- Email: [email protected]
CT fusion
Int. J. Radiation Oncology Biol. Phys., Vol. 65, No. 3, pp. 726 –732, 2006 Copyright © 2006 Elsevier Inc. Printed in the USA. All rights reserved 0360-3016/06/$–see front matter
doi:10.1016/j.ijrobp.2006.01.014
CLINICAL INVESTIGATION
Head and Neck
VARIABILITY OF GROSS TUMOR VOLUME DELINEATION IN HEAD-AND-NECK CANCER USING CT AND PET/CT FUSION ADAM C. RIEGEL, B.A.,* ANTHONY M. BERSON, M.D.,*† SYLVIE DESTIAN, M.D.,†‡ TRACY NG, M.D.,*† LAWRENCE B. TENA, M.D.,*† ROBIN J. MITNICK, M.D.,†‡ † AND PING S. WONG, PH.D.* *Department of Radiation Oncology, St. Vincent’s Comprehensive Cancer Center, New York, NY; †New York Medical College, Westchester, NY; and ‡Department of Radiology, St. Vincent’s Hospital, New York, NY Purpose: To assess the need for gross tumor volume (GTV) delineation protocols in head-and-neck cancer (HNC) treatment planning by use of positron emission tomography (PET)/computed tomography (CT) fusion imaging. Assessment will consist of interobserver and intermodality variation analysis. Methods and Materials: Sixteen HNC patients were accrued for the study. Four physicians (2 neuroradiologists and 2 radiation oncologists) contoured GTV on 16 patients. Physicians were asked to contour GTV on the basis of the CT alone, and then on PET/CT fusion. Statistical analysis included analysis of variance for interobserver variability and Student’s paired sample t test for intermodality and interdisciplinary variability. A Boolean pairwise analysis was included to measure degree of overlap. Results: Near-significant variation occurred across physicians’ CT volumes (p ⴝ 0.09) and significant variation occurred across physicians’ PET/CT volumes (p ⴝ 0.0002). The Boolean comparison correlates with statistical findings. One radiation oncologist’s PET/CT fusion volumes were significantly larger than his CT volumes (p < 0.01). Conversely, the other radiation oncologist’s CT volumes tended to be larger than his fusion volumes (p ⴝ 0.06). No significant interdisciplinary variation was seen. Significant disagreement occurred between radiation oncologists. Conclusion: Significant differences in GTV delineation were found between multiple observers contouring on PET/CT fusion. The need for delineation protocol has been confirmed. © 2006 Elsevier Inc. GTV delineation, PET/CT fusion, Head-and-neck cancer, Interobserver variation.
INTRODUCTION
small amount can result in a severely compromised plan, either overdosing of surrounding sensitive tissue, dangerously underdosing of cancerous tissue, or both (6 –10). One major source of inaccuracy or, technically, imprecision in GTV delineation is interobserver variability (11–14). Numerous studies have shown that target volumes drawn by multiple observers on conventional computed tomography (CT) vary significantly in areas such as the prostate and seminal vesicles (15, 16), brain (12, 14, 17), cervix (11), lung (7, 18 –21), bladder (5), cervical esophagus (9), and head and neck (6, 10, 13). Studies in the literature make various recommendations to reduce this variation, including more uncertainty analysis (7, 14), larger margins (15), multidisciplinary team contouring (1, 11, 17, 19, 20), and stricter delineation guidelines (1, 7, 11, 17, 20, 22). Two studies recommend the evaluation of multimodality imaging in target delineation (11, 20), which is the focus of this study.
The ever-increasing rate of technological innovation provides physicians around the world with access to cuttingedge equipment and a variety of treatment techniques, which ultimately allows them to individualize treatments to a greater degree than ever before. Intensity-modulated radiation therapy (IMRT), for example, assures high doses of radiation to the target volume and low doses to the surrounding sensitive tissues by creating a nonuniform dose distribution with the multileaf collimator, which is particularly useful in head-and-neck cancer (HNC), where normal and abnormal structures are in close proximity (1– 4). The treatment’s degree of success, however, begins with the treatment-planning process. Accurate delineation of the gross tumor volume (GTV) is critical in IMRT treatment planning because of the high dose gradient inherent to the technology (1, 5–7). Inaccuracy in the delineation of target volumes by a Reprint requests to: Anthony M. Berson, M.D., Radiation Oncology Department, St. Vincent’s Comprehensive Cancer Center, 325 W. 15th St., New York, NY 10011, Tel: (212) 367-1795; Fax: (212) 367-1742; E-mail: [email protected] The abstract of this study was presented at the Forty-Seventh
Annual Meeting of the American Society for Therapeutic Radiology and Oncology (ASTRO), October 19 –23, 2005, Denver, CO. Received June 8, 2005, and in revised form Jan 9, 2006. Accepted for publication Jan 10, 2006. 726
Variability of gross tumor volume delineation
Positron emission tomography that utilizes the radiotracer fluorodeoxyglucose-positron emission tomography (FDGPET) is another recent technological development now used in treatment planning to increase the accuracy of GTV delineation. PET images are coregistered with conventional CT images to create a “fused” PET/CT image, a 3D representation of a patient’s anatomic structure superimposed with metabolic function. Presently, no standard criteria exist for multimodality GTV delineation of HNC patients at our institution. The purpose of this study was to assess the need for institutional protocol by examination of the uniformity of GTV delineation across multiple observers by use of FDG-PET/CT fusion imaging. The investigation consisted of 2 comparisons: (1) evaluation of GTV-delineation variability between multiple physicians on CT and PET/CT fusion (including interdisciplinary variability), and (2) analysis of deviation between contours drawn on CT and PET/CT fusion. Any observed variability between physicians is justification for the development of standardized institutional protocol regarding the use of multimodality imaging in treatment planning. The basis for this protocol will be shaped by the results of this study. 18
METHODS AND MATERIALS Study participants Four physicians, labeled A, B, C, and D for the purposes of this study, volunteered to contour volumes on the 16 HNC patients. Physicians A and B were neuroradiologists with 17 and 15 years experience, respectively. Physicians C and D were radiation oncologists with 4.5 and 13 years experience, respectively. The following criteria were used to retrospectively select patients for this study: (1) The patient must have cancer in the head or neck. (2) Gross tumor must be present. (3) The patient must have received a PET/CT simulation. (4) The patient cannot have received any surgical or radiation treatment to the tumor in question before the PET/CT scan. Sixteen (16) patients, 13 men and 3 women, underwent PET/CT simulations between April 2004 and February 2005 and were enrolled in the study (Table 1). Mean patient age was 57 years (range, 41– 98 years). Tumor location varied from 5 in the nasopharynx, 4 in the base of tongue, to 1 in the tonsil, larynx, glottis, cervical esophagus, eye orbit, optic nerve, nasal vestibule, and maxillary sinus. Thirteen patients had squamous cell carcinoma, and 3 patients had non-Hodgkin’s lymphoma, B-cell lymphoma, and melanoma in the maxillary sinus, orbit, and optic nerve, respectively.
Scanning and contouring All 16 patients received an 18FDG PET/CT scan from a Discovery ST combined PET/CT scanner (GE Medical Systems, Milwaukee, WI). Each patient was injected with 15 to 20 mCi of 18 FDG and scanned on a flat scanning table an average of 2 hours after injection. Patients were immobilized with a Med-Tec MTAPS thermoplastic insert mask (Med-Tec, Orange City, IA) and a VacLok MT-VL-TYS-01 custom head rest. All CT scans were obtained in helical mode with a gantry rotation period of 0.8 sec and table pitch of 1.35 cm to 1 rotation. The number of slices was
● A. C. RIEGEL et al.
727
72 slices and 179 slices for 1 patient each, 91 slices for 5 patients, and 135 slices for 9 patients. Head-and-neck PET acquisition usually requires 2 bed positions, and scan duration is approximately 4 to 5 min per bed position. Images were reconstructed by use of OSEM iterative reconstruction, with post and loop filters at 6.71 and 4.30 full-width half-maximum, respectively. Images were managed through GE Xeleris software (GE Healthcare, Waukesha, WI). CT pixel size was 0.98 ⫻ 0.98 mm2 and PET pixel size was 4.30 ⫻ 4.30 mm2. The field of view was 55 cm in diameter, and slice thickness was 3.75 mm. Computed tomography and PET images were transferred via DICOM protocol to ECLIPSE (Varian, Palo Alto, CA) treatmentplanning workstations. Images were fused by use of DICOM coordinate registration and checked visually for accuracy. For each patient, separate treatment-planning profiles were created for the purpose of the study. Names were shielded to minimize bias. The primary investigator (PI) facilitated each contouring session. Each physician was asked questions regarding their familiarity with the treatment-planning system, and a tutorial customized to each physician’s skill level was given. Physicians were given the patient’s diagnosis (general tumor location and biopsy results), symptoms, brief clinical history, and physical examination. The PI instructed each physician to contour only gross tumor on the axial CT images. Lymph nodes were not included in the definition of GTV for simplification of the volumetric analysis. The CT window levels were discretionary. After the physician verbally acknowledged completion, the volume was saved and the PI instructed the physician to contour gross tumor only, again no lymph node involvement, on the PET/CT fused images. The standardized uptake-value (SUV) window was discretionary, but the default window was arbitrarily set at a lower limit of 3 and an upper limit of 4, as SUV ⫽ 3 has previously been used as a threshold for malignancy (23, 24). Again, after verbal acknowledgment of completion, the volume was saved. Experimental GTVs contoured for the purposes of the study were not explicitly used for treatment planning.
Analysis The analysis consisted of 3 categories of volumetric comparison: interobserver variation (across physicians within a modality), intermodality variation (between CT and PET/CT fusion for each physician and on average), and interdisciplinary variation (across disciplines within a modality and across modalities within a discipline). All 3 categories were analyzed with a mean-significance test (category-dependent) or overlap analysis. The significance threshold was set at ␣ ⫽ 0.05. “Near” significance or “tendency” was noted for 0.05⬍␣⬍0.10. An analysis of intraobserver variation was not included with this study, because several studies show interobserver variation has more impact than does intraobserver variation (9, 13, 15, 19, 21). We assessed interobserver (IO) variation for both CT and PET/CT fusion with a one-way analysis of variance (ANOVA) for correlated samples and a post-ANOVA Tukey HSD test to flush out individual mean differences (25). To compare degree of overlap, we applied a Boolean pairwise comparison introduced by Daisne et al. (26). In this method, we define Volume A ⬃ Volume B ⫽ Volume (A alone) ⁄ Volume B * 100%, where we define (⬃) as the “Boolean operator.” In other words, “Volume (A alone)” is the part of volume A that does not overlap
728
I. J. Radiation Oncology
● Biology ● Physics
with volume B, and “A⬃B” is the part of volume A that does not overlap with volume B expressed as a percentage of volume B (Figs. 1,2). Boolean percentages must be considered in reflexive pairs because, most often, A⬃B ⫽ B⬃A. Higher percentage pairs imply less overlap and vice versa. Only together can the values accurately represent the interaction of the 2 volumes. Intermodality (IM) variation for each physician was analyzed by application of a Student’s paired-sample t test and a Boolean pairwise comparison. Average IM variation was measured by taking the average magnitude of all 4 physicians’ volumes for each modality and comparing them with a Student’s paired-sample t test. Interdisciplinary (ID) variation was analyzed by averaging the CT volumes of physician A and B, and then C and D. The same was done for fusion volumes. The averages were compared by application of a Student’s paired-sample t test across disciplines (neuroradiologists vs. radiation oncologists within modalities) and modalities (CT vs. PET/CT fusion for neuradiologists, for example.) Boolean pairwise comparisons were not performed for any averaged volumes because of the geometric nature of the test. All statistics were performed on VassarStats (27), a Web-based statistical engine.
RESULTS Sixteen patients were contoured by each of the 4 physicians. The protocol evolved slightly over the first few contouring sessions, so any patients that were not contoured with the final protocol in place were recontoured at the conclusion of the study. This procedure affected Patients 1 and 2 for Physician A and Patients 1, 2, and 3 for Physician D. Physician A recontoured Patient 6 because of misunderstood instructions. Physician D recontoured Patient 8 because of data loss. An average of 3.6 months passed between contouring dates (range, 2– 4 months) to erase any memory of previous experience. Neither physician recog-
Table 1. Patient characteristics Patient
Age
Sex
Location
Type
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
53 45 47 56 58 97 75 74 58 53 49 77 51 41 54 41
M M M M M F M M M M M F M M F M
R base of tongue R maxillary sinus Nasopharynx L tonsil Nasopharynx L optic nerve Glottis L orbit R base of tongue Larynx Nasopharynx Cervical esophagus L nasal vestibule Nasopharynx L base of tongue L base of tongue
SCC NHL SCC SCC SCC Melanoma SCC B-cell lymphoma SCC SCC SCC SCC SCC SCC SCC SCC
Abbreviations: NHL ⫽ non-Hodgkin’s lymphoma; SCC ⫽ squamous cell carcinoma.
Volume 65, Number 3, 2006
Table 2. Results of interobserver Boolean analysis for average CT volumes A A B C D
31% 38% 16%
B
C
D
30%
26% 18%
60% 54% 60%
27% 11%
7%
Abbreviation: CT ⫽ computed tomography. Note: Read row to column (example, A⬃B ⫽ first row, second column).
nized the patients as part of the data set they had already contoured. Physician A was asked to modify the volumes for Patients 12 and 15 to eliminate lymph node involvement, which was not included in the GTV for the purposes of the study. Interobserver The analysis of variance (ANOVA) revealed significant tendencies for the CT volumes (F ⫽ 2.31, p ⫽ 0.09) and significance for the PET/CT fusion (FUS) volumes (F ⫽ 8.23, p ⫽ 0.0002) (Fig. 3). The Tukey HSD test specified that the mean fusion volumes of Physician A and Physician C were significantly larger than the mean fusion volume of Physician D (p ⬍ 0.05, p ⬍ 0.01, respectively). No significant difference occurred between Physician A and Physician C or any other physician pairs. The mean results of the IO Boolean comparison are listed in Table 2 (for CT) and Table 3 (for PET/CT fusion). Dramatic changes occurred in some Boolean comparisons from CT to fusion, especially in the comparison of Physician C and Physician D. Intermodality Physician C’s PET/CT fusion volumes were significantly larger than his CT volumes (p ⬍ 0.01). Conversely, Physician D’s CT volumes tended to be larger than his fusion volumes (p ⫽ 0.06). No significant differences were seen in the IM comparison for any other physician. On average, the physicians’ fusion volumes tended to be larger than their CT volumes (p ⫽ 0.07). Table 3. Results of interobserver Boolean analysis for average PET/CT fusion volumes A A B C D
15% 45% 5%
B
C
D
24%
12% 12%
90% 79% 137%
56% 8%
1%
Abbreviations: CT ⫽ computed tomography; PET ⫽ positron emission tomography. Note: Read row to column (example, A⬃B ⫽ first row, second column).
Variability of gross tumor volume delineation
Table 4. Results of intermodality Boolean analysis
A B C D
CT⬃FUS
FUS⬃CT
23% 34% 1% 23%
37% 39% 31% 7%
Abbreviations: CT ⫽ computed tomography; FUS ⫽ fusion.
The mean results of the IM Boolean comparison are listed in Table 4. Again, the data for Physicians C and D are particularly striking, as Physician C’s CT⬃FUS is nearly zero at 1% and Physician D’s CT⬃FUS is greater than his FUS⬃CT, contrary to the other physicians. Interdisciplinary No significant difference occurred between the mean CT volumes contoured by neuroradiologists and radiation oncologists, and no significant difference occurred between the mean PET/CT fusion volumes contoured by neuroradiologists and radiation oncologists. No significant difference occurred between the neuroradiologists’ CT and fusion volumes. The radiation oncologists’ mean fusion volume, however, was significantly larger than their mean CT volume (p ⬍ 0.05). DISCUSSION Several institutions have investigated the value of PET/CT fusion imaging in HNC target-volume delineation, with mixed results (28 –33). One study compares HNC target volumes defined by use of CT, MRI, and PET single-modality imaging with a surgical specimen in 9 cases. FDG-PET was the most accurate imaging modality when compared with the surgical specimen but still overestimated the tumor by 46% (26). This study contained a relatively small amount of raw data, but several analytical methods were used, and, therefore, many results were obtained. When taken out of context, the analyses utilized here can give a partial and some-
● A. C. RIEGEL et al.
729
times misleading explanation of the observed phenomena. The study results must be interpreted as a whole. Ketting et al. (10) concluded that physicians in their study failed to use “three-dimensional reasoning” to gain a correct geometric definition of the PTV. Qualitatively, Physicians A, B, and C in the current study were skilled at this technique. They often quickly scanned through adjacent slices to review their contours and gain a sense of the threedimensional nature of the tumor. Physician D, however, was clearly the most proficient at thinking three dimensionally, which is probably closely related to his technological savvy and his use of a wide array of contouring tools in the treatment-planning program. Physicians B and C used contradictory techniques. Physician B would contour the suspected GTV on the basis of the CT but often totally disregard this information when given the PET information and contour only PET avidity. Physician C, on the other hand, would contour the suspected GTV on the basis of the CT, and then, for the fusion contour, draw the union of the CT contour and PET avidity. Physician A’s method tended to be a mixture of the methods of Physicians B and C. Often, Physician A would “split the difference” and contour the compromise between the drawn CT contour and PET avidity. The statistical and relative pairwise results correlate closely with qualitative observations of each physician’s unique contouring method. It is evident that the differing methods and, hence, most of the variation in fusion contouring are caused by differing interpretations of contradictions between modalities. When given 2 modalities that disagree on extent of gross disease, physicians are unsure which modality should be given more weight in their decision-making process. The PET/CT fusion ANOVA revealed that the fusion volumes of Physician A and Physician C were significantly larger than those of Physician D. No significant difference was seen between the magnitudes of the CT volumes in the ANOVA, but a supplemental Student’s paired sample t test (performed in light of the fusion ANOVA) and the Boolean analysis exhibited a similar trend with regards to volumes A, C, and D. The difference between Physicians C and D is
Fig. 1. Two-dimensional intermodality Boolean analysis for Patient 12, Physician A. (a) CT and PET/CT fusion contours on a single slice. Geometrical interpretations of (b) FUSION⬃CT and (c) CT⬃FUSION. CT ⫽ computed tomography; PET ⫽ positron emission tomography.
730
I. J. Radiation Oncology
● Biology ● Physics
Volume 65, Number 3, 2006
Fig. 2. Three-dimensional intermodality Boolean analysis for Patient 12 and Physician A. (a) CT and PET/CT fusion volumes. Geometrical interpretations of (b) FUSION⬃CT and (c) CT⬃FUSION. CT ⫽ computed tomography; PET ⫽ positron emission tomography.
particularly noteworthy. Because the method employed in the ID variation analyses involved averaging the patient GTVs for each modality (comparing the mean CT volumes of Physicians C and D with the mean fusion volumes of Physicians C and D, respectively), the difference was great enough to overcome reciprocal methods that could offset. Even with Physician C’s large fusion volume averaged with Physician D’s small fusion volume (and vice versa for the CT volumes), the difference in averaged values was still great enough for statistical significance at p ⬍ 0.05. The neuroradiologists, however, exhibited no such statistical difference. In fact, the IO Boolean analyses show that their uniformity actually improved from CT to fusion (their percentages both decreased in the IO fusion Boolean analysis). The neuroradiologist pair was the only physician pair to exhibit this phenomenon. Three studies produced similar results when measuring IO variation between radiologists and radiation oncologists in GTV delineation for bladder cancer, lung cancer, and brain cancer. In the first 2 studies, the radiologist volumes
were more consistent than were radiation oncologist volumes (5, 20). In the third study, radiologist volumes were marginally more consistent than radiation oncologist volumes (12). Weiss et al. (11) compared gynecologists and radiation oncologists and found analogous results: Radiation oncologists contoured less common volume than did gynecologists in CTV delineation of cervical cancer. These studies (excluding Weltens et al. (12)) and 2 additional studies found that radiation oncologists contour larger volumes than do other specialists (5, 7, 11, 14, 20), which contradicts the results of our study. In PET/CT fusion, a radiation oncologist contoured the smallest volume for 13 out of 16 patients. The lack of significant difference between volumes contoured by radiation oncologists and radiologists in the ID comparison reflects the contrasting methodologies employed by the radiation oncologists, which most likely offset the significance. The results of our study directly contradict the findings of 3 similar studies. Caldwell et al. (18) found that PET/CT fusion (compared with CT) significantly decreased IO vari-
Fig. 3. Interobserver variation for Patient 5 and Physicians A to D. (a) Computed tomography and (b) positron emission tomography/computed tomography fusion.
Variability of gross tumor volume delineation
ation between 3 radiation oncologists who delineated GTV on NSCLC. Ciernik et al. (33) similarly reported that 2 radiation oncologists contoured 39 patients of mixed diagnosis (12 were HNC) more consistently on PET/CT fusion than on CT alone. Syed et al. (34) found that PET/CT significantly reduced IO variation as compared with PET alone in 3 observers and 24 HNC patients. Our study, which includes more physicians than all of the aforementioned studies and more patients than both head-and-neck studies, finds weakly significant IO variation with single-modality CT but statistically significant variation with dual-modality PET/CT fusion. Other studies have compared the amount of IO variation in MRI and CT. In comparing CT and CT viewed simultaneously with MRI, Weltens et al. (12) found significant IO variation between 9 observers who delineated GTV on brain tumors with CT. Supplementary MRI information did not decrease IO variation, and no significant ID variation between radiation oncologists, radiologists, and neurosurgeons was found. While measuring variation in prostate delineation, Rasch et al. (35) found that MRI did not significantly decrease IO variation when compared with CT, but IO variation was generally smaller on axial MRI, particularly near the plexus Santonini and the apex of the prostate. The aforementioned studies certainly suggest a relationship between imaging modality and the relative level of IO variation. Although previous studies show CT and MRI contouring to be comparable and PET/CT contouring to generate less IO variation than CT, the results of the current study contradict the latter claim and invite further investigation of the relationship between PET/CT and IO variation. One drawback to the present study is the focus on GTV rather than on planned target volume (PTV). The PTV is the volume to which a prescribed dose is actually delivered, whereas GTV is the volume of gross disease. Although studies have emphasized the misconception that GTV is easier to contour than other target volumes (7), and by definition the PTV must include the clinical target volume (which must include the GTV) (36), similar GTV volumes are not guaranteed to produce a uniform PTV or vice versa. A study of PTV reproducibility under the same circumstances should be performed. Error Various sources of error are noted. The radiation oncologists were unavoidably biased. Although names were shielded from the physicians to minimize recognition, almost all of the patients (except Patients 2 and 16) were
● A. C. RIEGEL et al.
731
treated by the participating radiation oncologists. They were also more familiar with the treatment-planning system than were the neuroradiologists, as it is the primary treatmentplanning system at our institution. Often in HNC, the tumor and malignant lymph nodes can be adjacent, which makes the definition of a tumor-only volume nearly impossible. The decision to omit the lymph nodes as a part of the GTV inherently caused some inaccuracy and variation not related to interobserver effects. Several sources of systematic error were present in our protocol. Although we tried to simulate a clinical situation, physicians were not given full diagnostic imaging data, including contrast CT, which, especially for lesions located in the nasopharynx, would make GTV delineation much easier. Because the physician was not blinded from his or her CT contour while drawing the fusion contour, the protocol lends itself to Physician C’s methodology; that is, drawing the union of a previously drawn contour and a newly suspected area. Without the previous contour present, this method would be much more difficult to replicate from patient to patient. CONCLUSIONS The purpose of this study was to assess the need for an institutional protocol regarding the use of multimodality imaging (specifically PET/CT fusion) in treatment planning. Each physician in the study was observed to use a unique contouring technique for fusion imaging, probably caused by varying interpretations of conflicting modalities, which manifested in statistically significant differences in fusion GTVs. Given the results of this analysis, in conjunction with overlap comparisons and qualitative observations, we report that institutional protocol is indeed necessary to increase the precision of GTV delineation. Guided by the results of this study and others in the literature, we are currently working with the nuclear medicine and radiology departments to shape an institutional protocol based on a “gold standard” concept for the use of multimodality imaging in treatment planning. One of the shortcomings of this study, as with all similar studies, is the inability to determine the location and volume of the “true” tumor, and, hence, the most accurate imaging modality. Eventually, the final result of our multidisciplinary efforts to define a delineation protocol will be tested in a comprehensive prospective study to determine its therapeutic efficacy on the basis of patient outcomes. However, further studies must be performed to analyze the effectiveness of a GTV-delineation protocol for use in PTV contouring.
REFERENCES 1. Lee N, Xia P, Fischbein NJ, et al. Intensity-modulated radiation therapy for head-and-neck cancer: The UCSF experience focusing on target volume delineation. Int J Radiat Oncol Biol Phys 2003;57:49 – 60. 2. Chao KS, Low DA, Perez CA, et al. Intensity-modulated radiation therapy in head and neck cancers: The Mallinckrodt experience [Abstract]. Int J Cancer 2000;90:92–103.
3. Mohan R, Qiuwen W, Manning M, et al. Radiobiological considerations in the design of fractionation strategies for intensity-modulated radiation therapy of head and neck cancers. Int J Radiat Oncol Biol Phys 2000;46: 619 – 630. 4. Nutting C, Dearnaley DP, Webb S, et al. Intensity modulated radiation therapy: A clinical review. Br J Radiol 2000;73:459 – 469.
732
I. J. Radiation Oncology
● Biology ● Physics
5. 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 –931. 6. Rasch C, Eisbruch A, Remeijer P, et al. Irradiation of paranasal sinus tumors, a delineation and dose comparison study. Int J Radiat Oncol Biol Phys 2002;52:120 –127. 7. 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. 8. Bradley J, Thorstad W, Mutic S, et al. Impact of FDG-PET on radiation therapy volume delineation in non-small-cell lung cancer. Int J Radiat Oncol Biol Phys 2004;59:78 – 86. 9. Tai P, Van Dyk J, Yu E, et al. Variability of target volume delineation in cervical esophageal cancer. Int J Radiat Oncol Biol Phys 1998;42:277–288. 10. Ketting C, Austin-Seymour M, Kalet I, et al. Consistency of three-dimensional planning target volumes across physicians and institutions. Int J Radiat Oncol Biol Phys 1997;37:445– 453. 11. 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. 12. 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. 13. Hermans R, Feron M, Bellon E, et al. Laryngeal tumor volume measurements determined with CT: A study on intra- and interobserver variation. Int J Radiat Oncol Biol Phys 1998; 40:553–557. 14. Leunens G, Menten J, Weltens C, et al. Quality assessment of medical decision making in radiation oncology: Variability in target volume delineation for brain tumours. Radiother Oncol 1993;29:169 –175. 15. Fiorino C, Reni M, Bolognesi A, et al. Intra- and interobserver variability in contouring prostate and seminal vesicles: Implications for conformal treatment planning. Radiother Oncol 1998;47:285–292. 16. Livsey JE, Wylie JP, Swindell R, et al. Do differences in target volume definition in prostate cancer lead to clinically relevant differences in normal tissue toxicity? Int J Radiat Oncol Biol Phys 2004;60:1076 –1081. 17. Coles CE, Hoole ACF, 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;60:189 –194. 18. Caldwell C, Mah K, Ung Y, et al. Observer variation in contouring gross tumor volume in patients with poorly defined non-small-cell lung tumors on CT: The impact of 18FDGhybrid PET fusion. Int J Radiat Oncol Biol Phys 2001;51: 923–931. 19. Senan S, van Sörnsen de Koste J, Samson M, et al. Evaluation of a target contouring protocol for 3D conformal radiotherapy in non-small cell lung cancer. Radiother Oncol 1999;53:247– 255. 20. Giraud P, Elles S, Helfre S, et al. Conformal radiotherapy for lung cancer: Different delineation of the gross tumor volume
Volume 65, Number 3, 2006
21.
22.
23.
24. 25. 26.
27. 28.
29.
30.
31. 32.
33.
34. 35. 36.
(GTV) by radiologists and radiation oncologists. Radiother Oncol 2002;62:27–36. Hamilton CS, Denham JW, Joseph DJ, et al. Treatment and planning decisions in non-small cell carcinoma of the lung: An Australasian patterns of practice study [Abstract]. Clin Oncol R Coll Radiol 1992;4:141–147. Saarnak A, Boersma M, Van Bunningen B, et al. Inter-observer variation in delineation of bladder and rectum contours for brachytherapy of cervical cancer. Radiother Oncol 2000; 56:37– 42. Oshida M, Uno K, Masato S, et al. Predicting the prognoses of breast carcinoma patients with positron emission tomography using 2-deoxy-2-fluoro[18F]-D-glucose. Cancer 1998;82:2227– 2234. Aoki J, Watanabe H, Shinozaki T, et al. FDG PET of primary benign and malignant bone tumors: Standardized uptake value in 52 lesions. Radiology 2001;219:774 –777. Lowry R. Concepts and applications of inferential statistics. Vassar College. Available at: http://faculty.vassar.edu/lowry/ webtext.html. 2005. Accessed February 16, 2005. Daisne JF, Duprez T, Weynand B, et al. Tumor volume in pharyngolarygneal squamous cell carcinoma: Comparison at CT, MR imaging, and FDG PET and validation with surgical specimen. Radiology 2004;233:93–100. Lowry R. Vassar stats: Web site for statistical computation. Vassar College. Available at: http://faculty.vassar.edu/lowry/ VassarStats.html. 2005. Accessed March 29, 2005. Paulino AC, Koshy M, Howell R, et al. Comparison of CTand FDG-PET-defined gross tumor volume in intensity-modulated radiotherapy for head-and-neck cancer. Int J Radiat Oncol Biol Phys 2005;61:1385–1392. Scarfone C, Lavely W, Cmelak A, et al. Prospective feasibility trial of radiotherapy target definition for head and neck cancer using 3-dimensional PET and CT imaging. J Nucl Med 2004; 45:542–552. Nishioka T, Tohru S, Shirato H, et al. Image fusion between 18 FDG-PET and MRI/CT for radiotherapy planning of oropharyngeal and nasopharyngeal carcinomas. Int J Radiat Oncol Biol Phys 2002;53:1051–1057. Schöder H, Yeung HWD, Gonen M, et al. Head and neck cancer: clinical usefulness and accuracy of PET/CT image fusion. Radiology 2004;231:65–72. Heron DE, Regiane SA, Flickinger J, et al. Hybrid PET-CT simulation for radiation treatment planning in head-and-neck cancers: A brief technical report. Int J Radiat Oncol Biol Phys 2004;60:1419 –1424. Ciernik F, Dizendorf E, Baumert B, et al. Radiation treatment planning with an integrated positron emission and computer tomography (PET/CT): A feasibility study. Int J Radiat Oncol Biol Phys 2003;57:853– 863. Syed R, Bomanji JB, Nagabhushan N, et al. Impact of combined (18)F-FDG PET/CT in head and neck tumours. Br J Cancer 2005;92:1046 –1050. 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. International Commission on Radiation Units and Measurements. Prescribing, recording and reporting photon beam therapy: ICRU Report 50. Washington, DC 1991.
doi:10.1016/j.ijrobp.2006.01.014
CLINICAL INVESTIGATION
Head and Neck
VARIABILITY OF GROSS TUMOR VOLUME DELINEATION IN HEAD-AND-NECK CANCER USING CT AND PET/CT FUSION ADAM C. RIEGEL, B.A.,* ANTHONY M. BERSON, M.D.,*† SYLVIE DESTIAN, M.D.,†‡ TRACY NG, M.D.,*† LAWRENCE B. TENA, M.D.,*† ROBIN J. MITNICK, M.D.,†‡ † AND PING S. WONG, PH.D.* *Department of Radiation Oncology, St. Vincent’s Comprehensive Cancer Center, New York, NY; †New York Medical College, Westchester, NY; and ‡Department of Radiology, St. Vincent’s Hospital, New York, NY Purpose: To assess the need for gross tumor volume (GTV) delineation protocols in head-and-neck cancer (HNC) treatment planning by use of positron emission tomography (PET)/computed tomography (CT) fusion imaging. Assessment will consist of interobserver and intermodality variation analysis. Methods and Materials: Sixteen HNC patients were accrued for the study. Four physicians (2 neuroradiologists and 2 radiation oncologists) contoured GTV on 16 patients. Physicians were asked to contour GTV on the basis of the CT alone, and then on PET/CT fusion. Statistical analysis included analysis of variance for interobserver variability and Student’s paired sample t test for intermodality and interdisciplinary variability. A Boolean pairwise analysis was included to measure degree of overlap. Results: Near-significant variation occurred across physicians’ CT volumes (p ⴝ 0.09) and significant variation occurred across physicians’ PET/CT volumes (p ⴝ 0.0002). The Boolean comparison correlates with statistical findings. One radiation oncologist’s PET/CT fusion volumes were significantly larger than his CT volumes (p < 0.01). Conversely, the other radiation oncologist’s CT volumes tended to be larger than his fusion volumes (p ⴝ 0.06). No significant interdisciplinary variation was seen. Significant disagreement occurred between radiation oncologists. Conclusion: Significant differences in GTV delineation were found between multiple observers contouring on PET/CT fusion. The need for delineation protocol has been confirmed. © 2006 Elsevier Inc. GTV delineation, PET/CT fusion, Head-and-neck cancer, Interobserver variation.
INTRODUCTION
small amount can result in a severely compromised plan, either overdosing of surrounding sensitive tissue, dangerously underdosing of cancerous tissue, or both (6 –10). One major source of inaccuracy or, technically, imprecision in GTV delineation is interobserver variability (11–14). Numerous studies have shown that target volumes drawn by multiple observers on conventional computed tomography (CT) vary significantly in areas such as the prostate and seminal vesicles (15, 16), brain (12, 14, 17), cervix (11), lung (7, 18 –21), bladder (5), cervical esophagus (9), and head and neck (6, 10, 13). Studies in the literature make various recommendations to reduce this variation, including more uncertainty analysis (7, 14), larger margins (15), multidisciplinary team contouring (1, 11, 17, 19, 20), and stricter delineation guidelines (1, 7, 11, 17, 20, 22). Two studies recommend the evaluation of multimodality imaging in target delineation (11, 20), which is the focus of this study.
The ever-increasing rate of technological innovation provides physicians around the world with access to cuttingedge equipment and a variety of treatment techniques, which ultimately allows them to individualize treatments to a greater degree than ever before. Intensity-modulated radiation therapy (IMRT), for example, assures high doses of radiation to the target volume and low doses to the surrounding sensitive tissues by creating a nonuniform dose distribution with the multileaf collimator, which is particularly useful in head-and-neck cancer (HNC), where normal and abnormal structures are in close proximity (1– 4). The treatment’s degree of success, however, begins with the treatment-planning process. Accurate delineation of the gross tumor volume (GTV) is critical in IMRT treatment planning because of the high dose gradient inherent to the technology (1, 5–7). Inaccuracy in the delineation of target volumes by a Reprint requests to: Anthony M. Berson, M.D., Radiation Oncology Department, St. Vincent’s Comprehensive Cancer Center, 325 W. 15th St., New York, NY 10011, Tel: (212) 367-1795; Fax: (212) 367-1742; E-mail: [email protected] The abstract of this study was presented at the Forty-Seventh
Annual Meeting of the American Society for Therapeutic Radiology and Oncology (ASTRO), October 19 –23, 2005, Denver, CO. Received June 8, 2005, and in revised form Jan 9, 2006. Accepted for publication Jan 10, 2006. 726
Variability of gross tumor volume delineation
Positron emission tomography that utilizes the radiotracer fluorodeoxyglucose-positron emission tomography (FDGPET) is another recent technological development now used in treatment planning to increase the accuracy of GTV delineation. PET images are coregistered with conventional CT images to create a “fused” PET/CT image, a 3D representation of a patient’s anatomic structure superimposed with metabolic function. Presently, no standard criteria exist for multimodality GTV delineation of HNC patients at our institution. The purpose of this study was to assess the need for institutional protocol by examination of the uniformity of GTV delineation across multiple observers by use of FDG-PET/CT fusion imaging. The investigation consisted of 2 comparisons: (1) evaluation of GTV-delineation variability between multiple physicians on CT and PET/CT fusion (including interdisciplinary variability), and (2) analysis of deviation between contours drawn on CT and PET/CT fusion. Any observed variability between physicians is justification for the development of standardized institutional protocol regarding the use of multimodality imaging in treatment planning. The basis for this protocol will be shaped by the results of this study. 18
METHODS AND MATERIALS Study participants Four physicians, labeled A, B, C, and D for the purposes of this study, volunteered to contour volumes on the 16 HNC patients. Physicians A and B were neuroradiologists with 17 and 15 years experience, respectively. Physicians C and D were radiation oncologists with 4.5 and 13 years experience, respectively. The following criteria were used to retrospectively select patients for this study: (1) The patient must have cancer in the head or neck. (2) Gross tumor must be present. (3) The patient must have received a PET/CT simulation. (4) The patient cannot have received any surgical or radiation treatment to the tumor in question before the PET/CT scan. Sixteen (16) patients, 13 men and 3 women, underwent PET/CT simulations between April 2004 and February 2005 and were enrolled in the study (Table 1). Mean patient age was 57 years (range, 41– 98 years). Tumor location varied from 5 in the nasopharynx, 4 in the base of tongue, to 1 in the tonsil, larynx, glottis, cervical esophagus, eye orbit, optic nerve, nasal vestibule, and maxillary sinus. Thirteen patients had squamous cell carcinoma, and 3 patients had non-Hodgkin’s lymphoma, B-cell lymphoma, and melanoma in the maxillary sinus, orbit, and optic nerve, respectively.
Scanning and contouring All 16 patients received an 18FDG PET/CT scan from a Discovery ST combined PET/CT scanner (GE Medical Systems, Milwaukee, WI). Each patient was injected with 15 to 20 mCi of 18 FDG and scanned on a flat scanning table an average of 2 hours after injection. Patients were immobilized with a Med-Tec MTAPS thermoplastic insert mask (Med-Tec, Orange City, IA) and a VacLok MT-VL-TYS-01 custom head rest. All CT scans were obtained in helical mode with a gantry rotation period of 0.8 sec and table pitch of 1.35 cm to 1 rotation. The number of slices was
● A. C. RIEGEL et al.
727
72 slices and 179 slices for 1 patient each, 91 slices for 5 patients, and 135 slices for 9 patients. Head-and-neck PET acquisition usually requires 2 bed positions, and scan duration is approximately 4 to 5 min per bed position. Images were reconstructed by use of OSEM iterative reconstruction, with post and loop filters at 6.71 and 4.30 full-width half-maximum, respectively. Images were managed through GE Xeleris software (GE Healthcare, Waukesha, WI). CT pixel size was 0.98 ⫻ 0.98 mm2 and PET pixel size was 4.30 ⫻ 4.30 mm2. The field of view was 55 cm in diameter, and slice thickness was 3.75 mm. Computed tomography and PET images were transferred via DICOM protocol to ECLIPSE (Varian, Palo Alto, CA) treatmentplanning workstations. Images were fused by use of DICOM coordinate registration and checked visually for accuracy. For each patient, separate treatment-planning profiles were created for the purpose of the study. Names were shielded to minimize bias. The primary investigator (PI) facilitated each contouring session. Each physician was asked questions regarding their familiarity with the treatment-planning system, and a tutorial customized to each physician’s skill level was given. Physicians were given the patient’s diagnosis (general tumor location and biopsy results), symptoms, brief clinical history, and physical examination. The PI instructed each physician to contour only gross tumor on the axial CT images. Lymph nodes were not included in the definition of GTV for simplification of the volumetric analysis. The CT window levels were discretionary. After the physician verbally acknowledged completion, the volume was saved and the PI instructed the physician to contour gross tumor only, again no lymph node involvement, on the PET/CT fused images. The standardized uptake-value (SUV) window was discretionary, but the default window was arbitrarily set at a lower limit of 3 and an upper limit of 4, as SUV ⫽ 3 has previously been used as a threshold for malignancy (23, 24). Again, after verbal acknowledgment of completion, the volume was saved. Experimental GTVs contoured for the purposes of the study were not explicitly used for treatment planning.
Analysis The analysis consisted of 3 categories of volumetric comparison: interobserver variation (across physicians within a modality), intermodality variation (between CT and PET/CT fusion for each physician and on average), and interdisciplinary variation (across disciplines within a modality and across modalities within a discipline). All 3 categories were analyzed with a mean-significance test (category-dependent) or overlap analysis. The significance threshold was set at ␣ ⫽ 0.05. “Near” significance or “tendency” was noted for 0.05⬍␣⬍0.10. An analysis of intraobserver variation was not included with this study, because several studies show interobserver variation has more impact than does intraobserver variation (9, 13, 15, 19, 21). We assessed interobserver (IO) variation for both CT and PET/CT fusion with a one-way analysis of variance (ANOVA) for correlated samples and a post-ANOVA Tukey HSD test to flush out individual mean differences (25). To compare degree of overlap, we applied a Boolean pairwise comparison introduced by Daisne et al. (26). In this method, we define Volume A ⬃ Volume B ⫽ Volume (A alone) ⁄ Volume B * 100%, where we define (⬃) as the “Boolean operator.” In other words, “Volume (A alone)” is the part of volume A that does not overlap
728
I. J. Radiation Oncology
● Biology ● Physics
with volume B, and “A⬃B” is the part of volume A that does not overlap with volume B expressed as a percentage of volume B (Figs. 1,2). Boolean percentages must be considered in reflexive pairs because, most often, A⬃B ⫽ B⬃A. Higher percentage pairs imply less overlap and vice versa. Only together can the values accurately represent the interaction of the 2 volumes. Intermodality (IM) variation for each physician was analyzed by application of a Student’s paired-sample t test and a Boolean pairwise comparison. Average IM variation was measured by taking the average magnitude of all 4 physicians’ volumes for each modality and comparing them with a Student’s paired-sample t test. Interdisciplinary (ID) variation was analyzed by averaging the CT volumes of physician A and B, and then C and D. The same was done for fusion volumes. The averages were compared by application of a Student’s paired-sample t test across disciplines (neuroradiologists vs. radiation oncologists within modalities) and modalities (CT vs. PET/CT fusion for neuradiologists, for example.) Boolean pairwise comparisons were not performed for any averaged volumes because of the geometric nature of the test. All statistics were performed on VassarStats (27), a Web-based statistical engine.
RESULTS Sixteen patients were contoured by each of the 4 physicians. The protocol evolved slightly over the first few contouring sessions, so any patients that were not contoured with the final protocol in place were recontoured at the conclusion of the study. This procedure affected Patients 1 and 2 for Physician A and Patients 1, 2, and 3 for Physician D. Physician A recontoured Patient 6 because of misunderstood instructions. Physician D recontoured Patient 8 because of data loss. An average of 3.6 months passed between contouring dates (range, 2– 4 months) to erase any memory of previous experience. Neither physician recog-
Table 1. Patient characteristics Patient
Age
Sex
Location
Type
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
53 45 47 56 58 97 75 74 58 53 49 77 51 41 54 41
M M M M M F M M M M M F M M F M
R base of tongue R maxillary sinus Nasopharynx L tonsil Nasopharynx L optic nerve Glottis L orbit R base of tongue Larynx Nasopharynx Cervical esophagus L nasal vestibule Nasopharynx L base of tongue L base of tongue
SCC NHL SCC SCC SCC Melanoma SCC B-cell lymphoma SCC SCC SCC SCC SCC SCC SCC SCC
Abbreviations: NHL ⫽ non-Hodgkin’s lymphoma; SCC ⫽ squamous cell carcinoma.
Volume 65, Number 3, 2006
Table 2. Results of interobserver Boolean analysis for average CT volumes A A B C D
31% 38% 16%
B
C
D
30%
26% 18%
60% 54% 60%
27% 11%
7%
Abbreviation: CT ⫽ computed tomography. Note: Read row to column (example, A⬃B ⫽ first row, second column).
nized the patients as part of the data set they had already contoured. Physician A was asked to modify the volumes for Patients 12 and 15 to eliminate lymph node involvement, which was not included in the GTV for the purposes of the study. Interobserver The analysis of variance (ANOVA) revealed significant tendencies for the CT volumes (F ⫽ 2.31, p ⫽ 0.09) and significance for the PET/CT fusion (FUS) volumes (F ⫽ 8.23, p ⫽ 0.0002) (Fig. 3). The Tukey HSD test specified that the mean fusion volumes of Physician A and Physician C were significantly larger than the mean fusion volume of Physician D (p ⬍ 0.05, p ⬍ 0.01, respectively). No significant difference occurred between Physician A and Physician C or any other physician pairs. The mean results of the IO Boolean comparison are listed in Table 2 (for CT) and Table 3 (for PET/CT fusion). Dramatic changes occurred in some Boolean comparisons from CT to fusion, especially in the comparison of Physician C and Physician D. Intermodality Physician C’s PET/CT fusion volumes were significantly larger than his CT volumes (p ⬍ 0.01). Conversely, Physician D’s CT volumes tended to be larger than his fusion volumes (p ⫽ 0.06). No significant differences were seen in the IM comparison for any other physician. On average, the physicians’ fusion volumes tended to be larger than their CT volumes (p ⫽ 0.07). Table 3. Results of interobserver Boolean analysis for average PET/CT fusion volumes A A B C D
15% 45% 5%
B
C
D
24%
12% 12%
90% 79% 137%
56% 8%
1%
Abbreviations: CT ⫽ computed tomography; PET ⫽ positron emission tomography. Note: Read row to column (example, A⬃B ⫽ first row, second column).
Variability of gross tumor volume delineation
Table 4. Results of intermodality Boolean analysis
A B C D
CT⬃FUS
FUS⬃CT
23% 34% 1% 23%
37% 39% 31% 7%
Abbreviations: CT ⫽ computed tomography; FUS ⫽ fusion.
The mean results of the IM Boolean comparison are listed in Table 4. Again, the data for Physicians C and D are particularly striking, as Physician C’s CT⬃FUS is nearly zero at 1% and Physician D’s CT⬃FUS is greater than his FUS⬃CT, contrary to the other physicians. Interdisciplinary No significant difference occurred between the mean CT volumes contoured by neuroradiologists and radiation oncologists, and no significant difference occurred between the mean PET/CT fusion volumes contoured by neuroradiologists and radiation oncologists. No significant difference occurred between the neuroradiologists’ CT and fusion volumes. The radiation oncologists’ mean fusion volume, however, was significantly larger than their mean CT volume (p ⬍ 0.05). DISCUSSION Several institutions have investigated the value of PET/CT fusion imaging in HNC target-volume delineation, with mixed results (28 –33). One study compares HNC target volumes defined by use of CT, MRI, and PET single-modality imaging with a surgical specimen in 9 cases. FDG-PET was the most accurate imaging modality when compared with the surgical specimen but still overestimated the tumor by 46% (26). This study contained a relatively small amount of raw data, but several analytical methods were used, and, therefore, many results were obtained. When taken out of context, the analyses utilized here can give a partial and some-
● A. C. RIEGEL et al.
729
times misleading explanation of the observed phenomena. The study results must be interpreted as a whole. Ketting et al. (10) concluded that physicians in their study failed to use “three-dimensional reasoning” to gain a correct geometric definition of the PTV. Qualitatively, Physicians A, B, and C in the current study were skilled at this technique. They often quickly scanned through adjacent slices to review their contours and gain a sense of the threedimensional nature of the tumor. Physician D, however, was clearly the most proficient at thinking three dimensionally, which is probably closely related to his technological savvy and his use of a wide array of contouring tools in the treatment-planning program. Physicians B and C used contradictory techniques. Physician B would contour the suspected GTV on the basis of the CT but often totally disregard this information when given the PET information and contour only PET avidity. Physician C, on the other hand, would contour the suspected GTV on the basis of the CT, and then, for the fusion contour, draw the union of the CT contour and PET avidity. Physician A’s method tended to be a mixture of the methods of Physicians B and C. Often, Physician A would “split the difference” and contour the compromise between the drawn CT contour and PET avidity. The statistical and relative pairwise results correlate closely with qualitative observations of each physician’s unique contouring method. It is evident that the differing methods and, hence, most of the variation in fusion contouring are caused by differing interpretations of contradictions between modalities. When given 2 modalities that disagree on extent of gross disease, physicians are unsure which modality should be given more weight in their decision-making process. The PET/CT fusion ANOVA revealed that the fusion volumes of Physician A and Physician C were significantly larger than those of Physician D. No significant difference was seen between the magnitudes of the CT volumes in the ANOVA, but a supplemental Student’s paired sample t test (performed in light of the fusion ANOVA) and the Boolean analysis exhibited a similar trend with regards to volumes A, C, and D. The difference between Physicians C and D is
Fig. 1. Two-dimensional intermodality Boolean analysis for Patient 12, Physician A. (a) CT and PET/CT fusion contours on a single slice. Geometrical interpretations of (b) FUSION⬃CT and (c) CT⬃FUSION. CT ⫽ computed tomography; PET ⫽ positron emission tomography.
730
I. J. Radiation Oncology
● Biology ● Physics
Volume 65, Number 3, 2006
Fig. 2. Three-dimensional intermodality Boolean analysis for Patient 12 and Physician A. (a) CT and PET/CT fusion volumes. Geometrical interpretations of (b) FUSION⬃CT and (c) CT⬃FUSION. CT ⫽ computed tomography; PET ⫽ positron emission tomography.
particularly noteworthy. Because the method employed in the ID variation analyses involved averaging the patient GTVs for each modality (comparing the mean CT volumes of Physicians C and D with the mean fusion volumes of Physicians C and D, respectively), the difference was great enough to overcome reciprocal methods that could offset. Even with Physician C’s large fusion volume averaged with Physician D’s small fusion volume (and vice versa for the CT volumes), the difference in averaged values was still great enough for statistical significance at p ⬍ 0.05. The neuroradiologists, however, exhibited no such statistical difference. In fact, the IO Boolean analyses show that their uniformity actually improved from CT to fusion (their percentages both decreased in the IO fusion Boolean analysis). The neuroradiologist pair was the only physician pair to exhibit this phenomenon. Three studies produced similar results when measuring IO variation between radiologists and radiation oncologists in GTV delineation for bladder cancer, lung cancer, and brain cancer. In the first 2 studies, the radiologist volumes
were more consistent than were radiation oncologist volumes (5, 20). In the third study, radiologist volumes were marginally more consistent than radiation oncologist volumes (12). Weiss et al. (11) compared gynecologists and radiation oncologists and found analogous results: Radiation oncologists contoured less common volume than did gynecologists in CTV delineation of cervical cancer. These studies (excluding Weltens et al. (12)) and 2 additional studies found that radiation oncologists contour larger volumes than do other specialists (5, 7, 11, 14, 20), which contradicts the results of our study. In PET/CT fusion, a radiation oncologist contoured the smallest volume for 13 out of 16 patients. The lack of significant difference between volumes contoured by radiation oncologists and radiologists in the ID comparison reflects the contrasting methodologies employed by the radiation oncologists, which most likely offset the significance. The results of our study directly contradict the findings of 3 similar studies. Caldwell et al. (18) found that PET/CT fusion (compared with CT) significantly decreased IO vari-
Fig. 3. Interobserver variation for Patient 5 and Physicians A to D. (a) Computed tomography and (b) positron emission tomography/computed tomography fusion.
Variability of gross tumor volume delineation
ation between 3 radiation oncologists who delineated GTV on NSCLC. Ciernik et al. (33) similarly reported that 2 radiation oncologists contoured 39 patients of mixed diagnosis (12 were HNC) more consistently on PET/CT fusion than on CT alone. Syed et al. (34) found that PET/CT significantly reduced IO variation as compared with PET alone in 3 observers and 24 HNC patients. Our study, which includes more physicians than all of the aforementioned studies and more patients than both head-and-neck studies, finds weakly significant IO variation with single-modality CT but statistically significant variation with dual-modality PET/CT fusion. Other studies have compared the amount of IO variation in MRI and CT. In comparing CT and CT viewed simultaneously with MRI, Weltens et al. (12) found significant IO variation between 9 observers who delineated GTV on brain tumors with CT. Supplementary MRI information did not decrease IO variation, and no significant ID variation between radiation oncologists, radiologists, and neurosurgeons was found. While measuring variation in prostate delineation, Rasch et al. (35) found that MRI did not significantly decrease IO variation when compared with CT, but IO variation was generally smaller on axial MRI, particularly near the plexus Santonini and the apex of the prostate. The aforementioned studies certainly suggest a relationship between imaging modality and the relative level of IO variation. Although previous studies show CT and MRI contouring to be comparable and PET/CT contouring to generate less IO variation than CT, the results of the current study contradict the latter claim and invite further investigation of the relationship between PET/CT and IO variation. One drawback to the present study is the focus on GTV rather than on planned target volume (PTV). The PTV is the volume to which a prescribed dose is actually delivered, whereas GTV is the volume of gross disease. Although studies have emphasized the misconception that GTV is easier to contour than other target volumes (7), and by definition the PTV must include the clinical target volume (which must include the GTV) (36), similar GTV volumes are not guaranteed to produce a uniform PTV or vice versa. A study of PTV reproducibility under the same circumstances should be performed. Error Various sources of error are noted. The radiation oncologists were unavoidably biased. Although names were shielded from the physicians to minimize recognition, almost all of the patients (except Patients 2 and 16) were
● A. C. RIEGEL et al.
731
treated by the participating radiation oncologists. They were also more familiar with the treatment-planning system than were the neuroradiologists, as it is the primary treatmentplanning system at our institution. Often in HNC, the tumor and malignant lymph nodes can be adjacent, which makes the definition of a tumor-only volume nearly impossible. The decision to omit the lymph nodes as a part of the GTV inherently caused some inaccuracy and variation not related to interobserver effects. Several sources of systematic error were present in our protocol. Although we tried to simulate a clinical situation, physicians were not given full diagnostic imaging data, including contrast CT, which, especially for lesions located in the nasopharynx, would make GTV delineation much easier. Because the physician was not blinded from his or her CT contour while drawing the fusion contour, the protocol lends itself to Physician C’s methodology; that is, drawing the union of a previously drawn contour and a newly suspected area. Without the previous contour present, this method would be much more difficult to replicate from patient to patient. CONCLUSIONS The purpose of this study was to assess the need for an institutional protocol regarding the use of multimodality imaging (specifically PET/CT fusion) in treatment planning. Each physician in the study was observed to use a unique contouring technique for fusion imaging, probably caused by varying interpretations of conflicting modalities, which manifested in statistically significant differences in fusion GTVs. Given the results of this analysis, in conjunction with overlap comparisons and qualitative observations, we report that institutional protocol is indeed necessary to increase the precision of GTV delineation. Guided by the results of this study and others in the literature, we are currently working with the nuclear medicine and radiology departments to shape an institutional protocol based on a “gold standard” concept for the use of multimodality imaging in treatment planning. One of the shortcomings of this study, as with all similar studies, is the inability to determine the location and volume of the “true” tumor, and, hence, the most accurate imaging modality. Eventually, the final result of our multidisciplinary efforts to define a delineation protocol will be tested in a comprehensive prospective study to determine its therapeutic efficacy on the basis of patient outcomes. However, further studies must be performed to analyze the effectiveness of a GTV-delineation protocol for use in PTV contouring.
REFERENCES 1. Lee N, Xia P, Fischbein NJ, et al. Intensity-modulated radiation therapy for head-and-neck cancer: The UCSF experience focusing on target volume delineation. Int J Radiat Oncol Biol Phys 2003;57:49 – 60. 2. Chao KS, Low DA, Perez CA, et al. Intensity-modulated radiation therapy in head and neck cancers: The Mallinckrodt experience [Abstract]. Int J Cancer 2000;90:92–103.
3. Mohan R, Qiuwen W, Manning M, et al. Radiobiological considerations in the design of fractionation strategies for intensity-modulated radiation therapy of head and neck cancers. Int J Radiat Oncol Biol Phys 2000;46: 619 – 630. 4. Nutting C, Dearnaley DP, Webb S, et al. Intensity modulated radiation therapy: A clinical review. Br J Radiol 2000;73:459 – 469.
732
I. J. Radiation Oncology
● Biology ● Physics
5. 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 –931. 6. Rasch C, Eisbruch A, Remeijer P, et al. Irradiation of paranasal sinus tumors, a delineation and dose comparison study. Int J Radiat Oncol Biol Phys 2002;52:120 –127. 7. 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. 8. Bradley J, Thorstad W, Mutic S, et al. Impact of FDG-PET on radiation therapy volume delineation in non-small-cell lung cancer. Int J Radiat Oncol Biol Phys 2004;59:78 – 86. 9. Tai P, Van Dyk J, Yu E, et al. Variability of target volume delineation in cervical esophageal cancer. Int J Radiat Oncol Biol Phys 1998;42:277–288. 10. Ketting C, Austin-Seymour M, Kalet I, et al. Consistency of three-dimensional planning target volumes across physicians and institutions. Int J Radiat Oncol Biol Phys 1997;37:445– 453. 11. 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. 12. 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. 13. Hermans R, Feron M, Bellon E, et al. Laryngeal tumor volume measurements determined with CT: A study on intra- and interobserver variation. Int J Radiat Oncol Biol Phys 1998; 40:553–557. 14. Leunens G, Menten J, Weltens C, et al. Quality assessment of medical decision making in radiation oncology: Variability in target volume delineation for brain tumours. Radiother Oncol 1993;29:169 –175. 15. Fiorino C, Reni M, Bolognesi A, et al. Intra- and interobserver variability in contouring prostate and seminal vesicles: Implications for conformal treatment planning. Radiother Oncol 1998;47:285–292. 16. Livsey JE, Wylie JP, Swindell R, et al. Do differences in target volume definition in prostate cancer lead to clinically relevant differences in normal tissue toxicity? Int J Radiat Oncol Biol Phys 2004;60:1076 –1081. 17. Coles CE, Hoole ACF, 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;60:189 –194. 18. Caldwell C, Mah K, Ung Y, et al. Observer variation in contouring gross tumor volume in patients with poorly defined non-small-cell lung tumors on CT: The impact of 18FDGhybrid PET fusion. Int J Radiat Oncol Biol Phys 2001;51: 923–931. 19. Senan S, van Sörnsen de Koste J, Samson M, et al. Evaluation of a target contouring protocol for 3D conformal radiotherapy in non-small cell lung cancer. Radiother Oncol 1999;53:247– 255. 20. Giraud P, Elles S, Helfre S, et al. Conformal radiotherapy for lung cancer: Different delineation of the gross tumor volume
Volume 65, Number 3, 2006
21.
22.
23.
24. 25. 26.
27. 28.
29.
30.
31. 32.
33.
34. 35. 36.
(GTV) by radiologists and radiation oncologists. Radiother Oncol 2002;62:27–36. Hamilton CS, Denham JW, Joseph DJ, et al. Treatment and planning decisions in non-small cell carcinoma of the lung: An Australasian patterns of practice study [Abstract]. Clin Oncol R Coll Radiol 1992;4:141–147. Saarnak A, Boersma M, Van Bunningen B, et al. Inter-observer variation in delineation of bladder and rectum contours for brachytherapy of cervical cancer. Radiother Oncol 2000; 56:37– 42. Oshida M, Uno K, Masato S, et al. Predicting the prognoses of breast carcinoma patients with positron emission tomography using 2-deoxy-2-fluoro[18F]-D-glucose. Cancer 1998;82:2227– 2234. Aoki J, Watanabe H, Shinozaki T, et al. FDG PET of primary benign and malignant bone tumors: Standardized uptake value in 52 lesions. Radiology 2001;219:774 –777. Lowry R. Concepts and applications of inferential statistics. Vassar College. Available at: http://faculty.vassar.edu/lowry/ webtext.html. 2005. Accessed February 16, 2005. Daisne JF, Duprez T, Weynand B, et al. Tumor volume in pharyngolarygneal squamous cell carcinoma: Comparison at CT, MR imaging, and FDG PET and validation with surgical specimen. Radiology 2004;233:93–100. Lowry R. Vassar stats: Web site for statistical computation. Vassar College. Available at: http://faculty.vassar.edu/lowry/ VassarStats.html. 2005. Accessed March 29, 2005. Paulino AC, Koshy M, Howell R, et al. Comparison of CTand FDG-PET-defined gross tumor volume in intensity-modulated radiotherapy for head-and-neck cancer. Int J Radiat Oncol Biol Phys 2005;61:1385–1392. Scarfone C, Lavely W, Cmelak A, et al. Prospective feasibility trial of radiotherapy target definition for head and neck cancer using 3-dimensional PET and CT imaging. J Nucl Med 2004; 45:542–552. Nishioka T, Tohru S, Shirato H, et al. Image fusion between 18 FDG-PET and MRI/CT for radiotherapy planning of oropharyngeal and nasopharyngeal carcinomas. Int J Radiat Oncol Biol Phys 2002;53:1051–1057. Schöder H, Yeung HWD, Gonen M, et al. Head and neck cancer: clinical usefulness and accuracy of PET/CT image fusion. Radiology 2004;231:65–72. Heron DE, Regiane SA, Flickinger J, et al. Hybrid PET-CT simulation for radiation treatment planning in head-and-neck cancers: A brief technical report. Int J Radiat Oncol Biol Phys 2004;60:1419 –1424. Ciernik F, Dizendorf E, Baumert B, et al. Radiation treatment planning with an integrated positron emission and computer tomography (PET/CT): A feasibility study. Int J Radiat Oncol Biol Phys 2003;57:853– 863. Syed R, Bomanji JB, Nagabhushan N, et al. Impact of combined (18)F-FDG PET/CT in head and neck tumours. Br J Cancer 2005;92:1046 –1050. 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. International Commission on Radiation Units and Measurements. Prescribing, recording and reporting photon beam therapy: ICRU Report 50. Washington, DC 1991.