Comparison of Sonography and CT for Differentiating Benign from Malignant Cervical Lymph Nodes in Patients with Squamous Cell

Carcinoma of the Head and Neck

Misa Sumi 1 Masafumi Ohki 2 Takashi Nakamura 1
OBJECTIVE. We compared the ability of sonography and CT to differentiate benign from malignant cervical lymph nodes in patients with squamous cell carcinoma of the head and neck. MATERIALS AND METHODS. We analyzed 209 cervical nodes (102 metastatic and 107 nonmetastatic) from 62 patients with head and neck cancer. These nodes were topographically correlated by node between images and surgical specimens, and accordingly between sonography and CT. RESULTS. The area under the receiver operating characteristic curve (Az value) for the overall impressions of metastatic or nonmetastatic nodes was signicantly greater for sonography (power Doppler sonography plus gray-scale sonography, 0.97 0.005; gray-scale sonography, 0.95 0.004) than for CT (0.87 0.018). Receiver operating characteristic curve analysis also showed that the greater ability of sonography to depict the internal architecture of the nodes (Az value, 0.96 0.006) compared with CT (Az value, 0.81 0.027) signicantly contributed to the better performance of sonography compared with CT in diagnosing metastatic nodes in the neck. On the other hand, size criterion (the short-axis diameter) was equally predictive in sonography and CT. The greater contributions of internal architectures relative to the size criterion of the node in the sonographic assessment for metastatic nodes were further evidenced by the ndings that sonography provided higher sensitivity and specicity than CT did, whereas the cutoff points for the short-axis diameter in both tests were equivalent. CONCLUSION. Sonography performed signicantly better than CT in depicting cervical metastatic nodes. Sonography could be a useful adjunct to CT in surveying cervical metastatic nodes.

Received August 14, 2000; accepted after revision September 25, 2000.
1 Department of Radiology and Cancer Biology, Nagasaki University School of Dentistry, 1-7-1 Sakamoto, Nagasaki 852-8588, Japan. Address correspondence to T. Nakamura. 2 Department of General Education, School of Health Sciences, Kyushu University, Fukuoka 812-0054, Japan.

AJR 2001;176:10191024
0361803X/01/17641019 American Roentgen Ray Society

taging of neck metastasis is a crucial step in treating patients with head and neck squamous cell carcinoma [1]. An increase in nodal size was found to be an effective imaging criterion for the detection of metastatic cervical nodes with CT and MR imaging [2, 3]. However, size determination alone is not effective enough for detecting metastatic nodes. Therefore, several studies have attempted to improve diagnostic accuracy by assessing the internal architecture of the node and using other tissue-specic imaging techniques such as sonographically guided ne-needle aspiration biopsy and 18Fuorodeoxyglucose positron emission tomography [1]. Recently, Curtin et al. [4] showed that combined information on the size and internal architecture of the node facilitated the detection of nodes that were metastatic from squamous cell carcinoma of the head and neck, conrming, at least in part, the signicance of

the assessment of the internal architecture of a node for the detection of metastatic nodes. Sonographic evaluation of enlarged nodes is also based on assessment of the internal architecture of the node as well as size determination of the node, and abnormalities in the node may be reected by increased parenchymal echogenicity or loss of hilar echogenicity in malignant disease of the node [57]. In addition, the recent development of Doppler sonography technology has shed light on the diagnostic signicance of changes in nodal blood ow in differentiating metastatic from nonmetastatic nodes [810]. These studies have shown that, albeit some controversies, the assessment of nodal blood ow patterns had some impact on improving the performance of sonography in depicting metastatic nodes in the neck [6]. Therefore, in this study, we compared the performance of gray-scale sonography, power Doppler sonography, and CT in staging meta-

AJR:176, April 2001

1019

Sumi et al. static nodes in the necks of patients with head and neck squamous cell carcinoma. Our focus was on the diagnostic performance of two imaging criterianamely, the size and internal architecture of the nodeon sonography and CT.
Materials and Methods
Patients We analyzed 209 lymph nodes (102 metastatic and 107 nonmetastatic) from 62 patients (38 men and 24 women; age range, 4592 years; mean age, 66 years) with head and neck squamous cell carcinoma. These carcinomas had arisen from the tongue (n = 12), the lower gingiva (n = 10), the upper gingiva (n = 10), the oral oor (n = 8), the hypopharynx (n = 7), the larynx (n = 6), the buccal mucosa (n = 4), the palate (n = 3), the oropharynx (n = 1), or the parotid gland (n = 1). All these nodes were examined on gray-scale and power Doppler sonography and also on CT. Topographic correlations, between dissected nodes and sonograms and between dissected nodes and CT images, were performed for each node, as previously described [8]. The indications for surgery included the presence of a clinically palpable, indurated node or nodes; or the presence of an enlarged node or nodes that were 1 cm or larger, that exhibited rim enhancement on CT images, or both. We facilitated the correlation between dissected nodes and the respective images by adding a report as a reference, describing the data concerning the approximate location relative to the surrounding anatomic structures, such as vessels, salivary glands, bones, and muscles. During surgery, surgeons identied the lymph nodes that were imaged on sonography and CT with the aid of the report. Thereafter, the nodes were excised en bloc with the adjacent tissue to ascertain more easily the spatial relationship between the excised nodes and the surrounding structures. The excised nodes that matched those on sonograms and CT images were then examined histopathologically. The nodes were grouped into four levels (levels I, II, III, and IV) as previously described [3]. Because there were only a small number of nodes in level IV, we categorized the nodes at levels III and IV as level III plus level IV. Sonographic Imaging Gray-scale and power Doppler sonography were performed using a unit (Logiq 700; General Electric Yokogawa Medical Systems, Tokyo, Japan) equipped with a wide-bandwidth transducer (range, 613 MHz). Gray-scale sonography was performed at 10 MHz; power Doppler sonography was performed at 8 MHz. Standard Doppler settings were chosen for optimal detection of the signals from the lymph node vessels, which had low-velocity ow. Common settings of pulse repetition frequency (500 Hz) and wall lter (75 or 62 Hz) were used. Representative images from each lymph node were obtained using grayscale and power Doppler sonography so that the maximal axial section of the nodes appeared on sonographic images. The following sonographic criteria were used in this study: short axial diameter of the node measured with a calliper on sonograms; the presence or absence of hilar echoes; and the presence or absence of hilar blood ow. Previous reports have shown that an increase in short axial diameter but not the long axial diameter is an efcient indicator for metastatic nodes [5, 6]. A preferential increase in short-axis diameter results in a round or oval node, so these features are suggestive of metastatic nodes. The hilum is identied as a highly echogenic structure in the central part of the node. Metastatic tumor cells frequently invade this part of the node and, in that case, the echogenicity of the hilum may be lost. Multivariate feature analysis showed that the presence or absence of hilar echoes, an increase in the short-axis diameter, and the presence of normal hilar ow were the sonographic features that were predictive of nonmetastatic (presence of hilar echoes and hilar ow) and metastatic (increases in short-axis diameter) lymph nodes [8]. Helical CT Patients were scanned using a CT imaging system (HiSpeed Advantage SG; General Electric Medical Systems, Milwaukee, WI). The scanning orientation was parallel to the Frankfurt horizontal line. Scanning was performed with a collimation of 3 mm, a pitch of 1:1, a matrix of 512 512, a display eld of view of 23 cm, 120 kVp, and 200 mA. CT examination was carried out after an IV bolus injection of approximately 100 mL (2 mL/kg of body weight) of iopamidol (Iopamiron 300; Schering, Berlin, Germany) at a rate of 1.0 mL/sec. We started scanning 80 sec after the start of contrast medium injection. CT for examination of metastatic nodes in the neck was usually completed in 5060 sec after the start of scanning. The scanning period (80140 sec after the start of contrast medium injection) was conrmed to be the time when the lymph node showed plateaulike staining kinetics and provided appropriate contrast against the neighboring muscles. We preferred this rate of contrast medium injection (1.0 mL/sec) instead of faster rates (e.g., 2.0 mL/sec) because the slower injection rate provided prolonged plateaulike periods for lymph nodes without any signicant loss of enhancement efciency during the following scanning period. We obtained reformatted axial images of 3 mm in thickness from these data. We used two CT criteria, both of which were deemed to be important indicators for differentiating benign from metastatic nodes in the preceding studies [1, 2]: short- and long-axis diameters of the node measured with a caliper on CT images and the presence or absence of abnormal staining patterns of the node, such as rim enhancement and irregularly increased staining of the node. Receiver Operating Characteristic (ROC) Curve Analysis We performed ROC curve analyses on overall impression of whether the node was metastatic or nonmetastatic, the presence or absence of an internal architecture suggestive of a metastatic or nonmetastatic node, and the size of the node on either sonographic or CT images. We used a total of 621 images (207 images each for gray-scale sonography, power Doppler sonography, and CT) of the metastatic and nonmetastatic lymph nodes. Sonographic and CT images were documented on print paper using a color digital printer (Pictrography; Fujilm, Tokyo, Japan) and were presented to four radiologists who were experienced in both sonography and CT in the head and neck region, had not previously seen these cases, and were not aware of any patient information. Before ROC curve analyses, the quality of printed images regarding noise of images, special resolution, and motion artifacts was assessed by two of the authors as excellent, good, fair, or poor. All 621 images were categorized as excellent, good, or fair. In advance, one of the authors measured the size of the nodes used in this study: the short-axis diameter of the node for sonography and the short- and long-axis diameters of the node for CT. Then the four observers were rst asked to score the presence or absence of hilar echoes on gray-scale sonographic images using a 5-point rating scale (1 = denitely present, 2 = probably present, 3 = unclear if the nding is present or not, 4 = probably not present, 5 = denitely not present). Next, they were asked to score the presence or absence of abnormal staining patterns of the node on CT images using a similar rating scale (1 = denitely not present, 2 = probably not present, 3 = unclear if the nding is present or not, 4 = probably present, 5 = denitely present). Then, at least 2 weeks after the second assessment, the observers were again asked to score the presence or absence of normal hilar blood ow or normal hilar echogenicity in the nodes on gray-scale plus power Doppler sonographic images. There may be an undened but important imaging feature or features that could affect the observers decision about differentiation between benign and malignant nodes. To estimate the relative importance of the imaging criteria tested in the present study, we also assessed the overall impression of the observers of the nodal images concerning benignancy and malignancy. Therefore, again at least 2 weeks after the third assessment, the overall impressions of these observers of the gray-scale images, gray-scale plus power Doppler sonographic images, and CT images were also scored using a 5-point rating scale (1 = definitely not metastatic, 2 = probably not metastatic, 3 = indeterminate, 4 = probably metastatic, 5 = denitely metastatic). We also assessed the performance of the size criterion for predicting the presence or absence of metastasis in the node. This test was based on an ordinal scale of increasing points for either the short- or long-axis diameter of lymph nodes. A rank of 1 represented nodes with a short- or long-axis diameter of 4 mm or less, and an increase of 1 mm in short- or longaxis diameter increased the rank by 1. Thus, for example, if a node was 10 mm in its short-axis diameter, then it would be assigned a rank of 7. For each imaging technique, a binormal ROC curve was tted to each observers rating data using software (ROCKIT; Metz CE, Chicago, IL) [11]. The diagnostic performance of each imaging technique was determined by calculating the area under each

1020

AJR:176, April 2001

CT and Sonography of Head and Neck Squamous Cell Carcinoma

observer-specic ROC curve (Az value) [12]. Twoway analysis of variance test showed that signicant differences in rating data were noted among different imaging techniques but not among the four observers. Therefore, averaged ROC curves representing the performance of the four observers as a group were calculated for comparing the imaging technique. Az values were expressed as means standard errors, and the signicance of the difference in the Az value between the imaging techniques was tested by the paired t test for each observer. ROC curves were also calculated and tted for size determination represented by the ordinal scale of increasing points described earlier. The signicance was tested by the Students t test for data obtained from assessment based on node size. We tested the signicance of the differences in Az values between the imaging techniques for a total of 207 nodes in the neck and for each node population categorized into three levels in the neck. Calculation of Sensitivity and Specicity for Each Imaging Technique We calculated the sensitivity (the number of nodes with positive ndings at imaging and at histology divided by the number of nodes with positive ndings at histology) and specicity (the number of nodes with negative ndings at imaging and at histology divided by the number of nodes with negative ndings at histology). The accuracy was calculated using the following formula: the number of nodes with positive ndings at imaging and at histology plus the number of nodes with negative ndings at imaging and at histology divided by the total number of nodes.

Results

First, we compared sonographic and CT features of metastatic and nonmetastatic lymph nodes. Nonmetastatic nodes were characteristically shown on CT images as discrete, kidney-

shaped soft-tissue structures with the hilum composed of fat tissue concaving into the central portion of the node (Fig. 1A). All the abnormal lymph nodes detected on CT were visible on sonography. Furthermore, sonography showed more details of the reactive nodes (Figs. 1B and 1C): The nodal parenchyma exhibited homogeneous and low echogenicity, and the hilum was identied as a highly echogenic structure in the central part of the node (Fig. 1B). On power Doppler sonograms, the hilum was superimposed by a blood ow signal from the hilar vessel, which was characteristically club- or Y-shaped and extended from the extranodal area into the deep portion of the node (Fig. 1C). In contrast, metastatic nodes were round and exhibited rim enhancement, central attenuation (Fig. 2A), or an irregular staining pattern in the parenchyma on CT images. Metastatic nodes commonly lacked normal hilar echogenicity (Fig. 2B), which was associated with abnormal peripheral or parenchymal blood ow signals (the absence of normal hilar ow) as shown on power Doppler sonography (Fig. 2C). A substantial part of the metastatic nodes displayed homogeneous staining patterns after contrast enhancement on CT images (Fig. 3A), whereas sonographic results were still highly predictive of the presence of metastasis, as evidenced by a loss of normal hilar echogenicity (Figs. 3B and 3C). This nding suggests that the homogeneous enhancement of the node on CT is a nonspecic nding. These ndings suggest that, compared with CT, sonography has a greater potential in depicting nodes metastatic from squamous cell carcinomas in the head and neck region. To substantiate this hypothesis, we assessed the

ability of sonography and CT to detect metastatic nodes by conducting ROC curve analyses on enlarged nodes that were topographically correlated by node between sonograms and CT images. ROC curves for the overall impression revealed that gray-scale sonography yielded signicantly better performance than CT (Az value for CT, 0.87 0.018; Az value for gray-scale sonography, 0.95 0.004; p = 0.017) (Fig. 4). Gray-scale plus power Doppler sonography performed as well as gray-scale sonography, with Az values at a similar level for these two techniques (0.97 0.005; p = 0.095). These ndings may indicate that blood ow signals as depicted by Doppler sonography did not signicantly contribute to the diagnostic performance of sonography in depicting metastatic nodes. We next conducted ROC curve analyses on data obtained by either size determination alone or by assessment of internal architecture alone for each of the three imaging techniques. ROC curve analysis for sonography showed that Az values calculated for the assessment using internal architectural patterns (0.92 0.005 for grayscale sonography; 0.96 0.006 for gray-scale plus power Doppler sonography) were signicantly greater than those for assessment using short-axis diameter (0.87 0.024) (p = 0.004 vs. gray-scale sonography; p < 0.001 vs. grayscale plus power Doppler sonography) (Fig. 5). In contrast, ROC analysis for CT showed that Az values calculated for the assessment using internal architecture (abnormal staining of the node) (0.81 0.027) did not differ signicantly from those for the assessment using short-axis diameter (0.86 0.026; p = 0.15) (Fig. 6). Furthermore, Az values for the assessment using the

A
Fig. 1.70-year-old man with nonmetastatic (reactive) node from squamous cell carcinoma of buccal mucosa. A, CT scan obtained after injection of contrast material shows reactive node (arrow ) that has failed to enhance. B, Gray-scale sonogram shows same reactive node (arrows ) as shown in A. Note associated hilar echogenicity. C, Power Doppler sonogram of reactive node shows blood ow signal overlapping hilar echogenicity.

AJR:176, April 2001

1021

Sumi et al.

A
Fig. 2.63-year-old man with squamous cell carcinoma of tongue. A, CT scan obtained after injection of contrast material shows metastatic node (arrow ). Note rim enhancement. B, Gray-scale sonogram shows same node (arrows ) as shown in A. Note absence of normal hilar echogenicity. C, Power Doppler sonogram shows metastatic node. Note peripheral blood ow signal.

Fig. 3.66-year-old man with squamous cell carcinoma of oral oor. A, CT scan obtained after injection of contrast material shows metastatic node (arrow ), exhibiting homogeneous enhancement. B, Gray-scale sonogram shows same node (arrows ) as seen in A. Note irregularly distributed echogenicity. C, Power Doppler sonogram shows abnormal parenchymal blood ow signals.

internal architecture of the node with gray-scale sonography (0.92 0.005) were greater than those obtained with CT (0.81 0.027; p = 0.020). Gray-scale plus power Doppler sonography was associated with a further increase in Az value in the assessment of internal architecture (0.96 0.006, p = 0.007 vs. gray-scale sonography). ROC curve analyses of CT images indicated that size determination using the short-axis diameter of the node was associated with an Az value (0.86 0.026) proximate to that of the overall impression (0.87 0.018) (Fig. 2C). The Az value obtained on CT images by size determination using long-axis diameter (0.73 0.034), on the other hand, was signi-

cantly smaller than that obtained by size determination using short-axis diameter (0.86 0.026; p < 0.005) (Fig. 6). The Az values for size determination by the short-axis diameter of the node were at similar levels in sonographic (0.87 0.024) and CT (0.86 0.026) evaluation. Consistent with the ndings of ROC analysis for assessment using the internal architecture of the node, combined assessment on gray-scale plus power Doppler sonography displayed the greatest sensitivity and specicity among the three different imaging techniques (sensitivity, 0.68, 0.69. and 0.74 for CT, gray-scale sonography, and gray-scale plus power Doppler sonography, respectively;

specicity, 0.82, 0.90, and 0.96 for CT, grayscale sonography, and gray-scale plus power Doppler sonography, respectively). The compromise cutoff points of the short axial diameter criterion for metastatic nodes that were associated with the best accuracy (0.79 for sonography and 0.80 for CT) were 9 and 10 mm for sonography and CT, respectively, supporting the notion that the size criterion was equally effective for sonography and CT. Then we asked whether the observed superiority in the performance of sonography in depicting metastatic nodes is dependent on levels in the neck. To address this question, we compared Az values at each level of the neck

1022

AJR:176, April 2001

CT and Sonography of Head and Neck Squamous Cell Carcinoma

Fig. 4.Graph shows receiver operating characteristic curve analysis for CT (dashed line ), gray-scale sonography (dotted line ), and gray-scale plus power Doppler sonography (solid line ) by overall impression. Note that sonography performs better than CT.

calculated from the averaged ROC curves tted for the overall impression of the four observers. At all three levels in the neck, grayscale plus power Doppler sonography yielded signicantly greater Az values (0.94 0.007, 0.96 0.010, and 0.97 0.008, for levels I, II, and III plus IV, respectively) than those obtained on CT (0.89 0.012, 0.92 0.015, and 0.86 0.003, for levels I, II, and III plus IV, respectively; p = 0.046, 0.040, and 0.042, for levels I, II, and III plus IV, respectively). Grayscale sonography, on the other hand, yielded signicantly greater Az values than did CT at levels I (0.94 0.008, p = 0.016) and III plus IV (0.95 0.007, p = 0.022), but not at level II (0.95 0.009, p = 0.171). This was, at least in part, because of a signicantly elevated Az value for CT at level II of the neck, compared with those at the other levels in the neck (paired t test, p = 0.0006). At all three levels in the neck, no signicant difference was found in Az values between gray-scale and power Doppler sonography.

Fig. 5.Graph shows receiver operating characteristic curve analysis for sonography by overall impression (gray-scale = dashed line, gray-scale and power Doppler sonography = upper solid line ), internal architecture (gray-scale = lower dotted line, gray-scale and power Doppler sonography = upper dotted line ), and short-axis length (lower solid line). Note higher performance by internal architectural assessment compared with that by short-axis diameter assessment.

Fig. 6.Graph shows receiver operating characteristic curve analysis for CT by overall impression (upper solid line ), internal architecture (dotted line ), longaxis diameter (dashed line ), and short-axis diameter (lower solid line ). Note that performance by internal architectural assessment does not exceed that by short-axis diameter assessment and that short-axis diameter assessment performs better than long-axis diameter assessment does.

Discussion

We have shown in this article that sonography can differentiate metastatic from nonmetastatic nodes in the neck better than CT. Furthermore, we found that the better performance of sonography for depicting metastatic nodes appears to be because of its greater ability to delineate changes in the internal architecture of the node. In contrast, the contribution of size determination to the diagnostic performance of sonography was relatively less than that of CT.

Sonographic evaluation was found to be adequate for differentiating metastatic nodes from nonmetastatic nodes in patients with head and neck cancer [1, 510]. Multivariate feature analysis showed that of the sonographic criteria advocated for metastatic or nonmetastatic nodes, the presence or absence of hilar echogenicity, an increase in short-axis diameter, and the presence of normal hilar blood ow were the only sonographic features that were predictive of metastatic (increases in short-axis length) or reactive (the presence of normal hilar blood ow, hilar echoes, or both) nodes [6]. ROC curve analyses in our study showed that the performance of gray-scale plus power Doppler sonography using internal architecture assessment was signicantly greater than that of gray-scale sonography using internal architecture assessment (Fig. 2B). However, the overall performance of gray-scale and gray-scale plus power Doppler sonography was not signicantly different (Fig. 2A). Taken together with the ndings from multivariate analysis [6], which showed the only three sonographic features (short-axis diameter, the presence or absence of hilar echoes, and the presence or absence of hilar blood ow) contributing signicantly to the diagnostic performance of sonography for detecting metastatic and reactive cervical nodes, the present ndings suggest that as yet an undened factor or factors may contribute, albeit not signicantly, to the diagnostic performance of gray-scale sonography for depicting metastatic nodes on the basis of overall impression assessment.

Diagnosis of metastatic nodes using CT also depends on size determination and assessment of changes in the internal architecture. In this context, Curtin et al. [4] studied the effect of size criteria and internal architectural changes of the node on diagnostic accuracy for metastatic nodes, and these researchers showed that the addition of information on internal architecture of the node resulted in a substantial improvement of the diagnostic performance of CT using longaxis length (maximum axial diameter). We found that CT did not perform well when the long-axis diameter criterion alone was used (Fig. 2C); however, when the short-axis diameter was used as a size criterion, even the size criterion alone, CT performed as well as, if not better than, when the internal architecture alone was used (Fig. 2C). The performance of CT using short-axis diameter assessment approximated that of CT using overall impression. Therefore, these ndings suggest that the contribution of size determination to the diagnostic performance was relatively greater in CT than in sonography. Different anatomic levels in the neck may affect the performance of diagnostic imaging for metastatic lymph nodes. Several studies of CT and sonography have conrmed that enlarged nodes at different levels require different size criteria for predicting whether nodes are metastatic [4, 7]. Therefore, we also assessed the performance of sonography and CT for detection of metastatic nodes in each of the three levels in the neck. As stated, at all levels in the neck, gray-scale plus power Doppler sonogra-

AJR:176, April 2001

1023

Sumi et al. phy performed best among the three imaging techniques. Our ndings also indicated that the performance of sonography was not signicantly different among the different levels in the neck. Again, gray-scale sonography as assessed by overall impression performed as well as gray-scale plus power Doppler sonography at all levels in the neck. It is interesting to note that level II was the only level in the neck in which CT performed as well as gray-scale sonography. At the other levels in the neck, gray-scale sonography performed better than CT. This nding could also be explained by the nding that nodes at level II were largest among the nodes at all three levels on CT images; the average short-axis lengths of nodes at levels I, II, and III plus IV, respectively, were 7 (reactive) and 11 mm (metastatic), 7 (reactive) and 13 mm (metastatic), and 6 (reactive) and 10 mm (metastatic). The long-axis lengths of reactive and metastatic nodes were also greatest for nodes at level II. Another study of sonography [7] also showed that metastatic nodes at level II were larger than those at the other levels of the neck. Thus, greater sizes of nodes at level II may facilitate precise interpretation of the internal architecture of metastatic nodes at this anatomic level using CT. The lack of a requirement for radiation and the low cost may be major advantages of sonographic examination. In addition, our ndings clearly indicated that sonography performed better than CT for detecting metastatic nodes in patients with head and neck squamous cell carcinoma. Nevertheless, the fact that sonographic examination takes more time than CT may greatly diminish the value of sonography; in general, CT for the whole neck is usually completed 23 min after the start of contrast medium injection, whereas sonographic examination of the bilateral neck requires at least 30 min. In addition, CT examination is necessary for the detection of deep cervical nodes such as those in the retropharyngeal space. Therefore, CT should be the rst choice for a survey of metastatic nodes in the neck, and it may be reasonable to perform sonographic examination for detailed study of suspected nodes in the neck after CT surveillance.
Acknowledgments
3. van den Brekel MWM, Stel HV, Castelijns JA, et al. Cervical lymph node metastasis: assessment of radiologic criteria. Radiology 1990;177:379384 4. Curtin HD, Ishwaran H, Mancuso AA, Dalley RW, Caudry DJ, McNeil BJ. Comparison of CT and MR imaging in staging of neck metastases. Radiology 1998;207:123130 5. Vassallo P, Wernecke K, Roos N, Peters PE. Differentiation of benign from malignant supercial lymphadenopathy: the role of high-resolution US. Radiology 1992;183:215220 6. Chikui T, Yonetsu K, Nakamura T. Multivariate analysis of sonographic ndings of metastatic cervical lymph nodes: contribution of blood ow features revealed by power Doppler sonography for predicting metastasis. AJNR 2000;21:561567 7. van den Brekel MWM, Castelijns JA, Snow GB. The size of lymph nodes in the neck on sonograms as a radiologic criterion for metastasis: how reliable is it? AJNR 1998;19:695700 8. Ariji Y, Kimura Y, Hayashi N, et al. Power Doppler sonography of cervical lymph nodes in patients with head and neck cancer. AJNR 1998;19:303307 9. Choi MY, Lee JW, Jang KJ. Distinction between benign and malignant causes of cervical, axillary, and inguinal lymphadenopathy: value of Doppler spectral waveform analysis. AJR 1995;165:981984 10. Tschammler A, Ott G, Schang T, SeelbachGoebel B, Schwager K, Hahn D. Lymphadenopathy: differentiation of benign from malignant diseasecolor Doppler US assessment of intranodal angioarchitecture. Radiology 1998;208:117123 11. Metz CE. ROC methodology in radiologic imaging. Invest Radiol 1986;21:720733 12. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:2936

We acknowledge Yoshiharu Higashida for his suggestions. We are also grateful to Charles E. Metz for the use of his ROCKIT program.
References
1. Anzai Y, Brunberg JA, Lufkin RB. Imaging of nodal metastases in head and neck. J Magn Reson 1997;7:774783 2. Som P. Detection of metastasis in cervical lymph nodes: CT and MR criteria and differential diagnosis. AJR 1992;158:961969

1024

AJR:176, April 2001