Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
ORIGINAL ARTICLE Semi-automated primary tumor volume measurements by dynamic contrast-enhanced MRI in patients with head and neck cancer Wouter L. Lodder, MD,1* Kenneth G. A. Gilhuijs, PhD,2 Charlotte A. H. Lange, MD,3 Frank A. Pameijer, MD, PhD,4 Alfons J. M. Balm, MD, PhD,1,5 Michiel W. M. van den Brekel, MD, PhD1,5 1 Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute—Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX, Amsterdam, The Netherlands, 2Department of Radiology, Image Sciences Institute, University Medical Centre Utrecht, Heidelberglaan 100, 3584 CX, Utrecht, The Netherlands, 3Department of Radiology, The Netherlands Cancer Institute—Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX, Amsterdam, The Netherlands, 4Department of Radiology, University Medical Centre Utrecht, Heidelberglaan 100, 3584 CX, Utrecht, The Netherlands, 5Department of Otorhinolaryngology, Academic Medical Center, University of Amsterdam, Meibergdreef 9, 1105 AZ, Amsterdam, The Netherlands. Accepted 25 January 2012 Published online 19 April 2012 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/hed.22988 ABSTRACT: Background. Tumor volume is a significant prognostic factor in the treatment of malignant head and neck tumors. Unfortunately, it is not routinely measured because of the workload involved. Methods. Twenty-one patients, between 2009 and 2010, were studied. Dynamic contrast-enhanced MRI (DCE-MRI) at 3.0T was performed. A workstation previously developed for semi-automated segmentation of breast cancers on DCE-MRI was used to segment the head and neck cancers. The Pearson correlation analysis was used to assess the agreement between volumetric measurements and the manually derived gross tumor volume (GTV). INTRODUCTION Tumor volume is a significant prognostic factor in the treatment of malignant head and neck tumors.1–8 Several studies have confirmed the prognostic value of CT-determined tumor volume for outcome after definitive radiation therapy in head and neck cancer, including tonsillar,3 hypopharyngeal,4 supraglottic,5 and glottic6 cancer. Van den Broek et al1 found that in patients with advanced, unresectable head and neck cancer, treated with targeted chemoradiation, primary tumor volume is the most important independent prognostic factor. The Union Internationale Contre le Cancer classification (TNM) represents the validated standard tool to describe tumor extent and includes prognostic information on the probability of disease control.9 Unfortunately, tumor volume has not been incorporated in the TNM staging system. The main reason for this is probably that tumor volume is more difficult to assess than the maximal diameter of a tumor. Gross tumor volume (GTV) might be a more relevant factor than a 2-dimensional descriptive TNM classification or stage *Corresponding author: W. L. Lodder, Department of Head and Neck Oncology and Surgery, Netherlands Cancer Institute – Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands. E-mail: [email protected] Contract grant sponsor: This research was produced without financial support. Results. In 90.5% of the patients (19 of 21) correlation could be made between DCE-MRI and the manually derived GTV. The Pearson correlation coefficient between the automatically derived tumor volume at DCE-MRI and the manually derived GTVs was R2 ¼ 0.95 (p < .001). Conclusion. Semi-automated tumor volumes on DCE-MRI were representative of those derived from the manually derived GTV (R2 ¼ 0.95; C 2012 Wiley Periodicals, Inc. Head Neck 35: 521–526, 2013 p < .001). V KEY WORDS: dynamic contrast-enhanced MRI, head and neck cancer, tumor volume, semi-automated measurement, volume measurement grouping as prognostic factor for local control in head and neck cancer.10 GTV represented the most important prognostic indicator in head and neck squamous cell carcinoma treated with intensity modulated radiation therapy and is recommended to be considered for therapeutic decisions and estimation of outcome.10 Unfortunately, it is not routinely measured because of the workload involved. Assessment of semi-automatic volume measurements of the primary tumor has been performed for prostate cancer and breast cancer11,12 using dynamic contrast-enhanced MRI (DCE-MRI). Alderliesten et al13 showed that semiautomatic volumetric measurement of tumor extent accurately correlates with histopathology (R2 ¼ 0.84). In 2004, Shah et al14 reviewed the use of DCE-MRI in head and neck cancer. They concluded that DCE-MRI had great promise in evaluating carcinoma of head and neck, monitoring the treatment, and differentiating tumorinvolved lymph nodes from non–tumor-involved nodes. Recently, Hadjiski et al15 investigated the feasibility of computerized segmentation of lesions on head and neck CT scans and evaluated its potential for estimating changes in tumor volume in response to treatment of head and neck cancers. To our knowledge, no previous study was conducted to evaluate the use of DCE-MRI for semi-automated tumor volume measurements in patients with head and neck cancer. HEAD & NECK—DOI 10.1002/HED APRIL 2013 521 LODDER ET AL. PATIENTS AND METHODS TABLE 1. Patient and tumor characteristics. Ethical considerations Patient characteristics Institutional approval for the study was received. As patient anonymity was preserved, patient consent was not required for the retrospective review of records and images. DCE-MRIs were acquired using standard MRI sequences used for imaging of head and neck tumors. However, typically, the imaging is performed at a fixed moment after contrast injection. In this method, imaging is performed dynamically directly after contrast injection throughout the normal fixed scanning moment. Neither the scanning time nor the contrast injection dose is different using the dynamic approach compared to the standard approach. Sex Male Female Primary site Oral cavity Oropharynx Pathologic T classification T1 T2 T3 T4 Pathologic N classification N0 N1 N2a N2b N2c N3 Manually derived volume Automatically derived volume Patient data Twenty-four consecutive patients with histologically proven head and neck squamous cell carcinoma, treated between September 2009 and March 2010, were included in this study. A total of 21 patients were used for subsequent analyses. Three cases were excluded due to inappropriate fieldof-view. For the evaluation of the semi-automated volume measurements, the tumor had to be in the area of interest (differences in volume in these cases are not caused by the failure of the segmentation technique). Clinical characteristics including sex, primary site, pathological T classification, and pathological N classification of the 21 patients are summarized in Table 1. The tumors were located in the oral cavity (57%; n ¼ 12) and oropharynx (43%; n ¼ 9). One patient had a recurrent carcinoma and 20 patients had a primary tumor. Imaging protocol All patients underwent a preoperative MRI. DCE-MRI examinations were performed at 3.0 T (Philips Achieva, release 3.2.1, Philips Medical Systems, Best, The Netherlands) using a dedicated 16-channel SENSE neurovascular coil. Fast field echo acquisitions were used. One series was acquired before the injection of a gadolinium-based contrast agent (intravenous injection of 15 cc, Prohance; Bracco-Byk Gulden, Konstanz, Germany) and 79 series after injection. The series were acquired at intervals of 0.95 seconds. The following series (produced during the same MRI procedure) were available to the operator performing manual tumor delineations: STIR TSE COR, TR (repetition time), IR (inversion time), TE (echo time) 3880/180/20 ms, ETL: 12, FOV 300/228/40 mm, matrix: 320/320, 2 nex, slice thickness 4 mm; STIR TSE TRA, TR/IR/TE 4,228/180/20, ETL: 12, FOV: 180/200/80 mm, matrix 300/312, 2 nex, SW 3.5 mm, T1 TSE TRA, TR/ TE: 780/10, ETL: 5, FOV 180/180/80, matrix 384/384, 2 nex, slice thickness: 3.5 mm; T1 3D Thrive, TR/TE: 5/ 2,22, ETL:90, TA: 10, FOV 230/272/220, matrix 288/ 288, 2 nex, slice thickness: 0.8 mm. T1 TSE COR (postcontrast): TR/TE: 812/10, ETL: 6, FOV: 180/150/96 mm, matrix: 320/320, 3 nex, slice thickness 3.5 mm. 522 HEAD & NECK—DOI 10.1002/HED APRIL 2013 No. of patients % 12 9 57.1 42.9 12 9 57.1 42.9 4 6 7 4 19 28.6 33.4 19 4 19 5 23.8 2 9.5 9 42.9 1 4.8 0 0 0.50–42.8 (mean, 10.7) cm3 0.65–42.3 (mean, 12.8) cm3 Measuring tumor extent MRI tumor delineation. To serve as reference, the primary tumor volume was manually delineated on the MRIs on STIR TSE axial series. For correlation, standard axial T1weighted and contrast-enhanced T1 3D THRIVE images were also available to select the region of interest designated as tumor. In prior studies, showing a prognostic significance of tumor volume, all measurements are performed on pretreatment MRI and CT images and therefore these volumes are considered to be the ‘‘gold standard.’’ In this study, tumor volume was measured by multiplying the total area of the outlined tumor on every slice with the slice thickness. One observer (F.P.), an experienced head and neck radiologist (>15 years), performed all measurements. Images were seen in random order. The observer was blinded for the results of the semi-automatically derived tumor volumes. Technique for semi-automated derived gross tumor volume A previously developed viewing station that permits simultaneous viewing of 2 linked series in 3 orthogonal directions was used.16 The system visualizes all acquired series (precontrast and 79 series postcontrast) as well as subtractions of all series. The viewing station was also used for semi-automated segmentation of head and neck tumors at DCE-MRI. The required user interaction involves indicating 1 single point in or near the location of the center of the tumor in the MRIs. Subsequently, the tumor is segmented in 3D resulting in a volumetric measurement of extent. After segmentation, visual inspection of the result was performed to determine whether the segmentation correctly encompassed the enhancement in the tumor. After approval of the segmentation result, the volume calculated from the segmented tumor volume was stored. DCE-MRI VOLUME MEASUREMENT IN HEAD AND NECK CANCER FIGURE 1. Illustration of a semi-automatic segmentation of a T3 base of tongue carcinoma in a 45-year-old patient. The top row shows the uptake images (subtraction of postcontrast after 45 seconds and precontrast) in which the user identifies the location of the tumor. By selecting a point in 1 view (cross hairs), the other views are updated to maintain correspondence in 3 directions. The bottom row shows the result of the segmentation for visual inspection in the corresponding planes. Semi-automated segmentation of dynamic contrast-enhanced MRI The method used for semi-automated segmentation of head and neck tumors in MRIs was adopted from previous work.13 In short, after manual selection of a seed point in or near the center of the tumor, the variance of the image intensities over time is computed at each voxel location, thus yielding a processed image in which the contrast of the tumor is increased relative to the background. Then, an ellipsoidal region of interest is automatically constructed around it, and volume growing is applied within the region of interest using a voxel valuebased stop criterion that is automatically derived from the enhanced tumor volume (Figure 1). Evaluation Linear regression analysis was used to assess the precision of volumetric tumor extent with respect to the extent obtained by manual delineation. Pearson correlation was used to test for significant associations. Dynamic contrast-enhanced measured tumor volume For the total group of 21 patients, the mean volume of the primary tumor was 12.78 6 12.23 cm (median volume, 7.79 cm; range, 0.65–42.3). Pearson correlation In 90.5% of the patients (19 of 21), correlation could be made between the semi-automatically derived volume and the manually derived GTV (Figure 2). In 2 cases, no reasonable visible agreement could be achieved (Figure 3). The first case was a patient with a recurrent carcinoma of the base of tongue. Clinically, a large necrotic tumor mass, with diameter of 4.5 cm, was seen. The second case was a T4N2c tongue carcinoma. Clinically, a large ulcerative tumor, with a diameter of 5 cm, was seen. Nineteen cases were therefore used for analysis, resulting in a significant correlation coefficient (R2 ¼ 0.95; p < .001; Figure 4). DISCUSSION RESULTS Synopsis of new findings In 3 excluded cases, the primary tumor was not totally covered within the craniocaudal field of view in the dynamic series. Therefore, semi-automated tumor volume measurement was not possible. This study shows that semi-automated assessment of tumor volume using DCE-MRI agrees well (R2 ¼ 0.95) with manually derived GTV. To the best of our knowledge, no prior studies have reported on the feasibility of semi-automated tumor volume measurements by DCEMRI in patients with head and neck cancer. Primary tumor volume has been proven to be an important prognostic indicator for a variety of head and neck squamous cell carcinomas. Unfortunately, in daily practice, tumor Manually derived gross tumor volume For the total group of 21 patients, the mean manually derived GTV of the primary tumor was 10.7 6 11.22 cm (median volume, 6.59 cm; range, 0.50–42.79). HEAD & NECK—DOI 10.1002/HED APRIL 2013 523 LODDER ET AL. FIGURE 2. Example of semi-automatic segmentation of primary tumor volume on dynamic contrast-enhanced MRI (DCE-MRI) with good visual agreement. (A) Axial postcontrast MR image at the level of the base of the tongue (same patients as in Figure 1). The tumor (T) involves the right tongue base with extension to the right lateral pharyngeal wall. In addition, there is extension over the midline to the left side. The uninvolved free edge of the epiglottis is slightly deviated to the left (long arrow). Short arrows: enhancement of normal lymphoid tissue at the left tongue base. (B) Auto-segmentation after placement of seed point. Note the delineation program automatically includes tumor enhancement, but excludes physiologic enhancing lymphoid tissue at the left tongue base. (C) Manual tumor delineation. volume measurements are performed infrequently probably due to the extra workload involved. Comparisons with other studies In a 2004 literature overview, van den Broek et al1 showed that the reported mean primary tumor volumes vary within the different T classifications. For a certain primary tumor volume, local control rates vary between different tumor locations. Only a few authors report on primary tumor volume.1–8 These studies showed that volume is an independent predictor for local and regional control. DCE-MRI allows the study of microcirculation of tumors and normal tissues. Enhancement of a tissue depends on a number of factors, including vascularity, capillary permeability, renal clearance and volume, and composition of the extracellular fluid.14 Based upon the differences in enhancement of body tissues, semi-automated measurements can be performed. At present, DCEMRI is frequently applied in diagnostic MRI protocols among others in the head and neck. Fischbein et al17 showed the different values for time-to-peak, peak enhancement, maximum slope, and washout slope for different tissues (nontumor node, sternocleoid mastoid muscle, tumor node, and submandibular gland). In 2007, Alderliesten et al13 studied the correlation between DCE-MRI measurements and histopathology in 43 patients with breast cancer. With R2 ¼ 0.84, they concluded that semi-automatic volumetric measurement of breast cancer accurately describes tumor extent. In this current study, we found a correlation of R2 ¼ 0.95. A possible explanation for this difference in Pearson correlation could be the difference in the gold standard used. A lower correlation coefficient between radiological and histopathological volume can be anticipated due to tissue under sampling and other technical limitations at histopathology. In 2010, Hadjiski et al15 described the feasibility of computerized segmentation of lesions on head and neck CT scans and evaluated its potential for objective tumor volume estimation in response to treatment. The computer-estimated change in tumor volume and percentage change in tumor volume between the pretreatment and posttreatment scans achieved a high correlation (intraclass correlation coefficient ¼ 0.98 and 0.98, respectively) with the estimates from manual segmentation in 13 primary tumors. These values approach our current study results. However, we think that MRI with its superior soft tissue contrast may be more suitable for semiautomated volume measurements. Study limitations Some study limitations should be discussed. Only 1 observer delineated the primary tumor. Knegjens et al2 studied interobserver variability in 50 cases. The mean difference in delineated primary tumor volume was 7.6% (range, 0.15% to 20.9%). This difference was constant across tumor volume. The Pearson correlation coefficient FIGURE 3. Example of semi-automatic segmentation of primary tumor volume on dynamic contrast-enhanced MRI (DCE-MRI) with bad visual agreement. (A) Axial contrast-enhanced MR image of a T3 tongue carcinoma (T). (B) Result of automatic segmentation after placement of the seed point. (C) Manual tumor delineation. 524 HEAD & NECK—DOI 10.1002/HED APRIL 2013 DCE-MRI FIGURE 4. Automated versus manual tumor volume measurements. Agreement between automatically and manually derived tumor volumes from dynamic contrast-enhanced MRI (DCE-MRI; R2 ¼ 0.95). was 0.99, which implies good agreement between the observers. In 1997, Rasch et al18 showed that MRIderived GTVs have less interobserver variation than CT-derived GTV. Furthermore, all studies showing the prognostic significance of tumor volume used delineated volumes determined on radiology. Therefore, we believe that MRI delineation by 1 observer may yield representative results in the current study. Also, 1 observer performed the placement of the seed point. In theory, the location of the seed point should not influence the segmented volume by much, because the technique automatically shifts the seed point to the approximate center of mass of enhancement before volume growing. However, we cannot exclude possible bias. Therefore, in future research the interobserver and intraobserver variation should be studied. A total of 24 patients were selected for semi-automated measurement of the primary tumor volume. In total, in 19 patients, the primary tumor volume could be assessed and compared with the manually derived GTV. In 3 cases, the primary tumor volume was not totally covered within the craniocaudal field. This was caused by an improper selection of the field of interest by the MRI physician. In 2 additional patients, no visual agreement was reached between DCE-MRI and tumor delineation on MRI. Several reasons exist for visual disagreement. First, the segmentation method was optimized for breast cancer. In future work, the system will be optimized to head and neck cancer. Another explanation is heterogeneity in contrast uptake in the tumor (eg, due to necrosis). Another limitation of our study is that most of the included patients had a primary untreated head and neck cancer; in our consecutive series only 1 patient had recurrent disease. Radiotherapy or carotid artery embolization may change the contrast enhancement of tissues, thus influencing the performance of the semi-automated segmentation. Clinical applicability of the study Currently, standard methods for tumor volume assessment (estimation of tumor volume based on the cubic relation with largest tumor diameters [volume ¼ a b c)19,20 VOLUME MEASUREMENT IN HEAD AND NECK CANCER or the ellipsoidal relation (volume ¼ 1/6 p a b c)21,22 as well as manual delineation of volume on CT or MRI, are time consuming and have limited reproducibility. The mean time required to perform a volume measurement using the summation-of-areas technique is approximately 10 minutes for experienced radiologists, which is impractical. Using the reported approach, it may be possible to assess tumor volume on a routine basis within 2 minutes. The only required user interaction involves indicating 1 single point in or near the center of the tumor on the MRI. Also, semi-automated segmentation may be useful for radiation therapy planning and treatment-response measurements. In the future, the reported method may allow for improved international standardization of tumor volume measurements in patients with head and neck cancer. In cases in which DCE-MRI was not possible due to previous irradiation/ treatment, contraindications for intravenous contrast, or in cases with no reasonable visible agreement, tumor delineation remains the gold standard. CONCLUSION Semi-automated measurements of tumor volume on DCE-MRI were representative of those derived from manual delineation (R2 ¼ 0.95; p < .001) in a group of 19 patients with head and neck cancer. When these results are validated in a larger prospective study, they could be implemented in routine practice. Acknowledgements We thank Dr. S. Muller and A. Paape for their assistance with the dynamic MRI protocol and we thank Josien de Bois for her help with the measurement of tumor volumes. REFERENCES 1. van den Broek GB, Rasch CR, Pameijer FA, et al. Pretreatment probability model for predicting outcome after intraarterial chemoradiation for advanced head and neck carcinoma. Cancer 2004;101:1809–1817. 2. Knegjens JL, Hauptmann M, Pameijer FA, et al. Tumor volume as prognostic factor in chemoradiation for advanced head and neck cancer. Head Neck 2011;33:375–382. 3. Hermans R, Op de beeck K, Van den Bogaert W, et al. The relation of CTdetermined tumor parameters and local and regional outcome of tonsillar cancer after definitive radiation treatment. Int J Radiat Oncol Biol Phys 2001;50:37–45. 4. Pameijer FA, Mancuso AA, Mendenhall WM, et al. Evaluation of pretreatment computed tomography as a predictor of local control in T1/T2 pyriform sinus carcinoma treated with definitive radiotherapy. Head Neck 1998;20:159–168. 5. Hermans R, Van den Bogaert W, Rijnders A, Baert AL. Value of computed tomography as outcome predictor of supraglottic squamous cell carcinoma treated by definitive radiation therapy. Int J Radiat Oncol Biol Phys 1999;44:755–765. 6. Pameijer FA, Mancuso AA, Mendenhall WM, Parsons JT, Kubilis PS. Can pretreatment computed tomography predict local control in T3 squamous cell carcinoma of the glottic larynx treated with definitive radiotherapy? Int J Radiat Oncol Biol Phys 1997;37:1011–1021. 7. Doweck I, Denys D, Robbins KT. Tumor volume predicts outcome for advanced head and neck cancer treated with targeted chemoradiotherapy. Laryngoscope 2002;112:1742–1749. 8. Mendenhall WM, Parsons JT, Mancuso AA, Pameijer FJ, Stringer SP, Cassisi NJ. Definitive radiotherapy for T3 squamous cell carcinoma of the glottic larynx. J Clin Oncol 1997;15:2394–2402. 9. Edge SB, Byrd DR, Compton CC, Fritz AG, Greene FL, Trotti A, editors. American Joint Committee of Cancer – AJCC Cancer Staging Manual. New York: Springer; 2010. 10. Studer G, Lütolf UM, El–Bassiouni M, Rousson V, Glanzmann C. Volumetric staging (VS) is superior to TNM and AJCC staging in predicting outcome of head and neck cancer treated with IMRT. Acta Oncol 2007;46: 386–394. HEAD & NECK—DOI 10.1002/HED APRIL 2013 525 LODDER ET AL. 11. Brown J, Buckley D, Coulthard A, et al. Magnetic resonance imaging screening in women at genetic risk of breast cancer: imaging and analysis protocol for the UK multicentre study. UK MRI Breast Screening Study Advisory Group. Magn Reson Imaging 2000;18:765–776. 12. Barentsz JO, Jager GJ, Witjes JA, Ruijs JH. Primary staging of urinary bladder carcinoma: the role of MRI and a comparison with CT. Eur Radiol 1996;6:129–133. 13. Alderliesten T, Schlief A, Peterse J, et al. Validation of semiautomatic measurement of the extent of breast tumors using contrast-enhanced magnetic resonance imaging. Invest Radiol 2007;42:42–49. 14. Shah GV, Fischbein NJ, Gandhi D, Mukherji SK. Dynamic contrastenhanced MR imaging. Top Magn Reson Imaging 2004;15:71–77. 15. Hadjiiski L, Mukherji SK, Ibrahim M, et al. Head and neck cancers on CT: preliminary study of treatment response assessment based on computerized volume analysis. AJR Am J Roentgenol 2010;194:1083–1089. 16. Gilhuijs KG, Deurloo EE, Muller SH, Peterse JL, Schultze Kool LJ. Breast MR imaging in women at increased lifetime risk of breast cancer: clinical system for computerized assessment of breast lesions initial results. Radiology 2002;225:907–916. 526 HEAD & NECK—DOI 10.1002/HED APRIL 2013 17. Fischbein NJ, Noworolski SM, Henry RG, Kaplan MJ, Dillon WP, Nelson SJ. Assessment of metastatic cervical adenopathy using dynamic contrastenhanced MR imaging. AJNR Am J Neuroradiol 2003;24:301–311. 18. Rasch C, Keus R, Pameijer FA, et al. The potential impact of CT-MRI matching on tumor volume delineation in advanced head and neck cancer. Int J Radiat Oncol Biol Phys 1997;39:841–848. 19. Grabenbauer GG, Steininger H, Meyer M, et al. Nodal CT density and total tumor volume as prognostic factors after radiation therapy of stage III/IV head and neck cancer. Radiother Oncol 1998;47:175–183. 20. Chufal KS, Rastogi M, Srivastava M, Pant MC, Bhatt ML, Srivastava K. Analysis of prognostic variables among patients with locally advanced head and neck cancer treated with late chemo-intensification protocol: impact of nodal density and total tumor volume. Jpn J Clin Oncol 2006;36:537–546. 21. Dunst J, Stadler P, Becker A, et al. Tumor volume and tumor hypoxia in head and neck cancers. The amount of the hypoxic volume is important. Strahlenther Onkol 2003;179:521–526. 22. Kuhnt T, Mueller AC, Pelz T, et al. Impact of tumor control and presence of visible necrosis in head and neck cancer patients treated with radiotherapy or radiochemotherapy. J Cancer Res Clin Oncol 2005;131:758–764.