Tel.: +1-215-615-0854, Fax: +1-215-898-4469
Molecular imaging assisted surgery improves survival in a murine head and neck cancer model
Article first published online: 14 DEC 2011
Copyright © 2011 UICC
International Journal of Cancer
Volume 131, Issue 5, pages 1235–1242, 1 September 2012
How to Cite
Miyamoto, S., Sperry, S., Yamashita, T., Reddy, N. P., O'Malley, B. W. and Li, D. (2012), Molecular imaging assisted surgery improves survival in a murine head and neck cancer model. Int. J. Cancer, 131: 1235–1242. doi: 10.1002/ijc.27328
- Issue published online: 27 JUN 2012
- Article first published online: 14 DEC 2011
- Accepted manuscript online: 27 OCT 2011 06:31AM EST
- Manuscript Accepted: 5 OCT 2011
- Manuscript Received: 24 MAY 2011
- head and neck cancer;
- surgical safety margin;
- molecular imaging;
- fluorescent protein
Surgery plays an important role in the treatment of head and neck cancer (HNC), and surgical margin status is a key prognostic factor. Molecular imaging (MI) can be applied to identify tumor extensions intraoperatively. We applied this technique in a murine HNC model to determine whether it improves outcomes from surgical intervention. An orthotopic murine model with HNC was established with SCC VII cells expressing a green fluorescent protein. To determine the diagnostic accuracy of MI, 20 murine models undergoing standard surgical resection were assessed with MI to identify residual tumor, which was compared to histology as the gold standard. Then, to assess the effect of MI as a therapeutic intervention for survival, 65 mice were randomly divided into standard surgical resection, MI-assisted surgical resection, and control groups. In the MI-assisted surgery group, residual signals identified by MI underwent further tissue excision to eliminate the signal positivity. In diagnostic accuracy analysis, sensitivity and specificity of intraoperative MI in the HNC murine model were 86% and 100%, respectively. The mice undergoing MI-assisted surgery showed a significantly improved 60-day survival rate compared to standard surgery, 37% versus 5%, respectively. Intraoperative MI guidance is a promising technique in oncologic surgery, which could increase the efficacy of tumor resection and the survival of patients with HNC. The hurdles in applying this technique in clinical practice are still considerable, and further research and development is warranted.
Head and neck cancer (HNC) that arises in the nasal cavity, paranasal sinuses, oral cavity, pharynx and larynx is the sixth most common type of cancer, representing about 6% of all cases in the world.1 In the USA, ∼56,000 new HNC cases have been reported every year.2 The overall survival of HNC patients has remained largely unchanged at 50–60% in the past 30 years despite the development of diagnosis and treatment.3 During this time period, the National Cancer Data Base demonstrated a decrease in the survival of laryngeal cancer patients that parallels the observed trend of increasing use of nonoperative therapy.4 Therefore, primary surgery continues to play an important role in the treatment of HNC,5, 6 although approximately two-thirds of HNC patients are diagnosed as an advanced stage at the initial presentation. In advanced disease, primary surgery is often followed by radiotherapy and/or chemotherapy.6, 7 Despite the intensive multimodality treatment, locoregional recurrences develop in 30–40% of patients and the 3-year disease-free survival rate is 35–55%.8 The rate of positive and close surgical margin after surgery in HNC is considerable, reported to be 3–23% and 25–43%, respectively.9, 10 It has been well known that the surgical margin status is closely associated with the prognosis.11 Distant metastases at initial presentation are uncommon, presenting in about 10% of HNC patients. However, local or regional recurrence and distant metastases are frequent after surgical treatment of advanced disease, particularly in patients with inadequate resection margins and extranodal spread.7, 8 Thus, the surgical margin status of the initial surgery for locoregional tumors might be critical for prognosis, and the improvement of the likelihood of complete tumor resection could increase the disease-free survival of HNC.
In recent years, noninvasive molecular imaging (MI), such as fluorescent imaging (FLI) and bioluminescent imaging (BLI), has been adapted to perform imaging of optical reporters in living animals.12, 13 The MI technique is cost-effective, portable, and radiation-free compared to conventional diagnostic imaging modalities, such as computed tomography (CT), magnetic resonance imaging (MRI), and positron-emission tomography.14 FLI offers advantages over BLI. The main advantage of FLI is that it does not require injection of substrate into the animal to visualize tumor cells before imaging.15 The light intensity of bioluminescence is variable, because the catalytic reaction that generates the signal is time and enzyme dependent.16 The peak of bioluminescence ranges from 10 to 15 min after substrate injection,17 so that it is time-consuming as well as variable. On the other hand, light intensity of fluorescence is relatively stable and does not require a substrate for imaging. In addition, the standard substrate of bioluminescence, luciferin, is reported to carry some toxicity.18 Thus, FLI has more potential for clinical application. The main disadvantage of FLI is high background noise due to tissue autofluorescence, mainly from the skin and hair.12 Although fluorescent proteins have allowed us to visualize, in real time, important aspects of cancer cells in vivo, including tumor cell mobility, invasion, metastasis, and angiogenesis,15 MI in living animals has been limited by autofluorescence and illuminating light attenuation by absorbance and scattering from surrounding tissue, which can prevent detection of low-intensity signals in the visible range.13, 19 However, these limitations would be diminished in the surgical setting where an open surgical wound would reduce the tissue depth that needs to be penetrated.
FLI is commonly macroscopic, which enables relatively large tissue areas to be scanned. On the other hand, MI utilizing fluorescence includes different rising methods, such as confocal microscopy and optical coherence tomography. These optical techniques provide high resolution identifying cellular or subcellular structures.14, 20 Meanwhile, it is difficult for them to scan an entire field-of-view in human surgical wound. These microscopic MI are also high-potential techniques, however, macroscopic FLI was used as MI in this study, because relatively large areas including primary lesion and regional lymph nodes (LNs) would be commonly screened in head and neck surgery.
MI is a promising technique with possible application in intraoperative guidance in oncologic surgery, with the goal of increasing the efficacy of tumor resection by identifying tumor extensions which otherwise would not be excised in conventional surgery.14 A MI technique has already found application in oncologic surgery, mainly for sentinel LN mapping in breast, skin, gastric, colorectal, oropharyngeal cancer and melanoma.21–23 However, the application of MI to intraoperative guidance for primary tumor extirpation is still limited.
The application of MI for primary tumor extirpation has been reported only in brain and liver tumors, in which survival was not estimated.24, 25 On the other hand, there are many reports of intraoperative MI guidance for tumor resection in the experimental stage.26–29 However, there is only one report investigating survival as a treatment outcome using MI, in which an ectopic model was used.30 All of reported surgical outcomes from these studies vary from each other because of the inconsistent biomarkers and labeling techniques for targeted tumor cells. As the targeted tumor cells cannot be maximally labeled, there has been no data to demonstrate the maximal benefits from MI-assisted oncological surgery when looking at the tumor survival.
Therefore, in this study, we created an orthotopic animal model of HNC with a stable tumor cell line expressing fluorescence signals, in which, nearly all of the tumor cells can be consistently detected and monitored with MI and quantified with imaging software. We hypothesized that intraoperative MI guidance might improve treatment outcome in HNC surgery by increasing the reliability of a complete tumor resection. To prove this hypothesis, we applied this technique during surgery in an animal model of HNC to evaluate the accuracy of the method and the survival in mice.
Materials and Methods
The SCC VII is a murine squamous cell carcinoma cell line originating spontaneously in a C3H mouse.31 Green fluorescent protein (GFP) is widely used to visualize cancer in living animals,15 so it was selected as a candidate probe to establish a stable fluorescence expressing cell line. TdTomato, a red fluorescent protein, was another candidate, because it has reduced interference from autofluorescence. A GFP gene was transduced into the cells with a lentiviral vector, GFP lentiviral particles (GenTarget, San Diego, CA), according to the manufacture's protocol (SCC VII-GFP). SCC VII cells transduced with tdTomato gene (SCC VII-tdTomato) were also established by retroviral transduction; briefly, pBabe-Blast-tdTomato plasmid was added to Phoenix Eco and Phoenix Ampho retroviral packaging cell lines (ATCC, Manassas, VA) in FuGENE HD transfection media (Roche, Indianapolis, IN). After 72 hr of incubation, the supernatant was added to SCC VII cells for transduction of tdTomato gene. SCC VII-GFP and SCC VII-tdTomato cells were cultured in RPMI-1640 medium (Lonza, Walkersville, MD) supplemented with 10% fetal bovine serum (Sigma–Aldrich, St. Louis, MO) and 1% Penicillin–Streptomycin (Invitrogen, Grand Island, NY) at 37°C and 5% CO2. The fluorescent and bright field photographs of cell cultures were captured with a Nikon ECLIPSE TS100 microscope (Nikon, Tokyo, Japan) equipped with a Nikon Coolpix 4500 digital camera to compare the brightness of both cells.
In vitro MI
The brightness of SCC VII-GFP and SCC VII-tdTomato was compared using Kodak In-Vivo Imaging System FX Pro (Carestream, Rochester, NY). Each stable cell line was seeded onto three wells of a 96-well culture plate at 106, 105 and 104 cells per well. A fluorescence image of the plate was captured with 5 sec of exposure using excitation and emission filters of 480 nm and 535 nm wavelength for SCC VII-GFP, and 550 nm and 600 nm for SCC VII-tdTomato, respectively. The fluorescence image was overlaid on the grayscale image of the same view. To measure the amount of fluorescent signal, an equal-sized region of interest (ROI) was set on each well, and the photon intensity of the ROI was quantified using supplemental software.
A murine HNC model using SCC VII cells and the C3H mouse was first described in 1997.32 This orthotopic model is considered to imitate locally advanced HNC in humans, because it demonstrates initial locoregional aggressiveness due to poor immunogenicity.33 We established the murine HNC model with GFP expressing cells to visualize tumor cells in animal tissues. Briefly, 6- to 8-week old female C3H/HeJ mice (Jackson Laboratory, Bar Harbor, ME) were inoculated with 5 × 105 SCC VII-GFP cells in 100 μL of phosphate buffered saline (PBS) (no calcium and no magnesium) subcutaneously into the floor of the mouth (FOM) with a 28-gauge needle. All animal procedures were conducted in accordance with the guidelines of the University of Pennsylvania School of Medicine Institutional Animal Care and Use Committee (IACUC).
In vivo MI
After tumor resection, the HNC murine models were placed on the chamber of the MI system. Multispectral fluorescence images were captured for each mouse with 2 sec of exposure using excitation filters of 420, 430, 440, 450, 460, 470 and 480 nm wavelengths and an emission filter of 535 nm wavelength. The images were unmixed using the supplemental software to minimize background fluorescence. The unmixed images were overlaid on the grayscale images of the same view to locate tumors to determine the localization of tumors in the surgical field. Equal-sized ROIs were applied to all the mice to quantify the photon intensities of surgical fields.
Diagnostic accuracy of intraoperative MI
In this assay, 20 mice were used (Fig 1a for experiment design). Following general anesthesia with intraperitoneal injection of 2,2,2-tribromoethanol (Sigma–Aldrich, St. Louis, MO) at a dosage of 0.4–0.5 mg/g and I-shaped skin incisions over tumors, standard total tumor resection was performed on all mice by a head and neck surgeon with the unaided eye 6 days after tumor cell inoculation, when the size of the tumors had grown to ∼7 mm in diameter. The intended surgical margin was only 1–2 mm, because the head and neck region of mice is compact and contains a variety of important structures, including the carotid arteries, jugular veins, and trachea. One mouse expired from excessive blood loss during the operation. Immediately after the standard total tumor resection, the surgical field was examined with MI. The mice with obvious foci of fluorescence signal were instantaneously judged positive in residual signal. To estimate the validity of the instantaneous judgments of residual signals, each postoperative image was obtained with a negative control mouse, and then the relative photon intensities of the ROIs in the operated mouse versus the control mouse were calculated. The skin incisions were closed by 4-0 nylon sutures. Histological analysis with hematoxylin and eosin (H&E) stain, which is gold standard for the diagnosis of HNC, was conducted every 0.5 mm on the FOM tissue including muscles, salivary glands and soft tissues around the jugular veins to determine accurately whether tumor cells remained or not. However, when surgery is intended complete resection, it is difficult to provide negative proof such as the absence of any minimal residues of tumor cells immediately after surgery, because specimens are not truly serial-sectioned. New tumorigenesis in 1 month is practically impossible and tumor regrowth reflects the remaining of tumor cells, so that we put an observation period for potentially residual tumors to reduce histological false negatives: either on the day when the mice revealed obvious swellings in the FOM or 28 days after tumor cell inoculation, if no swelling occurred. According to the results from the MI guidance and the histological analysis, the diagnostic accuracy of intraoperative MI in the orthotopic murine model was calculated, including sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).
In this assay, 65 mice were used (Fig. 1b for experiment design). Seven days after tumor cell inoculation, all the mice were randomly divided into two groups: 55 mice which underwent standard total tumor resection by two head and neck surgeons as described above and 10 mice without treatment. The average tumor size was ∼9 mm in diameter. Because five mice expired from either blood loss or anesthesia side effects during or after the surgery, 50 mice were subsequently randomized to two groups; 30 mice underwent intraoperative MI (MI-assisted surgery group), and 20 had no additional treatment (standard surgery group). The mice with positive fluorescence signal in the MI-assisted surgery group underwent an additional resection immediately after MI to diminish signal positive tissues (signal (+) additional resection group), whereas the mice without fluorescence signal had no additional treatment (signal (−) group). The tissues removed in the additional resection were examined with histological analysis to verify the presence of tumor cells. The photographs of specimens were captured with a Nikon ECLIPSE 80i microscope. After the tissue resection was completed, the skin incisions were closed with 4-0 nylon suture. All mice were observed for up to 60 days following tumor cell inoculation to determine survival. The survival of the MI-assisted surgery group was compared with that of the standard surgery group and the no treatment group, to evaluate the impact of intraoperative MI on tumor resection. The survival of the two subgroups in the MI-assisted surgery group, that is the signal (+) additional resection group and the signal (−) group, were compared with that of the standard surgery group to investigate the effect of the additional tumor resection as guided by MI. Tumor-specific survival rates were used in this analysis, because it was difficult to detect postoperative tumor recurrence early as well as the timing of detection was variable due to scar formation in the surgical area. In accordance with IACUC guidelines, the mice with tumors reaching 20 mm in maximum diameter or cachexia were euthanized and were considered to be tumor-specific deaths in this study.
The comparison of fluorescence photon intensities in the ROI was evaluated using the Mann–Whitney U test. The survival curves of mice in the treatment groups were modeled by the Kaplan–Meier method. Differences in survival between treatment groups were evaluated using the logrank test. The Cox regression method was used to estimate the hazard ratio (HR). The statistical analysis and generation of survival curves were performed with PASW statistic software (Version 18; IBM, Somers, NY). p values less than 0.05 were considered significant.
Comparison of fluorescence in tumor cell lines
The brightness of fluorescence in SCC VII-GFP and SCC VII-tdTomato cells was compared to select an appropriate tumor cell line for the study. The expression rates of fluorescence in SCC VII-GFP and SCC VII-tdTomato cells were confirmed to be almost 100% with fluorescent microscopy. The fluorescence photon intensity of the two cell lines as measured with MI was compared. Fluorescence quantification with our system showed that fluorescence was roughly proportional to the number of cells, and that SCC VII-GFP was 25–90 times brighter than SCC VII-tdTomato in vitro (Table 1). Because there was a large difference in the brightness of fluorescence between these cell lines, we selected SCC VII-GFP as the probe to visualize tumor cells in the murine HNC model.
Assessment of residual signals with MI
Murine HNC models with GFP expressing cells were established, and tumors were successfully visualized by MI (Figs. 2a and 2b). By comparing to the negative control, 19 mice undergoing tumor resection were assessed whether they were positive or negative in residual signal, and then the relative photon intensities were calculated (Fig. 2c). Twelve mice were judged positive, whereas seven were negative. The relative photon intensity in the 12 positive mice theoretically must be greater than 1; the measured minimum value was actually 2.01, although the measured positive values had a wide variability (mean ± standard deviation 26.9 ± 50.0) due to large amount of residual tumor in some cases. On the other hand, the relative photon intensity in the seven negative mice was 1.02 ± 0.47 [95% confidence interval (CI) 0.56–1.46]. There were significant differences between those of the positive and negative mice with the Mann–Whitney U test (p < 0.001). These results demonstrate that the operator's assessment of the intraoperative MI correlated with the measurable photon intensity, and we can judge residual signal foci as residual tumors.
Diagnostic accuracy of intraoperative MI
In the 19 mice as discussed in the last section, FOM tissues were analyzed histologically to evaluate the accuracy of intraoperative MI guidance (Table 2). Sensitivity and specificity were 86% and 100%, with PPV and NPV of 100% and 71%, respectively. In addition, the false positive and false negative rates were 0% and 14%, respectively. As a result, intraoperative MI proved to be effective in detecting residual tumors that were missed with the unaided eye.
Histological findings of the additional resection specimens
In the survival analysis, the tissues removed in the additional resection following MI guidance were analyzed histologically to investigate the presence of residual tumor that was overlooked with the unaided eye. Twenty-two specimens from 18 mice were examined, and 19 of them were histologically positive in tumor cells (86.4%). Most specimens showed salivary gland (Fig. 3a) or muscle invasion of tumor cells (Fig. 3b), and only 2 showed LN metastases (Fig. 3c). These findings are only 7 days after the tumor cells inoculation, reflecting the aggressiveness of SCC VII cells. MI detected microscopic tumor lesions measuring less than 1 mm (Fig. 3d).
MI-assisted surgery improved survival in the HNC murine model. Sixty days after tumor cell inoculation, the tumor-specific survival rates of the MI-assisted surgery group, standard surgery group and no treatment group were 37%, 5% and 0%, respectively (Fig. 4a). The survival rate of the MI-assisted surgery group was significantly improved compared with that of the standard surgery group (HR 0.25, 95% CI 0.13–0.49, p < 0.001 by logrank test) as well as that of the no treatment group (HR 0.20, 95% CI 0.09–0.43, p < 0.001). This result suggested that intraoperative MI was more reliable than the unaided eye to confirm complete tumor resection. The survival rates of the two subgroups in the MI-assisted surgery group (i.e., the signal (+) additional resection group and the signal (−) group) were 33% and 42%, respectively (Fig. 4b). Survival in the signal (+) additional resection group was significantly improved compared with the standard surgery group (HR 0.28, 95% CI 0.13–0.60, p < 0.001), as well as that in the signal (−group versus the standard surgery group (HR 0.51, 95% CI 0.32–0.80, p = 0.001). Moreover, a comparison of those two subgroup showed no significant difference in survival (HR 1.12, 95% CI 0.47–3.04, p = 0.70). These results suggest that the improved survival rate of the MI-assisted surgery group was attributable to the additional resection of macroscopically undetectable tumors that, if otherwise left untreated, would cause tumor-related death.
Although the surgical margin status in HNC is associated with prognosis,11 conventional tumor resection is limited in its ability to discriminate invasive tumor cells from surrounding normal tissues when performed with the unaided eyes. Surgeons have traditionally relied on their senses developed through experience, such as visual inspection and palpation, to detect tumors and determine the extent of resection during surgery.34 However, this often results in easily missing invisible or microscopic tumors, skip lesions and micrometastases of LNs. It is difficult even in microscopic surgery to distinguish microscopic lesions from surrounding tissues without histological analysis. Frozen-section examination has been the gold standard for performing rapid microscopic analysis of a specimen in oncological surgery. However, the quality of the slides produced by frozen section is of lower quality than formalin fixed, wax embedded tissue processing.35, 36 In the case, invisible skip lesions or LN metastases are located outside the scheduled surgical margin or extent of LN dissection, it is impossible to detect them by the conventional method. Moreover, the minimal lesions in the specimens, although they are resected, could be overlooked in the frozen-section examination, because multiple sections of tissue are inaccurate for intraoperative rapid diagnosis. Therefore, the accuracy diagnosis of frozen-section examination can be compromised. Our study indicates that intraoperative MI guidance is a promising technique in oncologic surgery, which could increase the efficacy of tumor resection and therefore improve treatment outcome. We believe that combined MI with frozen-section examination will significantly improve the surgical outcome for HNC.
Even permanent histological analysis has the propensity to miss microscopic tumors, because specimens are not entirely serial-sectioned and generally sectioned every 1 to several mm at a minimum.37, 38 Accordingly, surgeons have to take sufficient safe surgical margins and often perform regional LN dissections whether or not those actually include known tumor cells to remove undetected tumor cells. Nevertheless, head and neck surgery is frequently limited by the proximity of vital structures, including carotid arteries, cranial nerves and cranial bone, to the tumor.7 Excessive resection of head and neck structures, such as the tongue, pharynx and larynx, causes a function loss of phonating and swallowing. These limitations might be associated with the high incidence of positive margins in surgical resections and locoregional recurrence, which lead to poor prognosis. Intraoperative MI might overcome these limitations by increasing the rate of complete tumor resection and avoiding excess removal of normal tissue.
The sensitivity and specificity of intraoperative MI to detect tumor in the orthotopic murine HNC model were high, similar to the reported rates of 86% and 100%, respectively, in murine xenograft models.28 As reported previously,34 this technique could identify a microscopic tumor less than 1 mm in size in this study. These results indicate that this technique is worth further investigations for intraoperative guidance for tumor detection.
The survival rate of mice treated with intraoperative MI guidance was significantly improved compared with that of mice operated on with the unaided eye, suggesting that the detection of invisible residual tumor cells with MI increased the thoroughness of total tumor resection. The survival benefits of intraoperative MI guidance in an animal model were previously reported.30 However, we have confirmed similar results in an orthotopic model. Although the 60-day survival rate of the MI-assisted surgery group was just 37%, it was considered to be a remarkable improvement, given that this locally advanced tumor model was treated only with surgery and surgical margins were insufficient. The survival analysis of the subgroups demonstrated that the benefits of the additional resection of tumor directly associated with significantly improved tumor survival. The signal (+) additional resection group would likely have developed disease recurrence with subsequent decreased survival without the salvage step of additional tumor resection. It is also remarkable that the survival rate of the signal (+) additional resection group showed no difference compared with that of the signal (−) group. These indicate that intraoperative guidance with MI could redeem margin status, even if the initial definitive resection failed.
Because of the limitations of conventional surgery and intraoperative frozen-section examination, they should be assisted with other techniques. The intraoperative MI technique might have huge potential impact, particularly for advanced stage disease that tends to develop irregular invasion into surrounding tissue, skip lesions and regional LN metastasis.
When surgery is the primary treatment in advanced disease, neck dissection is carried out as part of surgical management.7 The extent of neck dissections is currently based on the tumor origin site and the stage of disease,39, 40 and can carry varying amounts of risk for significant esthetic and functional morbidity.41 MI technique could be applied to customize the extent of the neck dissection, based on the invasive or metastatic tumor cells. MI could avoid the additional morbidity of over-resection of negative tissues in the neck, as well as the tongue, pharynx and larynx. In addition, surgery in advanced disease is commonly followed by radiation and/or chemotherapy, causing severe toxicity.7 MI-assisted surgery could reduce the toxicity by decreasing the dose and duration of the adjunctive treatments.
There are still considerable hurdles to overcome for the application of MI technique to clinical practice. First, MI is limited by a depth penetration, as well as a lower resolution relative to conventional imaging like CT and MRI.14 These limitations would be relieved by near-infrared fluorescence (NIRF) probes, including infrared-fluorescent protein, Cy5.5, indocyanine green (ICG) and quantum dots, because autofluorescence, absorption and scattering of light by tissues are greatly reduced at near-infrared wavelengths.42, 43 Gaining Food and Drug Administration (FDA) approval for clinical use of NIRF proteins and dyes will be a key step toward using MI for clinical applications, because currently ICG is the only FDA approved fluorescent dye.22, 42 And also, those limitations may be reduced by an increased sensitivity of a charge-coupled device camera, as well as combined application of microscopic MI. Second, specific biomarkers for tumors are necessary for targeted labeling of tumor cells with fluorescent probes. Conjugation of these probes to tumor-specific biomarkers can lead to highly selective fluorescence imaging of tumor tissue. There are candidate biomarkers for HNC including epidermal growth factor receptor (EGFR), vascular endothelial growth factor (VEGF), VEGF receptor and Hsp47.44–46 Actually, Rosenthal et al. labeled HNC cells with Cy5.5 using an anti-EGFR antibody in a mouse model.47 Nguyen et al. bound fluorescently labeled cell-penetrating peptides to tumor cells.30
In conclusion, diagnostic accuracy of MI is high to the point of increasing the reliability of a complete tumor resection. Intraoperative MI guidance is a promising new technique in oncologic surgery to improve the survival of HNC patients. However, the hurdles are still considerable in applying this technique to clinical practice, and further research and development is warranted.
- 2American Cancer Society. Learn about cancer, 2010. http://www.cancer.org/Cancer/index [Updated October 7, 2010].
- 3http://seer.cancer.gov/csr/1975_2007/index.html National Cancer Institute [Updated October 8, 2010]., , , . SEER cancer statistics review 1975–2007,
- 6National Comprehensive Cancer Network. NCCN guidelines, 2011. http://www.nccn.org/professionals/physician_gls/f_guidelines.asp [Updated March 30, 2011].