. [10] Rasch C, Barillot I, Remeijer P, et al. Definition of the prostate in CT and MRI: a multi-observer study. Int J Radiat Oncol Biol Phys 1999;43:57–66.
460
Interobserver variability for lung cancer
[11] Weltens C, Menten J, Feron M, et al. Interobserver variations in gross tumor volume delineation of brain tumors on computed tomography and impact of magnetic resonance imaging. Radiother Oncol 2001;60:49–59. [12] Valley JF, Bernier J, Tercier PA, et al. Quality assurance of the EORTC radiotherapy trial 22931 for head and neck carcinomas: the dummy run. Radiother Oncol 1998;47:37–44. [13] Logue JP, Sharrock CL, Cowan RA, et al. Clinical variability of target volume description in conformal radiotherapy planning. Int J Radiat Oncol Biol Phys 1998;41:929–31. [14] Petersen RP, Truong PT, Kader HA, et al. Target volume delineation for partial breast radiotherapy planning: clinical characteristics associated with low interobserver concordance. Int J Radiat Oncol Biol Phys 2007;69:41–8. [15] Smith WL, Lewis C, Bauman G, et al. Prostate volume contouring: a 3D analysis of segmentation using 3DTRUS, CT, and MR. Int J Radiat Oncol Biol Phys 2007;67:1238–47. [16] Weiss E, Richter S, Krauss T, et al. Conformal radiotherapy planning of cervix carcinoma: differences in the delineation of the clinical target volume. A comparison between gynaecologic and radiation oncologists. Radiother Oncol 2003;67:87–95. [17] Yamazaki H, Nishiyama K, Tanaka E, et al. Dummy run for a phase II multi-institute trial of chemoradiotherapy for unresectable pancreatic cancer: inter-observer variance in contour delineation. Anticancer Res 2007;27: 2965–71. [18] Coles CE, Hoole AC, Harden SV, et al. Quantitative assessment of inter-clinician variability of target volume delineation for medulloblastoma: quality assurance for the SIOP PNET 4 trial protocol. Radiother Oncol 2003;69: 189–94.
[19] Stroom J, Blaauwgeers H, van Baardwijk A, et al. Feasibility of pathologycorrelated lung imaging for accurate target definition of lung tumors. Int J Radiat Oncol Biol Phys 2007;69:267–75. [20] van Baardwijk A, Bosmans G, Boersma L, et al. PET-CT-based auto-contouring in non-small-cell lung cancer correlates with pathology and reduces interobserver variability in the delineation of the primary tumor and involved nodal volumes. Int J Radiat Oncol Biol Phys 2007;68:771–8. [21] Fox JL, Rengan R, O’Meara W, et al. Does registration of PET and planning CT images decrease interobserver and intraobserver variation in delineating tumor volumes for non-small-cell lung cancer? Int J Radiat Oncol Biol Phys 2005;62:70–5. [22] Jeong SY, Lee KS, Shin KM, et al. Efficacy of PET/CT in the characterization of solid or partly solid solitary pulmonary nodules. Lung Cancer 2008;61:186–94. [23] Fitton I, Steenbakkers RJ, Gilhuijs K, et al. Impact of anatomical location on value of CT-PET co-registration for delineation of lung tumors. Int J Radiat Oncol Biol Phys 2008;70:1403–7. [24] Grills IS, Fitch DL, Goldstein NS, et al. Clinicopathologic analysis of microscopic extension in lung adenocarcinoma: defining clinical target volume for radiotherapy. Int J Radiat Oncol Biol Phys 2007;69:334–41. [25] Kupelian PA, Forbes A, Willoughby TR, et al. Implantation and stability of metallic fiducials within pulmonary lesions. Int J Radiat Oncol Biol Phys 2007;69:777–85. [26] Seddon B, Bidmead M, Wilson J, Khoo V, Dearnaley D. Target volume definition in conformal radiotherapy for prostate cancer: quality assurance in the MRC RT-01 trial. Radiother Oncol 2000;56:73–83. [27] Roy AE, Wells P. Volume definition in radiotherapy planning for lung cancer: how the radiologist can help. Cancer Imaging 2006;6:116–23.
Contents lists available at ScienceDirect
Radiotherapy and Oncology journal homepage: www.thegreenjournal.com
Treatment planning
The delineation of target volumes for radiotherapy of lung cancer patients Hilke Vorwerk a,*, Gabriele Beckmann b, Michael Bremer c, Maria Degen d, Barbara Dietl e, Rainer Fietkau f, Tammo Gsänger a, Robert Michael Hermann a, Markus Karl Alfred Herrmann a, Ulrike Höller g, Michael van Kampen h, Wolfgang Körber i, Burkhard Maier j, Thomas Martin k, Michael Metz l, Ronald Richter m, Birgit Siekmeyer a, Martin Steder n, Daniela Wagner a, Clemens Friedrich Hess a, Elisabeth Weiss a,o, Hans Christiansen a a
Department of Radiotherapy and Radiooncology, University Hospital Göttingen, Robert-Koch-Straße, Germany Department of Radiotherapy and Radiooncology, University Hospital Würzburg, Josef-Schneider-Straße, Germany Department of Radiotherapy and Radiooncology, Medical School Hannover, Carl-Neubergstraße, Germany d Department of Pneumology, Lungenfachklinik, Waldhof Elgershausen, Greifenstein, Germany e Department of Radiotherapy and Radiooncology, University Hospital Regensburg, Franz-Josef-Strauß-Allee, Germany f Department of Radiotherapy and Radiooncology, University Hospital Erlangen, Universitätsstraße, Germany g Department of Radiotherapy and Radiooncology, Vivantes MVZ Neukölln, Rudower Straße, Berlin, Germany h Department of Radiotherapy and Radiooncology, Krankenhaus Nordwest, Steinbacher Hohl, Frankfurt, Germany i Department of Pneumology, Ev. Krankenhaus Weende, Pappelweg, Bovenden-Lenglern, Germany j Department of Radiotherapy and Radiooncology, Klinikum Bayreuth, Preuschwitzer Straße, Germany k Department of Radiotherapy and Radiooncology, Klinikum Bremen Mitte, St. Jürgen-Straße, Germany l Department of Haematooncology, University Hospital Göttingen, Robert-Koch-Straße, Germany m Department of Radiotherapy and Radiooncology, Klinikum St. Georg, Delitzscher Straße, Leipzig, Germany n Department of Haematooncology, Pius-Hospital, Georgstraße, Oldenburg, Germany o Department of Radiation Oncology, Virginia Commonwealth University, Richmond, VA, USA b c
a r t i c l e
i n f o
Article history: Received 23 May 2008 Received in revised form 5 March 2009 Accepted 7 March 2009 Available online 30 March 2009 Keywords: Radiation therapy Lung cancer Interobserver variability Delineation Target volume
a b s t r a c t Purpose: Differences in the delineation of the gross target volume (GTV) and planning target volume (PTV) in patients with non-small-cell lung cancer are considerable. The focus of this work is on the analysis of observer-related reasons while controlling for other variables. Methods: In three consecutive patients, eighteen physicians from fourteen different departments delineated the GTV and PTV in CT-slices using a detailed instruction for target delineation. Differences in the volumes, the delineated anatomic lymph node compartments and differences in every delineated pixel of the contoured volumes in the CT-slices (pixel-by-pixel-analysis) were evaluated for different groups: ten radiation oncologists from ten departments (ROs), four haematologic oncologists and chest physicians from four departments (HOs) and five radiation oncologists from one department (RO1D). Results: Agreement (overlap P 70% of the contoured pixels) for the GTV and PTV delineation was found in 16.3% and 23.7% (ROs), 30.4% and 38.6% (HOs) and 32.8% and 35.9% (RO1D), respectively. Conclusion: A large interobserver variability in the PTV and much more in the GTV delineation were observed in spite of a detailed instruction for delineation. The variability was smallest for group ROID where due to repeated discussions and uniform teaching a better agreement was achieved. Ó 2009 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 91 (2009) 455–460
Patients with unresectable or medically inoperable non-smallcell lung cancer (NSCLC) are treated with primary radiochemotherapy [1,2]. Although this is a curative approach, these patients still have a poor prognosis with a curative potential of 10–20% for stage IIIA/B. Published literature suggests that tumor control is improved with dose escalation using conformal radiation therapy [3,4].
* Corresponding author. Address: Department of Radiotherapy and Radiooncology, University Hospital Göttingen, Robert-Koch-Straße 40, 37099 Göttingen, Germany. Tel.: +49 551 398866; fax: +49 551 396192. E-mail address: [email protected] (H. Vorwerk). 0167-8140/$ - see front matter Ó 2009 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.radonc.2009.03.014
Whether adjuvant chemotherapy is able to improve survival after primary radiochemotherapy for inoperable NSCLC is still unclear. To answer this question, the GILT-CRT-1-study (multicenter, open-label, randomised, phase III study of oral Vinorelbine in combination with Cisplatin concurrently with radiotherapy to a total dose of 66 Gy) randomises to either two more cycles of consolidation therapy with oral Vinorelbine and Cisplatin plus Best Supportive Care or Best Supportive Care alone in these patients with unresectable locally advanced NSCLC. The comparability between different participants of multicenter studies is compromised by contouring errors which may cause inaccurate radiation treatments
456
Interobserver variability for lung cancer
that result from different RT volumes. Contouring variations are caused by observer-, image- and instruction-related reasons. Within the GILT-study, we wanted to evaluate the agreement of target volume delineation by participants of the study. The target volume definition for conformal radiotherapy may be the most important factor for adequate tumor coverage and sufficient sparing of critical structures [5,6]. The aim of our analysis was to examine, whether we could minimise the variability in GTV and PTV delineation for primary nonsmall-cell lung cancer patients by using a strict presetting for GTV definition and for the determination of the PTV (observer-related reason) with simultaneous exclusion of image- and instruction-related reasons for testing quality assurance. Such a precise delineation protocol to homogenize delineation was demanded by different authors [5,7]. We analysed different groups of physicians: radiation oncologist from different departments, haematologic oncologists and chest physicians from different departments and radiation oncologist from one department.
Methods and materials Delineation was performed during a workshop of the participants of the GILT-CRT-1-study. The physicians, who agreed to participate in the workshop, could obtain all patient informations digitally in the forefront of the meeting. At the beginning of the meeting all participants received a folder with all patient data to make themselves familiar with the facts. Contouring was done in the planning rooms of one department under survey of two physicists. Interaction with any physician during contouring was not allowed. Eighteen physicians were asked to contour the GTV and PTV for three patients with non-small-cell lung cancer stage IIIB scheduled for primary radiation therapy. Contouring was done by experienced physicians from fourteen different departments, divided into three groups: RO ten senior radiation oncologists (G.B., M.B., B.D., R.F., U.H., M.v.K., B.M., T.M., R.R., H.V.) from ten different departments used to 3D planning. HO four senior haematologic oncologists/and chest physicians (M.D., W.K., M.S., M.M.) from four different departments (further noticed as haematologic oncologists). RO1D four senior and one experienced junior radiation oncologists (R.M.H., M.H., M.N., B.S., H.V.) from one department, all familiar with 3D planning. Patient data were acquired from the first three men, who were treated in the GILT-CRT-1-study (Pierre Fabre Study Code: PM 0259 CA 304 J1; EudraCT number: 2004-005135-26) in the University of Göttingen, Germany. Before contouring every physician was provided with written information on every patient, including history, clinical examination, general performance status, bronchoscopic findings, lung function tests and computed tomography (CT) scans with iodine contrast enhancement. The CT-scans were also available in digital form. All data were made anonymous. The median age of the patients was 60 years. All patients were planned for curative radiochemotherapy in combination with Cisplatin and Vinorelbine and a total radiation dose of 66.0 Gy (5 2.0 Gy per week) with reduced field according to the study protocol. The field was reduced after 50.0 Gy utilizing a new restaging CT according to the study protocol. For our analysis we used the first CT before beginning of the treatment and analysed the contouring planned for the first part of the treatment up to 50.0 Gy.
The stage and location of the tumors were as follows: 1. cT3 cN3 M0 (left central, partial atelectasis of the lower left lobe, central and upper mediastinal lymph nodes), 2. cT3 cN3 M0 (right central, central mediastinal lymph nodes), 3. cT2 cN3 M0 (right central, central mediastinal lymph nodes). Continuous 5 mm CT-scans (10 mm in one patient) of the treatment region in the thorax were obtained with a single-slice spiral CT scanner (Somatom Balance, Siemens Medical Systems, Forchheim, Germany) for all three patients without contrast enhancement. We used a resolution of 0.94 mm in axial plane, a couch shift of 7.5 mm and a field-of-view of 480 mm (130 kV, 100 mAs). Physicians entered their contours directly at the treatmentplanning console (Eclipse, Version 6.5, Varian Medical Systems, Palo Alto, CA, USA). Radiation physicians not familiar with this planning system and haematologic oncologists beforehand underwent introductions in the management of the contouring workspace by a physicist. Particularly, it was demonstrated how to change the window level between soft tissue and lung window and how to measure distances. Sagittal and coronar CT reconstruction were available to allow orientation in craniocaudal direction. All physicians had permanent technical assistance during contouring by two physicists (D.W., T.G.). All physicians worked on their own CT image-set. Contouring was performed independently without information of other participants. Two physicians from the group ROs contoured only the CT from one patient. All physicians obtained a detailed instruction, which is provided by the GILT-CRT-1-study protocol, for definition of the gross target volume (GTV) and planning target volume (PTV) as follows: Gross tumor volume: 1. extension of the primary tumor in the soft tissue and lung window, 2. lymph nodes larger than 1.5 cm in cross diameter in the transversal CT-scans, 3. lymph nodes with central hypodensity, 4. lymph nodes larger than 1.0 cm in cross diameter in the transversal CT-scans, if multiple nodes are seen in the same compartment (defined by Naruke et al. [8]). It was possible to contour different structures in one volume (for example the GTV). The limits for the different window levels could be chosen by the participants. The structures could be contoured by a paint brush or pixel-by-pixel. Every physician received a detailed anatomic sagittal picture of the lymph node compartments according to the Naruke Classification [8] and described in more detail by Kirikuta et al. [9]. For easier distinction of the compartments we enhanced a transversal CTmap created by Kirikuta et al. [9] with coloured flags for all lymph node compartments. Planning target volume: 1. PTV is defined as GTV plus a safety margin of 1.0 cm in lateral direction, 1.0 cm in ventrodorsal direction and 1.5 cm in craniocaudal direction via automatic expansion (This considers both oncological (subclinical extension [CTV]) and geometrical (setup and motion uncertainty) factors). 2. If infracarinal involvement is suspected, caudal rim of the PTV should be placed at least 3 cm below the carina. It was possible to modify the PTV contour after automatic expansion of the GTV. We performed a volume analysis, an anatomic analysis and a pixel-by-pixel analysis of all contoured parts of all CT-slices similar to that explained in more detail below for different groups of
H. Vorwerk et al. / Radiotherapy and Oncology 91 (2009) 455–460
physicians. In the first part a comparison of ten radiation oncologists (ROs) from ten different departments with four haematologic oncologists (HOs) from four different departments were performed. In the second part the ten radiation oncologists from the first part were compared with five radiation oncologists from one department (RO1D). 3D-volumes of the GTV and PTV were calculated by use of the planning system EclipseÒ. The medians, standard deviations and ratio of maximum to minimum volume (Vmax/Vmin) were calculated for every patient and physician. Additionally, we assessed the deviation from the mean volume, calculated from the fourteen different departments, i.e. from RO’s and HO’s. We analysed which anatomic sites of the lymph node compartments were considered to be involved. A compartment was considered as contoured by a participant when at least a part of the compartment was contoured. A checklist was created by us which contained the following items: upper and lower paratracheal lymph nodes left and right, preaortic lymph nodes, subaortic lymph nodes (A-P window), hilar lymph nodes left and right, subcarinal lymph nodes, bronchopulmonal lymph nodes left and right, paraesophageal lymph nodes (below carina). This checklist was completed for every outlined GTV and analysed for agreement between the physicians. Agreement was stated, when 70% or more of the observers included or excluded the lymph node compartment in their contour of the structure. In our clinic we developed a mapping tool in order to analyse the concordance between the observers pixel-by-pixel. All contoured structures were exported from EclipseÒ with Dicom format, were converted into ASCII format and imported in MicrosoftÒ Office Excel 2003. The structures were specified with different points on the circumference of the structure. We composed a graticule with one cubic centimetre equivalent to 100 pixel points. Every pixel point of the contoured circumference from the structure was plotted as a ‘‘1” into one table map of ExcelÒ per physician, patient and z-axis of the CT-scans. We developed a source code to provide a linear regression between the pixel points on the circumference from the contoured structure, like the planning system EclipseÒ, and filled the area between the lines with the number
Fig. 1. Picture of a map from the pixel-by-pixel analysis of the contoured structures by 13 physicians. All structures per physician were displayed by the number 1 for all pixels in the contoured area and afterwards added for each pixel point. For better visualization we coloured the different areas of numbers (The darker the colour the higher the number). Hundred pixel points are equivalent to one cubic centimetre.
457
‘‘1” using an in house source code. Along this way we received a map of the structures in all transversal CT-scans of all patients and physicians, pixel-by-pixel. Afterwards we added up all maps of each group, who should be investigated, and achieved a new map with numbers equivalent to the number of physician, who had contoured this pixel (Fig. 1). This analysis detected differences in the 2D-transversal CT-slices for all slices. On this way a statement for the 3D-volume was possible. Therewith we calculated the agreement between the physician groups. Agreement was stated, if 70% or more of the observers included the pixels on the transversal CT slice in their contour of the structure. To identify the physicians’ level of agreement, encompassing and common volumes were calculated for every patient. The encompassing volumes are the smallest volumes integrating all entered volumes of one patient. The common volumes are the largest volumes that are part of all GTVs or PTVs of one patient [10–12].
Results Firstly, we analysed the delineation of the fourteen different clinics. Especially, we compared the ROs against the HOs. Secondly, we analysed the differences between the ROs from ten different departments and the RO1D from one department. The mean volume and the standard deviation of the GTV and PTV volumes contoured by the haematologic oncologists (HO’s) were significantly smaller than the contours of the radiation oncologists (RO’s) (Table 1) with 50–68% of the HO volumes relative to the RO volumes. The mean volumes of both the groups of radiation oncologists (ROs and RO1D) are comparable for the GTV and PTV with a much smaller standard deviation by the RO1D. The volumes delineated by the RO1D ranged from 83% to 90% of the volumes delineated by the ROs. The deviation from the mean GTV of all patients calculated for all fourteen different physicians (ROs and HOs), were largest considering the ROs (30.5 ± 86.2 cm3) and comparable considering the HOs ( 34.0 ± 75.2 cm3) and RO1D ( 1.9 ± 39.0 cm3). The differences between the three groups get larger regarding the contoured PTV (ROs: 46.4 ± 126.4 cm3, HOs: 30.3 ± 171.88 cm3, RO1D: 0.7 ± 89.9 cm3). For ROs and RO1D, the deviation of the PTV was similar to the deviation of the GTV, but the standard deviation was larger. The standard deviation of the PTV contours is larger than that of the GTV contour, especially for the HOs. We found a good agreement (P70% of the observers were conform) in 81% (92%, 92%) of the investigated lymph node compartments for the ROs (HOs, RO1D). On average the ROs and RO1D included 6.3 of all 12 lymph node compartments as defined in methods and materials in their contours in contrast to only 4.4 of the 12 lymph node compartments by the HOs. We compared the agreement of the contoured parts of all CTslices for all patients and physicians. We divided all slices of the CT in small parts of 0.001 cm3 and compared all of them with one another. The accurate analysis of this pixel-by-pixel mapping procedure leads to less agreement than the volume and anatomic analysis. Only 70% of the ROs contoured 16.3% of the encompassing GTV volume (smallest contoured part of the slices integrating all entered contours), 70% of the HOs and RO1D contoured 30.4% and 32.8% of the encompassing GTV, respectively (Table 2). The analysis of the fourteen different departments showed even fewer agreements (12.4%). The agreement between the physicians is better concerning the encompassing PTV volume. 70% of the ROs contoured 23.7% of the encompassing PTV volume in comparison to 38.6% (35.9%) by the HOs (RO1D). The common GTV volumes were very small for the 14 departments but also for the ROs and HOs (6.2–14.9 cm3) and not much larger for the RO1D (36.5 cm3). The smallest encompassing GTV
458
Interobserver variability for lung cancer
Table 1 Magnitude of volumes for each observer group. Mean volume over all physicians (cm3) GTV
14 Departments ROs HOs RO1D
PTV
Patient 1
Patient 2
Patient 3
Patient 1
Patient 2
Patient 3
137.5 ± 98.8 162.6 ± 105.6 81.1 ± 45.1 132.8 ± 50.8
157.1 ± 84.5 183.7 ± 86.9 97.5 ± 33.1 151.6 ± .47.8
179.6 ± 110.5 213.1 ± 116.3 122.5 ± 52.8 180.9 ± 38.6
458.0 ± 219.1 531.8 ± 206.4 291.8 ± 142.7 438.2 ± 104.8
400.1 ± 118.1 450.0 ± 94.1 287.7 ± 84.7 393.8 ± 125.2
555.8 ± 165.8 590.1 ± 183.6 487.2 ± 89.7 582.2 ± 85.5
ROs, radiation oncologists from 10 departments; HOs, haematologic oncologist/s pneumologists; RO1D, radiation oncologists from one department; GTV, gross target volume; PTV, planning target volume.
Table 2 Agreement of the contoured parts of the transversal CT-slices of the physicians with a pixel-by-pixel analysis. Agreement in 70% of contoured pixels (%) GTV 14 1.2 10 ROs 4 HOs 5 RO1D PTV 14 3.3 10 ROs 4 HOs 5 RO1D
Departments
16.3 ± 4.1 30.4 ± 8.0 32.8 ± 3.7 Departments
23.7 ± 3.9 38.6 ± 6.1 35.9 ± 2.2
Common volume (cm3) 12.4 ± 4.3
Encompassing volume (cm3)
Common/ encompassing volume (%)
6.2
507.5
14.9 9.5 36.5
422.9 227.4 225.0
3.3 4.0 15.8
19.3 ± 4.3
52.9
1446.2
1124.0 402.7 643.5
7.5 15.1 17.4
83.1 70.8 121.1
Agreement as stated, when 70% of the physicians had contoured the part of the transversal CT slice. The encompassing volumes are the smallest volumes integrating all entered volumes of one patient. The common volumes are the largest volumes that are part of all GTVs or PTVs of one patient. The described volumes are the sum of all three patients. ROs, radiation oncologists from 10 departments; HOs, haematologic oncologist/s pneumologists; RO1D, radiation oncologists from one department; GTV, gross target volume; PTV, planning target volume.
volumes could be found contoured by the RO1D and the HOs, whereas the encompassing volume of the 14 departments showed more than twice as much cubic centimetre. The relation between the common and encompassing volumes was less than 5% except for the RO1D with 15.8%. The observations for the common and encompassing GTV volumes are nearly the same for the PTV volumes with altogether larger volume. But the relation between the common and encompassing volumes is larger for all groups of physicians, especially for the HOs. Discussion Within the GILT-CRT-1-study (multicenter study for patients with unresectable locally advanced NSCLC), we wanted to evaluate the agreement of target volume delineation by participants of this study. The target volume definition for conformal radiotherapy may be the most important factor for adequate tumor coverage and sufficient sparing of critical structures [5]. Within a study protocol every effort should be taken to ensure that the contours are standardized and accurate with high interobserver reproducibility. Contouring variations result in systematic errors of radiation treatment that are a confounding factor for the interpretation of treatment results. Besides, the comparability between different departments, who take part in multicenter studies, will be very difficult. More accurate delineation of the GTV and PTV may result in
better conformal target coverage, improved sparing of normal tissue, and ultimately enhanced local control. Several reports on the variability of GTV and CTV delineation have been published for a variety of malignancies, such as tumors of the lung, breast, cervix, bladder, pancreas and prostate and of the head and neck region [5,10,13–18]. The principle reasons for interobserver variability’s can be image-, instruction- and observer-related. For our analysis, we wanted to analyse only observer-related reasons for interobserver variability. One major factor for interobserver variability is the different assessments regarding the risk of spreading of tumor [5,7]. The purpose of our analysis was the difference of interpretations whether a lymph node was involved or not and the extension of tumor size. The first patient presented an atelectasis of the left lower lobe, which could increase the variations. Other reasons are training, differences in the attention to detail in the contouring process itself and subjective interpretation of image knowledge of topographical cross-sectional anatomy. The observed wide range in volume size for the GTV (22.7– 486.9 cm3) and PTV (119.0–927.1 cm3) is comparable to the results observed for lung cancer [5]. The mean volume of tumors, suffer from rather large neoplasms, may cause a problem regarding interobserver variation [5]. The maximum to minimum volume (Vmax/ Vmin) from the GTV is larger for the fourteen departments (10.1) and comparable for the different physician groups (2.3–7.9) then described in the published literature. Van de Steene et al. found a ratio between 1.1 and 7.6 for the Vmax/Vmin of the GTV in patients with lung tumor. For some other tumors a quite good agreement on target volumes was observed, especially for the prostate [10,13]. Obviously, tumors that have obvious borders like the prostate can be outlined more consistently than tumors without clear border to the neighbouring organs. The Vmax/Vmin of the GTV (7.9) delineated by the ROs is much larger than that of the PTV (2.4). The rigid instruction for delineation gave the order to enlarge the GTV volume (possible automatically) with a default offset, so that we could assume, that the Vmax/ Vmin of the GTV and PTV could be of the same order, similar to that observed in the groups of the HOs and RO1D. The ROs have varied their PTV after automatic enlargement, especially in lateral directions, although this was not part of the study protocol. Thereby the large differences in the GTV delineation were reduced for the PTV. This is reflected in the lesser differences between the ROs and the HOs/RO1D in the PTV contours, whereby the clinical impact of the differences in GTV delineation will be much reduced. The accurate definition of the GTV can be assed by pathologycorrelated imaging, if patients with small tumors underwent surgery after radiotherapy. Thereby most GTV definitions are too larger, but can also be too small [19]. We recommend to focus not only on the GTV delineation but also more on the PTV delineation, because subsequent steps use the initial GTV data. Dosimetric and field planning decisions are based primary on the PTV and secondarily on the GTV.
H. Vorwerk et al. / Radiotherapy and Oncology 91 (2009) 455–460
The definition of lymph nodes classified as pathologic was exactly defined with P1.5 cm or many lymph nodes with P1.0 cm in a lymph node compartment (definition by Naruke et al. [8] and Kirikuta et al. [9]). Every participant received a detailed CTmap of the lymph node compartments. Particularly it was demonstrated to all physicians how to measure diameters of the lymph nodes. This led to a good agreement in 100% and P80% of all analysed lymph node regions in 42% and 75% of all participants, respectively. We found only disagreement in the paratracheal medial left paraesophageal and praeaortal lymph node regions. Another error was due to confusion between lymph node, esophagus and vessels. Other authors found 63% of involved lymph node regions were delineated correctly by the clinicians and 22% of the drawn lymph node regions were accepted to be false positive [5]. This correlates exactly with 75% agreement of all physicians in our analysis. So that we must state, that strict measuring of lesions with a rigid instruction for delineation could not notably reduce interobserver variations in lymph node judgement in our analysis. The pixel-by-pixel analysis compares every part of the contours against the contours of the other physicians instead of only mathematical values and anatomical sites. This reflects the exact differences between the participants. The differences in the GTV contours of the ROs and HOs (16.3% and 30.4%) were respectable in the analysis group by group, but less compared with each other groups (12.4%), because the HOs delineated smaller parts of the CTslices than the ROs but are more consistent among each other (Table 2). The relation between these two groups was reflected in the PTV delineation with better agreement within the groups. The agreement between the RO1D is higher than that between the ROs even in the GTV (32.8%) but also in the PTV contours (35.9%) and comparable to the HOs. This was also reflected in the relation between the common and encompassing volumes. Our analysis showed large discrepancies regarding the delineation of the lymph node regions by the HOs. This may be caused by less knowledge regarding the large mediastinal vessels and the position of esophagus in CT-scans by the HOs in comparison to the ROs. On the other hand, we found small discrepancies between the HOs regarding the delineation of the tumor. The HOs had another point of view than the ROs regarding the tumor because they are used to define the tumor expansion based on combination of bronchoscopy and CT. This could be the reason for this good agreement. The contoured volumes by the HOs are significantly smaller than contoured by the ROs and RO1D. This could be a consequence of the RO’s fear of missing a part of the tumor tissue and therefore design larger contours, whereas the HOs are more afraid of making a false positive error. The agreement in the group of RO1D regarding the delineated lymph node regions was superior with 92% agreement compared to the other groups and published data with 75%. We found the same good agreement in the GTV and also the PTV contoured by the RO1D in all analysis in contrast to the ROs. This implicates that the RO1D used the automatic expansion software of the contouring program and did not correct the PTV free-hand correction afterwards. Remarkably we found the same agreement in the PTV delineation after free-handly correction by the ROs and only automatically enlarged GTV by the RO1D. We suggest, that we could detect better agreement because the oncologists from one department are used to compare and discuss their target volumes daily for better accordance concerning the delineation of the tumor as well as the lymph node regions. The interobserver differences would not be relevantly improved by use of a restrictive instruction, but intra-institutional agreement can diminish differences in target volume delineation. The interobserver variabilities vary by use of different imaging techniques. PET-CT is superior to CT in delineation of the tumor and the lymph nodes [3,20–22], but PET was not requested in
459
GILT-study. By use of PET the variability between the participants would probably be smaller, especially in regions of atelectasis [23]. Delineation accuracy can be improved by using CT window level information. The CT lung windows correlate better with the pathologic size of lung adenocarcinoma than do the mediastinal windows [24]. Grills et al. described an overestimation of the gross pathologic size by a mean of 5.8 mm and an underestimation of the gross tumor size plus microscopic extension by a mean of 1.2 mm by use of the mediastinal window. All participants were instructed to delineate the tumor expansion in the lung and mediastinal window and pathologic lymph nodes – not the whole lymph node region – to deal with this problem. Nevertheless this led to no diminishment of the variation. Difficulties in discrimination between tumor and atelectasis are widely known and a major factor in interobserver variations [12]. In our analysis we could not detect higher discrepancies of the contours of patient 1 (with atelectasis) in comparison to patient 2 or 3. Possibly a microinvasive insertion of radio-opaque markers may be helpful in diminishing interobserver variabilities [25]. Conclusions No ‘‘golden standard” in GTV or PTV delineation can be defined so far. We will not till then be able to make appraisals of the clinical impact, if all physicians define the PTV in the same way with a much better agreement, than is currently detected. A good training of radiation oncologists, a good collaboration between radiation oncologists and radiologist, dummy runs for multicenter studies, establishment of central reviewing boards or detailed instructions for delineation (tested in this analysis) are recommended. However, up to now these procedures have not been shown to decrease the interobserver variability substantially [6,12,17,18,26,27]. Nevertheless, we could not demonstrate an improvement in agreement between the physicians by use of a detailed instruction for delineation compared to existing literature with no detailed instructions, but we could demonstrate, that physicians from one department agree more than physicians from different departments. We recommend online available examples and the possibility of comparison of the own delineation in order to control oneself periodically for better agreement in contouring of target volumes. A ‘‘golden standard” for target delineation should be developed taking relapse patterns into account. References [1] Jeremic B, Machtay M. Concurrent radiochemotherapy in the treatment of locally advanced non-small cell lung cancer. Hematol Oncol Clin North Am 2004;18:91–100. [2] Curran Jr WJ. Therapy for stage III non-small-cell lung cancer: Bob Beamon or Chemo-Beamo? Clin Lung Cancer 2006;8:84. [3] Hong R, Halama J, Bova D, Sethi A, Emami B. Correlation of PET standard uptake value and CT window-level thresholds for target delineation in CT-based radiation treatment planning. Int J Radiat Oncol Biol Phys 2007;67:720–6. [4] Martel MK, Ten Haken RK, Hazuka MB, et al. Estimation of tumor control probability model parameters from 3-D dose distributions of non-small cell lung cancer patients. Lung Cancer 1999;24:31–7. [5] Van de Steene J, Linthout N, de Mey J, et al. Definition of gross tumor volume in lung cancer: inter-observer variability. Radiother Oncol 2002;62:37–49. [6] Weiss E, Hess CF. The impact of gross tumor volume (GTV) and clinical target volume (CTV) definition on the total accuracy in radiotherapy theoretical aspects and practical experiences. Strahlenther Onkol 2003;179:21–30. [7] Giraud P, Elles S, Helfre S, et al. Conformal radiotherapy for lung cancer: different delineation of the gross tumor volume (GTV) by radiologists and radiation oncologists. Radiother Oncol 2002;62:27–36. [8] Naruke T, Suemasu K, Ishikawa S. Lymph node mapping and curability at various levels of metastasis in resected lung cancer. J Thorac Cardiovasc Surg 1978;76:832–9. [9] Kiricuta IC. Selection and delineation of lymph node target volume for lung cancer conformal radiotherapy. Proposal for standardizing terminology based on surgical experience. Strahlenther Onkol 2001;177:410–23. [10] Rasch C, Barillot I, Remeijer P, et al. Definition of the prostate in CT and MRI: a multi-observer study. Int J Radiat Oncol Biol Phys 1999;43:57–66.
460
Interobserver variability for lung cancer
[11] Weltens C, Menten J, Feron M, et al. Interobserver variations in gross tumor volume delineation of brain tumors on computed tomography and impact of magnetic resonance imaging. Radiother Oncol 2001;60:49–59. [12] Valley JF, Bernier J, Tercier PA, et al. Quality assurance of the EORTC radiotherapy trial 22931 for head and neck carcinomas: the dummy run. Radiother Oncol 1998;47:37–44. [13] Logue JP, Sharrock CL, Cowan RA, et al. Clinical variability of target volume description in conformal radiotherapy planning. Int J Radiat Oncol Biol Phys 1998;41:929–31. [14] Petersen RP, Truong PT, Kader HA, et al. Target volume delineation for partial breast radiotherapy planning: clinical characteristics associated with low interobserver concordance. Int J Radiat Oncol Biol Phys 2007;69:41–8. [15] Smith WL, Lewis C, Bauman G, et al. Prostate volume contouring: a 3D analysis of segmentation using 3DTRUS, CT, and MR. Int J Radiat Oncol Biol Phys 2007;67:1238–47. [16] Weiss E, Richter S, Krauss T, et al. Conformal radiotherapy planning of cervix carcinoma: differences in the delineation of the clinical target volume. A comparison between gynaecologic and radiation oncologists. Radiother Oncol 2003;67:87–95. [17] Yamazaki H, Nishiyama K, Tanaka E, et al. Dummy run for a phase II multi-institute trial of chemoradiotherapy for unresectable pancreatic cancer: inter-observer variance in contour delineation. Anticancer Res 2007;27: 2965–71. [18] Coles CE, Hoole AC, Harden SV, et al. Quantitative assessment of inter-clinician variability of target volume delineation for medulloblastoma: quality assurance for the SIOP PNET 4 trial protocol. Radiother Oncol 2003;69: 189–94.
[19] Stroom J, Blaauwgeers H, van Baardwijk A, et al. Feasibility of pathologycorrelated lung imaging for accurate target definition of lung tumors. Int J Radiat Oncol Biol Phys 2007;69:267–75. [20] van Baardwijk A, Bosmans G, Boersma L, et al. PET-CT-based auto-contouring in non-small-cell lung cancer correlates with pathology and reduces interobserver variability in the delineation of the primary tumor and involved nodal volumes. Int J Radiat Oncol Biol Phys 2007;68:771–8. [21] Fox JL, Rengan R, O’Meara W, et al. Does registration of PET and planning CT images decrease interobserver and intraobserver variation in delineating tumor volumes for non-small-cell lung cancer? Int J Radiat Oncol Biol Phys 2005;62:70–5. [22] Jeong SY, Lee KS, Shin KM, et al. Efficacy of PET/CT in the characterization of solid or partly solid solitary pulmonary nodules. Lung Cancer 2008;61:186–94. [23] Fitton I, Steenbakkers RJ, Gilhuijs K, et al. Impact of anatomical location on value of CT-PET co-registration for delineation of lung tumors. Int J Radiat Oncol Biol Phys 2008;70:1403–7. [24] Grills IS, Fitch DL, Goldstein NS, et al. Clinicopathologic analysis of microscopic extension in lung adenocarcinoma: defining clinical target volume for radiotherapy. Int J Radiat Oncol Biol Phys 2007;69:334–41. [25] Kupelian PA, Forbes A, Willoughby TR, et al. Implantation and stability of metallic fiducials within pulmonary lesions. Int J Radiat Oncol Biol Phys 2007;69:777–85. [26] Seddon B, Bidmead M, Wilson J, Khoo V, Dearnaley D. Target volume definition in conformal radiotherapy for prostate cancer: quality assurance in the MRC RT-01 trial. Radiother Oncol 2000;56:73–83. [27] Roy AE, Wells P. Volume definition in radiotherapy planning for lung cancer: how the radiologist can help. Cancer Imaging 2006;6:116–23.