OBJECTIVE. The purpose of this study is to evaluate the CT densities of small head and neck mucosal cancers as a means of deriving a CT mucosal window display of narrower window width and higher window level to better detect and delineate head and neck carcinomas.
MATERIALS AND METHODS. We retrospectively studied 19 subjects with T1–2 head and neck carcinomas. The density of tumor and adjacent normal mucosa on CT were measured. CT scans for the 19 patients with tumors and 35 subjects without mucosal tumors were anonymized and interpreted by two readers using standard soft-tissue windows and were reviewed again 1 week later with the addition of mucosal windows.
RESULTS. The mean (± SD) attenuation of 17 visible tumors was 85.5 ± 18.3 Hounsfield units (HU) and that of the surrounding normal mucosa was 55.3 ± 15.2 HU (p < 0.0001). From our data, we derived guideline mucosal window settings—a window width of 120 HU and a window level of 60 HU. On blinded review, reader A detected 12 tumors with the addition of mucosal windows (sensitivity, 63%; specificity, 82%) and nine tumors on soft-tissue windows alone (sensitivity, 47%; specificity, 94%). Reader B detected nine tumors with use of mucosal windows (sensitivity, 47%; specificity, 71%) and eight tumors on soft-tissue windows alone (sensitivity, 42%; specificity, 74%).
CONCLUSION. Early T-stage tumors have higher CT density than normal mucosa. Their conspicuity can be amplified using display windows with narrower window width and higher window level. The potential clinical applications are for the improved detection of unknown primary tumors and delineation of a known mucosal tumor.
Primary mucosal head and neck squamous cell carcinomas (SCCs) can be difficult to identify on contrast-enhanced CT (CECT) scans of the neck when there is no exophytic component or clear evidence of deep invasion across fat planes. Small tumors are often clinically silent but may present with large cervical nodal metastases. When the primary tumor cannot be located, the patient is treated as having carcinoma of unknown primary (CUP) and will undergo more radical multimodality treatment [1, 2]. Before tumors are labeled as CUP, an extensive diagnostic workup is performed to try to identify the primary site in the upper aerodigestive tract. This workup may involve additional imaging tests, such as MRI or PET, and a more invasive endoscopic procedure with multiple biopsies and even tonsillectomy [1, 3]. This diagnostic workup, with its associated expense and potential hazards, might be averted if the primary mucosal tumor could be detected on the initial CT scan of the neck. The potential advantages to the patient are reduced medical expense, earlier treatment, and possibly more effective treatment.
When early T-stage head and neck SCCs are missed on CT, the most common reasons are dental metal artifacts, small size, and small differences in attenuation compared with surrounding mucosa [4, 5]. Regarding the last reason, increased enhancement of head and neck SCC may be extremely subtle and difficult to detect with routine soft-tissue windows. It is known that subtle pathologic abnormalities in other body regions can have greater conspicuity when specialized CT window settings are used in addition to soft-tissue windows. The strategy of using adjunct window settings has helped in the detection of small focal liver lesions, pulmonary emboli, and acute cerebral infarction [6–9]. To our knowledge, specialized CT window settings have not been reported for head and neck imaging.
The primary purpose of this study was to acquire data for the range of densities in small mucosal tumors to optimize CT window settings for the detection of early T-stage SCC. A secondary aim was to evaluate the premise that the use of narrowed window width and higher window level display (mucosal windows) may be helpful in the detection of small head and neck SCCs. We hypothesized that CT mucosal window settings would have greater sensitivity for detecting small head and neck SCC compared with the standard window and level settings normally used for neck CT review.
Materials and Methods
We performed a retrospective case-control study. From the Radiation Oncology Tumor Registry at one institution we identified 96 consecutive patients with T1 or T2 head and neck SCC over a 3-year period between January 2005 and April 2008. Patients were excluded if a CT scan of the neck was not available before biopsy and treatment. Nineteen patients (15 men and four women; mean age, 59 years; range, 48–71 years) met all inclusion criteria. Primary locations included laryngeal, pharyngeal, and oral cavity SCC.
We matched patients with mucosal tumors to control patients without mucosal tumors. Because no prior data existed for power analysis, we oversampled control subjects to patients with tumors in an approximate 2:1 ratio. There were two control groups. The first control group consisted of 29 age-matched subjects who were referred for CT scans of the neck in a 4-month period between September 1, 2007, and January 1, 2008. We chose this time period to overlap with the tumor group and to allow time for follow-up. Patients were excluded if there was a history of malignancy or if there was severe dental amalgam artifact limiting the evaluation of the pharynx or oral cavity. The second control group consisted of six patients from our Radiation Oncology Tumor Registry with a final diagnosis of CUP between January 2005 and April 2008. In this second control group, the absence of a primary tumor was confirmed by negative panendoscopy (including blind mucosal biopsies of the tongue base, nasopharynx, and palatine tonsils), negative FDG PET, and no primary tumor on clinical follow-up. This second control group was used during the blinded visual assessment because we suspected that the presence of lymphadenopathy would influence the study readers. Control groups 1 and 2 together consisted of 35 patients (14 men and 21 women; mean age, 58 years; range, 46–75 years).
Institutional review board approval was obtained. Informed consent was waived because our study involved retrospective review of previously obtained images. The study was HIPAA compliant.
All examinations were performed with 16- or 64-MDCT scanners. Patients were imaged during quiet respiration and were instructed not to swallow during the examination. The plane of section was prescribed parallel to the hyoid bone using lateral scout images. A contiguous helical axial data set was obtained from the sella turcica to the inferior aortic arch with the following parameters: 2.5-mm slice thickness, 19- to 22-cm field of view, 120-kV tube voltage, and 100- to 440-mAs tube current (automatic dose reduction). Iodinated contrast material (100 mL of iopamidol [Isovue 300, Bristol-Myers Squibb]) was given IV with an automated power injection at a rate of 2 mL/s. Imaging was performed after a 60- and 80-second delay for 16- and 64-MDCT, respectively.
Image Analysis Using Mucosal and Standard Soft-Tissue Windows
First, the CT neck scans were anonymized and evaluated independently by two neuroradiologists who were blinded to the diagnosis. These radiologists had 7 and 17 years of experience in reading neck CT examinations. At the evaluation sessions, CT studies of the patients with mucosal tumor and the control subjects were randomly loaded and reviewed.
During the first session, the readers used our institution's standard window settings for neck soft tissue (window width, 400 Hounsfield units [HU]; window level, 30 HU). One week later, the same images were read using mucosal window settings with the option of additional using soft-tissue windows during the second reading. The option of also using soft-tissue windows was more reflective of the normal reading conditions. At the time of blinded review, the optimal CT mucosal window setting from our data had not been calculated. A mucosal window setting of a window width of 150 HU and a window level of 50 HU (i.e., a narrower window width and a higher window level) was used by the readers on the basis of previously published densities for tumor and was modified by consensus to suit our institution's PACS display and the contrast timing of our CT neck studies [10, 11].
The readers recorded the presence of a mucosal lesion graded on a 5-point ordinal degree-of-confidence scale: 1, lesion definitely not present; 2, lesion probably not present; 3, unsure; 4, lesion probably present; and 5, lesion definitely present. They also recorded the side and location of the lesion.
Measurement of Density of Tumor and Normal Mucosa
After the blinded read, the same two readers reviewed the CT studies of the 19 patients with mucosal tumors and correlated the appearance on CT with knowledge of the primary site as reported in the surgical pathologic analysis. The readers recorded whether the tumor was seen on CT. For the tumors that were seen, one reader measured the CT densities in Hounsfield units of the primary tumor, the corresponding site in the contralateral mucosa, and the ipsilateral normal-appearing mucosa adjacent to the primary tumor. These values were obtained by drawing regions of interest at the same image level on a magnified image. The mean of three measurements was taken from each site and compared.
Using the tumor density values, optimal CT mucosal window settings were derived with methods similar to those in the published literature for pulmonary embolus windows: window width equals the mean number of tumor Hounsfield units plus two SD; window level is window width divided by two .
Data were recorded on case record forms by the readers and later were entered into a Microsoft Excel spreadsheet. Statistical analyses were performed using SAS (version 9.1, SAS Institute). All imaging results were compared with the endoscopic and histopathologic findings as the reference standard.
The density of the tumor and adjacent normal mucosa were measured and compared using Student's t-test. A two-tailed p value of < 0.05 was considered statistically significant.
Sensitivity and specificity of the soft tissue and the addition of mucosal window settings were calculated on the basis of accurate localization of tumor. The data were dichotomized by grouping the grade on the degree-of-confidence scale: levels 1 and 2 were regarded as negative findings, and levels 3–5 were regarded as a positive diagnosis of malignancy. Tumors had to match by site and side to be considered as a true-positive diagnosis.
There were 19 primary squamous cell carcinomas in 19 patients, consisting of 12 oropharyngeal tumors, two hypopharyngeal tumors, two supraglottic laryngeal carcinomas, two oral cavity carcinomas, and one nasopharyngeal SCC. Eleven tumors were on the left side, six tumors were on the right side, and two tumors were bilateral. None of the control patients had primary mucosal tumors on imaging or on 18-month follow-up review.
Mucosal Density of Tumor and Normal Mucosa
The distribution of mucosal densities in Hounsfield units for tumor and normal mucosa are shown in Figure 1. Two of 19 primary tumors in 19 patients were not seen by either reader on neck CT, even with knowledge of the biopsy-proven primary site. The mean (± SD) density of the 17 visible tumors was 85.5 ± 18.3 HU. The mean density of the mucosa in the contralateral neck was 64.8 ± 14.9 HU, and the mean density of the ipsilateral normal-appearing mucosa was 55.3 ± 15.2 HU. The differences in density between tumor and normal mucosa (ipsilateral and contralateral) were statistically significant (p < 0.0001). There was no statistically significant difference between the normal mucosa on the ipsilateral and contralateral neck (p = 0.09).
The differences in density between tumor and normal mucosa (density difference equals HU of tumor minus HU of normal mucosa) were as follows: the contralateral mucosa mean difference was 20.6 ± 14.0 HU, and the ipsilateral mucosa mean difference was 30.2 ± 11.2 HU. The differences in density between normal ipsilateral and contralateral mucosa were not significant (9.6 ± 11.9 HU).
Using the tumor densities values, we derived optimal CT mucosal window settings of a window width of 120 HU and a window level of 60 HU to display the density differences between tumor and normal mucosa. These values were obtained with methods similar to the published literature for pulmonary embolus windows . Window width is the mean tumor HU plus 2 SD, or 85.6 + 2(18.3). Window level equals window level divided by 2, or 122/2.
Detection of Tumor Using Mucosal and Standard Soft-Tissue Windows
Reader A detected 12 of 19 tumors using mucosal windows with or without soft-tissue windows, compared with nine of 19 tumors on standard soft-tissue windows alone (sensitivity, 63% and 47%, respectively). Reader B detected nine of 19 tumors with use of mucosal windows with or without soft-tissue windows, compared with eight of 19 tumors with standard soft-tissue windows alone (sensitivity, 47% and 42%, respectively). The number of false-positive results for reader A was six for mucosal window and two for soft-tissue window (specificity, 82% and 94%, respectively). The number of false-positive results for reader B was 10 for mucosal window and nine for soft-tissue window (specificity, 71% and 74%, respectively). The differences in sensitivity and specificity were not statistically significant with the addition of mucosal windows.
Five of 19 primary tumors were not seen by either reader on the blinded evaluation, including two tumors that were not visible on neck CT even with knowledge of the biopsy-proven primary site. The locations of these latter two tumors were the oral cavity buccal mucosa and the palatine tonsil. Three of the five primary tumors that were not seen on initial evaluation by either reader could be seen in hindsight with knowledge of the biopsy results.
CT is often the first imaging technique used for workup of a new neck mass because it has relatively low cost, is widely available, and has good spatial and contrast resolution. Although head and neck SCC is a common indication for CT of the neck, at least one study suggests that CT may be less sensitive than PET and MRI for detection of mucosal tumors . Although it can be argued that it is not the role of the radiologist to identify small mucosal lesions, identification of an area of high suspicion can direct the surgeon to focused biopsies, potentially increasing the diagnostic yield. This allows identification of a small tumor that would otherwise be labeled as an unknown primary tumor and thereby enables directed, less morbid treatment. On CECT, small tumors may be missed because traditionally images are reviewed using soft-tissue window settings (window width, 350–400 HU; window level, 20–60 HU) designed to provide an overall balanced contrast for looking at a range of tissues and structures seen throughout the neck [13, 14]. In our practice, we observed that some mucosal tumors are poorly seen with such standard soft-tissue window settings but may be better seen with use of a narrower CT window width and higher window level. This prompted our study to measure CT attenuation of small tumors and to use these data to introduce a novel method of evaluating the mucosa of the larynx, pharynx, and oral cavity with CT mucosal windows.
The differences in density between tumor and normal contralateral and ipsilateral mucosa were more than 20 HU and 30 HU, respectively. Although these differences in density may sometimes be appreciated on conventional soft-tissue window settings, CT mucosal window settings display the differences with greater contrast (Figs. 2A, 2B, 3A, 3B, 4A, 4B). It is important to emphasize that the optimal window setting derived from our density data—window width of 120 HU and window level of 60 HU—serves as a starting point and can be modified depending on radiologist preference and equipment. In the first part of our study, we assigned a window width of 150 HU and a window level of 150 HU for the blind reviews on the basis of previously reported tumor densities and our PACS display [10, 11]. In a study by Bae et al. , all seven observers of 25 CTA chest studies preferred their own personal settings for detection of pulmonary embolus over the preset pulmonary embolus window setting. Those authors recognized that optimal window settings serve as a guideline and that most radiologists prefer to adjust the window settings by visual assessment. Furthermore, adjustment of window settings may be necessary depending on degree of mucosal enhancement, which can differ between patients and even vary in the same patient depending on CT acquisition technique.
Our sensitivity for tumor detection was lower than that reported in some prior studies (61–78%) [4, 15]. We think this is most likely because we focused solely on small (T1 and T2) tumors, whereas other studies included larger tumors. Use of mucosal window settings together with soft-tissue windows detected more tumors (sensitivity, 47–63%) than did the use of soft-tissue windows alone (sensitivity, 42–47%). Although the difference in sensitivity did not prove to be statistically significant, this proof-of-concept study introduces mucosal window settings and shows that, in certain individuals, examples of the tumors were substantially more conspicuous with mucosal windows (Figs. 2A, 2B, 3A, 3B, 4A, 4B). In the age of PACS, it is very easy for the radiologist to use window “preset value” keys on their PACS machine. We found that the time required to review each case in mucosal windows averaged less than 2–3 minutes, so we think that, for select patients, the method is a practical one.
Another potential application of CT mucosal window settings may be for accurate delineation of tumor extent in patients with larger mucosal tumors. This is important for both tumor staging and radiotherapy treatment planning. In a study that compared CT to surgical specimens for oropharyngeal, laryngeal, and hypopharyngeal tumors, CT both overestimated tumor volume and failed to depict a small fraction of macroscopic tumor extension . Because mucosal window settings presumably are more sensitive than standard neck CT settings, it will be important to examine in future studies whether use of mucosal window settings improves accuracy.
A possible disadvantage of CT mucosal windows settings, as seen from the preliminary results from our readers, is a lower specificity and higher false-positive rate for tumor compared with soft-tissue windows. For this reason, mucosal windows should not be applied for screening patients with low risk of mucosal malignancy. In patients with cervical lymphadenopathy and CUP, the balance between sensitivity and specificity will need to be carefully considered. Although these patients ordinarily undergo endoscopy and biopsy after imaging, false-positive results could mislead or bias the endoscopist in favor of the incorrect location. Ultimately, the potential value of CT assessment of the mucosa in selected patients will need to be continuously assessed. The next steps for this research will involve prospective assessment of patients with the collaboration of the endoscopist. For example, it remains unknown whether peritumoral inflammation could be mistaken for tumor in some cases.
There are several limitations to our study. First, this was a retrospective study with a small study population. Although we have acquired data regarding tumor densities and optimal CT mucosal windows, future validation of this method will require study of more patients with more readers. Ultimately, a prospective study is required to investigate whether identifying “suspicious” mucosal areas before endoscopy can help the clinician to locate and confirm cancer in the mucosa. One-to-one correlation will need to be performed between possible mucosal abnormality, endoscopy findings of that mucosa, and histopathologic analysis of that specific mucosa. Second, the detection of tumor may have been biased in cases of cervical lymphadenopathy. We tried to control for this by including a control group with lymphadenopathy but no mucosal lesion (control group 2). Our readers were aware of this group so that they were less inclined to diagnose tumor on the basis of lymphadenopathy alone. If lymphadenopathy did bias the readers, we could expect similar rates of bias for both window settings, and this would not influence differences in tumor detection.
In conclusion, early T-stage mucosal tumors have higher density than normal mucosa on CECT. Their conspicuity could be amplified with the use of CT mucosal windows with narrower window width and higher window level. We propose settings of a window width of 120 HU and a window level of 60 HU as a guideline to be adjusted according to personal visual preferences, differences in CT technique, and variations in PACS display. Future studies are required to validate this method, but there are important clinical dilemmas for which this reviewing method might potentially be applied. Specifically, mucosal windows could aid in the detection of unknown primary tumors, which may then reduce the need for further imaging studies and allow timelier focused biopsy and treatment. In addition, mucosal windows may aid in delineation of the full extent of a known mucosal tumor, which has implications for accurate pretreatment tumor staging and radiation planning.
C. M. Glastonbury is an investor in and a consultant for Amirsys.
Pavlidis N, Pentheroudakis G, Plataniotis G. Cervical lymph node metastases of squamous cell carcinoma from an unknown primary site: a favourable prognosis subset of patients with CUP. Clin Transl Oncol 2009; 11:340 –348
Keberle M, Tschammler A, Hahn D. Single-bolus technique for spiral CT of laryngopharyngeal squamous cell carcinoma: comparison of different contrast material volumes, flow rates, and start delays. Radiology 2002; 224:171–176
Stokkel MP, ten Broek FW, Hordijk GJ, Koole R, van Rijk PP. Preoperative evaluation of patients with primary head and neck cancer using dual-head 18fluorodeoxyglucose positron emission tomography. Ann Surg 2000; 231:229 –234
Daisne JF, Duprez T, Weynand B, et al. Tumor volume in pharyngolaryngeal squamous cell carcinoma: comparison at CT, MR imaging, and FDG PET and validation with surgical specimen. Radiology 2004; 233:93–100