Evaluating oropharyngeal carcinoma with transcervical ultrasound, CT, and MRI

Oral Oncology 78 (2018) 177–185

Contents lists available at ScienceDirect

Oral Oncology journal homepage: www.elsevier.com/locate/oraloncology

Evaluating oropharyngeal carcinoma with transcervical ultrasound, CT, and MRI

T

Farhoud Farajia, Stephanie F. Coquiab,c, Meghan B. Wenderothb, Ericka S. Padillab, Dana Blitzb, ⁎ M. Robert DeJongb, Nafi Aygunb, Ulrike M. Hamperb, Carole Fakhrya, a

Department of Otolaryngology-Head and Neck Surgery, Johns Hopkins School of Medicine, Baltimore, MD, United States The Russell H. Morgan Department of Radiology and Radiological Sciences, Johns Hopkins Hospital, Baltimore, MD, United States c Division of Medical Imaging Products, Office of New Drugs, Center for Drug Evaluation and Research, Food and Drug Administration, Silver Spring, MD, United States b

A R T I C L E I N F O

A B S T R A C T

Keywords: Head and neck Oropharynx Squamous cell carcinoma Ultrasound Computed tomography Magnetic resonance imaging

Objective: To compare transcervical ultrasonography (US) to standard cross-sectional imaging for the visualization of human papillomavirus-related oropharyngeal cancer (HPV-OPC). Materials and methods: Patients with HPV-OPC and available standard imaging (CT and/or MRI) were identified in clinic and prospectively enrolled. US was performed to visualize the oropharynx and lymph nodes. Tumor characteristics across imaging modalities were evaluated (CT versus MRI, and US versus standard imaging (SI)). Results: Forty-three patients were included. The overall blinded detection rates for CT and MRI were 83% and 71%, respectively. The unblinded detection rate for US was 98%. Agreement of tumor anatomic subsite was moderate for both CT vs MRI (κ = 0.59) and US vs SI (κ = 0.47). Comparison of tumor size by CT and MRI showed statistically significant correlations in craniocaudal (CC), anteroposterior (AP), and mediolateral (ML) dimensions (RhoCC = 0.51, pCC = 0.038; RhoAP = 0.81, pAP < 0.0001; RhoML = 0.57, pML = 0.012). Tumor size estimates by US and SI showed statistically significant correlations in CC and AP, but not ML (RhoCC = 0.60, pCC = 0.003; RhoAP = 0.71, pAP < 0.0001; RhoML = 0.30, pML = 0.08). Tumor volume estimates improved correlations between US and SI (Rho = 0.66, p < 0.0001). Stratification of US patients into early and late imaging studies demonstrated an increase in correlation strength from early (Rho = 0.32, p = 0.32) to late groups (Rho = 0.77, p < 0.0001) demonstrating that ultrasound accuracy improved with experience. Conclusions: Our findings suggest that transcervical ultrasonography is a sensitive and relatively accurate adjunct to standard imaging for the evaluation of oropharyngeal tumors. Its cost, portability, and potential for inclinic and serial imaging render US an attractive modality to further develop for imaging oropharyngeal tumors.

Introduction The incidence of oropharyngeal squamous cell carcinoma (OPC) has increased significantly worldwide [1]. Trends in OPC are attributed to increasing oral exposure to human papillomavirus (HPV) infection [2]. Indeed, approximately 70% of OPCs in the United States are attributable to HPV [3]. HPV-positive OPC (HPV-OPC) are typically small tumors that arise within lymphoid-associated epithelial crypts of the palatine and lingual tonsils [4], rendering clinical and radiographic imaging difficult [5–11]. While computed tomography (CT) represents the most commonly available modality for imaging OPC, it is limited by dental filling artifact and poor soft tissue contrast resolution. Despite superior soft tissue contrast resolution to CT, magnetic resonance imaging (MRI) can be

susceptible to significant motion artifact [9]. Furthermore, while intravenous radiocontrast agents enhance tumor visualization relative to surrounding tissue at most head and neck sites, the normal lymphoid tissue of the oropharynx enhances to a similar extent and precludes clear delineation of tumors [10]. Oropharyngeal lymphoid tissue also exhibits basal metabolic activity resulting in diffuse, mild FDG uptake that can obscure positron emission tomography (PET) signal from small tumors [11]. As a result, neither CT, MRI, nor PET/CT independently serve as an optimal imaging modality for the evaluation of oropharyngeal tumors [12,13]. Our group has investigated the potential role of transcervical ultrasonography (US) in the evaluation of oropharyngeal tumors. We have shown that US can identify the primary tumor site of unknown primary head and neck carcinoma and visualize clinically relevant

⁎ Corresponding author at: Johns Hopkins University School of Medicine, Otolaryngology-Head and Neck Surgery, 601 N. Caroline Street, Sixth Floor, Baltimore, MD 21287, United States. E-mail address: [email protected] (C. Fakhry).

https://doi.org/10.1016/j.oraloncology.2018.01.016 Received 8 October 2017; Received in revised form 15 January 2018; Accepted 22 January 2018 Available online 20 February 2018 1368-8375/ © 2018 Elsevier Ltd. All rights reserved.

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

features of BOT tumors [14–16]. We recently reported an optimized US protocol implementing anatomic landmarks for high-confidence visualization of tonsillar and BOT tumors [17]. In the current report, we apply this optimized protocol to compare the performance of CT, MRI, and US in accurately identifying and evaluating the dimensions of 43 tonsillar and BOT tumors.

Table 1 Patient characteristics. n

%

Age

Median Mean Range

59 59.6 42–81

Materials and methods

Sex

Male Female

60 7

90 10

Patient population

Race

White Non-white

61 6

91 9

Patients with OPC and available cross-sectional imaging (CT and/or MRI) were identified in clinic and prospectively enrolled for pretreatment transcervical ultrasonography of the oropharynx and lymph nodes. Since all patients enrolled for US were HPV-positive, an additional set of patients with HPV-OPC who had both CT and MRI imaging available were selected retrospectively to compare the performance of CT and MRI in evaluating OPC. This latter group underwent primary surgical treatment at Johns Hopkins Hospital between 2009 and 2015 and had both CT and MRI imaging available. The Johns Hopkins Institutional Review Board approved all procedures described in this study.

Anatomic subsite

Base of tongue Tonsil Overlap Unknown

37 24 2 4

55 36 3 6

cT category

cTx cT1 cT2 cT3 cT4a cT4b

4 30 24 4 4 1

6 45 36 6 6 1

AJCC8 cN category

cN0 cN1 cN2 cN3

6 56 4 1

9 84 6 1

AJCC8 stage group

I II III

56 5 6

84 7 9

Primary treatment

Surgery RT ± chemo

43 24

64 36

Data collection CT and MRI images were read by a neuroradiologist (N.A.) blinded to diagnosis and with more than 15 years of experience. Separate standardized CT/MRI data collection forms documented imaging modality and quality, anatomic subsite, lesion size in craniocaudal (CC), anteroposterior (AP), and mediolateral (ML) dimensions, number of radiologically suspicious lymph nodes, and CC, AP, and ML dimensions of the largest radiologically suspicious lymph node. Prospectively enrolled patients underwent ultrasonography per standardized protocol for imaging oropharyngeal structures as previously described [17]. Prior to ultrasound, the study coordinator and sonographer reviewed patient clinical records, including standard imaging, biopsy, and medical history. Sonographic imaging included transverse, sagittal, and parasagittal still images and cine clips of the BOT, bilateral tonsils, and largest suspicious lymph node. Ultrasonography was performed by trained sonographers (M.B.W., E.S.P., and D.B.) with at least 5 years of experience in general ultrasound. The sonographers received specialized training in oropharyngeal transcervical ultrasound imaging. The Philips iU22 or Philips EPIQ7 (Koninklijke Philips N.V., Amsterdam, Netherlands) ultrasound systems were used with the C5-1, C8-5, L12-5, and X6-1 transducers. Data was collected on standardized US data collection forms. During ultrasound imaging, the sonographer provided relevant information to the study coordinator (F.F.), who completed the standardized data collection form. Clinical data including patient age, sex, race, clinical tumor and lymph node stage, primary treatment modality, greatest tumor dimension on pathological inspection were abstracted from patient electronic medical records.

0.41–0.6 (moderate), 0.61–0.80 (substantial), 0.81–1.00 (near perfect) [19].

Results Patient characteristics The study population (n = 67) included 43 patients prospectively enrolled for ultrasonography and an additional 24 patients with available CT and MRI. The study population included 60 (90%) men and 61 (91%) white patients ranging from 42 to 81 years in age (Table 1). Diagnostic biopsies confirmed 37 (55%) base of tongue (BOT) tumors, 24 (36%) tonsil tumors, and 2 (3%) tumors involving both the BOT and tonsil (overlapping lesions). Four primary tumors (6%) were not clinically identified and considered tumors of unknown primary origin. All tumors were HPV-positive. Most primary tumors were cT1 (45%) or cT2 (36%). Forty-three patients (64%) were treated with primary surgery and 24 (36%) with primary radiotherapy.

Standard imaging characteristics Sixty-three patients had available CT imaging (Table 2). Eighty-six percent of CT studies were good (54%) or fair (32%) quality. The 5 (8%) CT studies performed without intravenous contrast were considered poor or fair quality. CT failed to identify 11 primary lesions, resulting in an 82.5% overall detection rate. Excluding non-contrast CT studies (of poor or fair quality), the detection rate was 82.8% (48 of 58). Gadolinium-based contrast-enhanced MRIs were available for 36 patients, of which 65% were good quality (Table 2). The single noncontrast enhanced MRI was considered fair quality. The primary tumor detection rate by MRI was 71.4% (25 of 35).

Statistical analysis Statistical analyses were performed with GraphPad Prism® 6.0 g. Scatterplots were generated for cross-modality comparisons of tumor sizes and volumes. Tumor volumes were calculated assuming ellipsoid 4 shape: Volume = 3 π (CC)(AP)(ML). Correlation analyses were performed with Spearman’s test. Anatomic subsite concordance was defined as the reliability of identifying the same anatomic subsite with different imaging modalities and was determined by generating contingency tables to calculate Cohen’s kappa for inter-imaging-modality agreement [18]. Strength of agreement is indicated by the following ranges of kappa values: < 0 (poor), 0–0.20 (slight), 0.21–0.4 (fair), 178

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

MRI were compared. There were statistically significant correlations in all dimensions (Fig. 2A). Correlations in the craniocaudal (CC), anteroposterior (AP), and mediolateral (ML) dimensions were RhoCC = 0.51 (p = 0.038), RhoAP = 0.81 (p < 0.001), and RhoML = 0.57 (pML = 0.012), respectively. To determine whether large outlier tumors were driving correlations, patients with clinical tumor stage 4 (cT4) disease (n = 2) were excluded. Correlations remained significant in the anteroposterior dimension (RhoAP = 0.77, pAP = 0.0008) and but not in craniocaudal and mediolateral (RhoCC = 0.48, pCC = 0.068; RhoML = 0.45, pML = 0.085), suggesting that the correlations in CC and ML dimensions were driven by cT4 tumors. Next, correlation analyses were stratified by anatomic subsite to determine whether subsite-specific differences in tumor imaging were an influence. BOT tumors (n = 10) showed significant correlation between CT and MRI in the AP and ML but not in CC (RhoCC = −0.076, pCC = 0.846; RhoAP = 0.80, pAP = 0.007; RhoML = 0.71, pML = 0.025). Since no BOT tumors were cT4, analysis excluding cT4 samples was not performed. Tonsillar tumors (n = 6) demonstrated no significant correlations between CT and MRI (p-value ranges: 0.48–0.67). To understand if discrepancies in primary tumor size by CT and MRI may have been a result of modality-dependent resolution of tumor borders at the primary site, we next evaluated analogous correlations for the largest lymph node, a structure that is typically well-defined on imaging. Thirty patients exhibited radiologically suspicious nodal disease. Lymph node dimensions in all orientations demonstrated strong and highly significant correlations (Fig. 2B; RhoCC = 0.70, pCC < 0.0001; RhoAP = 0.90, pAP < 0.0001; RhoML = 0.84, pML < 0.0001).

Table 2 Standard imaging characteristics. Overall n

%

CT contrast

Yes Noa

58 5

92 8

CT study quality

Good Fair Poor

34 20 9

54 32 14

CT site

Base of tongue Tonsil Overlap Not identified

34 14 4 11

54 22 6 17

MRI contrast

Yes Nob

36 1

97 3

MRI study quality

Good Fair Poor

24 12 1

65 32 3

MRI site

Base of tongue Tonsil Overlap Not identified

17 6 2 12

46 16 5 32

a b

Quality: 3 poor, 2 fair. Quality: 1 fair.

Subsite concordance of CT, MRI, and biopsy Anatomic subsite concordance between diagnostic biopsy, CT, and MRI was evaluated. Overall agreement for anatomic subsite was 69.7% (biopsy vs CT), 65.6% (CT vs MRI), and 53.1% (biopsy vs MRI) (Fig. 1A). After accounting for concordance by chance, agreement for biopsy as compared to MRI (κ = 0.29) remained lower than biopsy vs. CT (κ = 0.55) and CT vs. MRI (κ = 0.59) (Fig. 1B).

Correlation in tumor dimensions by pathology and imaging Forty-three patients had surgical resections, enabling evaluation of the relationship between tumor size on imaging and actual tumor size. Correlations were assessed between greatest dimension on imaging and the primary tumor on gross pathological examination. Both CT and MRI demonstrated significant correlations in tumor size with respect to pathology (RhoCT = 0.56, pCT = 0.002, nCT = 28; RhoMRI = 0.55, pMRI = 0.01, nMRI = 19; Fig. 2C and D).

Correlation in tumor dimensions by CT and MRI Of the 36 patients with MRI and CT, a primary lesion was identified on both in 50% (n = 18). Tumor dimensions as determined by CT and

Fig. 1. Comparison of oropharyngeal tumor anatomic subsite concordance. (A) Bar chart comparing crossmodality anatomic subsite concordance by CT, MRI, and biopsy. (B) Bar chart summarizing Cohen’s kappa values for cross-modality anatomic subsite concordance by CT, MRI, and biopsy. (C) Bar chart comparing cross-modality anatomic subsite concordance by ultrasound (US), standard imaging (SI), and biopsy. (D) Bar chart summarizing Cohen’s kappa values for cross-modality anatomic subsite concordance by US, SI, and biopsy.

179

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

Fig. 2. Comparison of CT and MRI in estimating unidimensional tumor and lymph node lengths. (A) Comparison of unidimensional size estimates by CT and MRI for oropharyngeal tumors in craniocaudal (left), anteroposterior (middle), and mediolateral (right) dimensions. (B) Comparison of unidimensional size estimates by CT and MRI for Group 1 lymph nodes in craniocaudal (left), anteroposterior (middle), and mediolateral (right) dimensions. Comparison of greatest dimension of oropharyngeal tumor by (C) pathology versus CT, (D) pathology versus MRI. Abbreviations: m, slope of regression line; ρ , Spearman’s rank correlation coefficient; mm, millimeters.

Normal palatine tonsils were visualized in 66.7% of patients. All sonographically visualized tonsils were hypoechoic; 75.8% also contained heterogeneously distributed echogenic foci. These echogenic foci were likely indicative of tonsillar crypts containing air. Sixteen of 40 patients (40%) with ascertainable history had tonsillectomy in childhood. Ultrasonography failed to visualize palatine tonsillar tissue in 68.8% of patients with a history of pediatric tonsillectomy but only 12.5% of patients without history of pediatric tonsillectomy.

Ultrasound imaging characteristics Forty-three patients underwent ultrasonography of the oropharynx and cervical lymph nodes with a standardized protocol [17] and prospective data collection. The median time between standard and ultrasound imaging was 16.5 days. In all cases, parts of the BOT uninvolved by tumor were visualized and noted to have a characteristic heterogeneous echotexture (Table 3). 180

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

Table 3 Ultrasound imaging characteristics. Patients 1–16

Patients 17–43

p

n

%

n

%

16 0

100 0

27 0

100 0

1.00

Overall n

%

43 0

100 0

Normal BOT visualized by US

Yes No

Echotexture of normal BOT

Heterogeneous

16

100

27

100

1.00

43

100

Normal tonsils visualized by US

Yes No

11 5

69 31

18 9

67 33

1.00

29 14

67 33

Echotexture of normal tonsils

Hypoechoic Hypoechoic with echogenic foci Not identified

2 9 5

13 56 31

5 13 9

19 48 33

0.68

7 22 14

16 51 33

Mass visualized by US

Yes No

16 0

100 0

26 1

96 4

1.00

42 1

98 2

Echotexture of mass

Hypoechoic Isoechoic Not identified

15 1 0

94 6 0

25 1 1

93 4 4

1.00

40 2 1

93 5 2

Mass margin

Regular/clear Irregular/unclear Not identified

0 16 0

0 100 0

2 24 1

7 89 4

0.52

2 40 1

5 93 2

CC (RhoCC = 0.47, pCC = 0.051), and remained insignificant in ML (RhoML = 0.27, pML = 0.27). Restricting analyses to tonsillar tumors (n = 9) showed significant correlation in CC but not in AP or ML (RhoCC = 0.91, pCC = 0.004; RhoAP = 0.27, pAP = 0.48; RhoML = 0.27, pML = 0.48). The correlation in CC remained significant upon exclusion of cT4 tumors. Thirty-six patients had radiologically suspicious nodal disease. Comparison of the largest lymph node in each patient demonstrated significant correlations between US and SI (RhoCC = 0.54, pCC = 0.0009; RhoAP = 0.43, pAP = 0.008; RhoML = 0.40, pML = 0.015) (Fig. 3B).

A mass consistent with primary OPC was visualized in 42 of 43 cases, indicating an unblinded overall detection rate of 98%. Of visualized masses, 95.2% were hypoechoic, 4.8% were isoechoic, and 95.2% of masses had irregular, unclear margins (Table 3). Subsite concordance of ultrasound, standard imaging, and biopsy To determine whether ultrasound was comparable in tumor subsite assignment, US was compared with the site determined by biopsy and standard imaging (SI). Given that contrast-enhanced CT of the head and neck is more commonly available and utilized in the initial evaluation of OPC, patients with available contrast-enhanced CT (n = 39) were included in the standard imaging group. For patients without available CT imaging (n = 4), MRI studies were included as SI. Subsite agreement was highest for biopsy vs SI (72.1%), but similar for US vs SI (60.5%), and biopsy vs US (62.7%; Fig. 1C). Similar patterns of agreement were observed when considering potential agreement by chance: kappa values were 0.57 for biopsy vs SI, 0.47 for US vs SI, and 0.43 for biopsy vs US (Fig. 1D). Subsite concordance analyses were also performed separately comparing biopsy, CT, and US as well as biopsy, MRI, and US. In patients who underwent US and CT, subsite concordance by Cohen’s test was κ = 0.54 (biopsy vs CT), κ = 0.46 (US vs CT), κ = 0.38 (biopsy vs US). In patients who underwent US and MRI, subsite concordance was κ = 0.67 (biopsy vs MRI), κ = 0.58 (US vs MRI), and κ = 0.83 (biopsy vs US).

Correlation in tumor dimension by pathology and ultrasound Sixteen patients who underwent transcervical US had surgical resections. The median time between US imaging and surgical resection was 8 days. Correlations were assessed between greatest dimension on ultrasound and greatest dimension of the resected primary tumor on gross pathological examination. No correlations were observed between tumor size by US and pathology (Rho = −0.23, p = 0.39, n = 16; Fig. 3C). In this group, there were also no significant correlations in primary tumor size by standard imaging and pathology (Rho = 0.43, p = 0.16, n = 12; Fig. 3D). Correlation in tumor volume by ultrasound and standard imaging US uniquely enables imaging outside of standard orthogonal planes, making comparisons with SI difficult. US imaging of the tonsils and BOT from a transcervical approach is performed in (sagittal and transverse) oblique planes that optimize visualization and measurement of oropharyngeal structures [17]. We thus hypothesized that weaker correlations observed between US and SI relative to CT and MRI may be a result of off-axis (oblique) imaging. To address this possibility, we assumed tumors to be ellipsoid in shape and calculated volumes using the CC, AP, and ML dimensions for both US and SI [20,21]. We observed improved correlations for volumes calculated from US and SI derived dimensions (Rho = 0.66, p < 0.0001, n = 33, Fig. 4A). Correlation strength remained significant with exclusion of cT4 cases (Rho = 0.56, p = 0.0015, n = 29) (Fig. S1). Similar observations were made upon stratification by tumor subsite (Data not shown). Volume estimates also improved US and SI correlations in lymph nodes (Rho = 0.61, p < 0.0001, n = 35; Fig. 4B) and in comparisons

Correlation in tumor dimensions by ultrasound and standard imaging Next, tumor dimensions observed by SI and US were compared. Thirty-five patients had radiologically identifiable lesions on both SI and US. SI and US showed statistically significant correlations in CC and AP but not ML (RhoCC = 0.60, pCC = 0.003; RhoAP = 0.71, pAP < 0.0001; RhoML = 0.30, pML = 0.08) (Fig. 3A). Exclusion of cT4 samples showed significant correlations in CC and AP but not ML (RhoCC = 0.64, pCC < 0.0001; RhoAP = 0.48, pAP = 0.008; RhoML = 0.19, pML = 0.30), suggesting that cT4 tumors did not drive correlations between SI and US. Correlation analyses restricted to BOT tumors (n = 20) showed significant correlations in the CC and AP but not in ML (RhoCC = 0.55, pCC = 0.016; RhoAP = 0.72, pAP = 0.0004; RhoML = 0.37, pML = 0.10). Upon exclusion of cT4 BOT tumors, correlations remained significant in AP (RhoAP = 0.67, pAP = 0.0016), reached borderline significance in 181

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

Fig. 3. Comparison of Ultrasound, Standard Imaging, and Pathology in estimating unidimensional tumor and lymph node lengths. (A) Comparison of unidimensional size estimates by ultrasound (US) and standard imaging (SI) for oropharyngeal tumors in craniocaudal (left), anteroposterior (middle), and mediolateral (right) dimensions. (B) Comparison of unidimensional size estimates by US and SI of lymph nodes in craniocaudal (left), anteroposterior (middle), and mediolateral (right) dimensions. (C) Comparison of primary tumor greatest dimension by pathology and ultrasound. (D) Comparison of primary tumor greatest dimension by pathology and standard imaging (SI). Abbreviations: m, slope of regression line; ρ , Spearman’s rank correlation coefficient; mm, millimeters.

statistically significant differences in ultrasound imaging characteristics (Table 3). Correlation strength of tumor volume between US and SI increased from the training (Rho = 0.32, p = 0.32, n = 13) to the test set (Rho = 0.77, p < 0.0001, n = 20) (Fig. 4C). In addition, the slope of the regression line improved from the training (m = 0.47) to the test set (m = 0.83), indicating a reduction in tumor size overestimation. Similar improvements in correlation strength and relative accuracy from training to test set were observed on unidimensional analyses and

between CT and MRI (Fig. 4C). Ultrasound accuracy improves with experience To determine whether diminished correlations between US and SI were a result of operator dependence, we explored if US accuracy improved with experience. Patients were stratified into training (patients 1–16) and test (patients 17–43) sets. These groups exhibited no 182

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

Fig. 4. Comparison of CT, MRI, Ultrasound, and Standard Imaging in estimating tumor and lymph node volume. Comparison of ellipsoid volume estimates by ultrasound (US) and standard imaging (SI) for Group 2 (A) oropharyngeal tumors and (B) lymph nodes. Abbreviations: m, slope of regression line; ρ , Spearman’s rank correlation coefficient; mm, millimeters. Bar graphs demonstrating correlation strength ( ρ ) for orthogonal dimensions (CC, AP, and ML) and ellipsoid volume (Vol). Comparisons of CT versus MRI are represented by grey bars. Comparisons of US versus SI are represented by black bars and divided into training and test sets. Abbreviations and symbols: m, slope of regression line; ρ , Spearman’s rank correlation coefficient; mm, millimeters; n.s.: not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

that of our study [26]. Nonetheless, both CT and MRI can contribute information that improves the clinical assessment of OPC [22,23,25]. Our data suggest that transcervical ultrasound is a useful adjunct to multimodal imaging strategies currently implemented to evaluate oropharyngeal tumors. The unblinded detection rate of US was 98%. Consistent with previous findings [14,15], US demonstrated a high rate of detection of tumors unidentifiable by CT or MRI. Since sonographers were not blinded to clinical and imaging results, while the neuroradiologist was blinded, our detection rate likely overestimates the true sensitivity of US in single modality detection of OPC. Nonetheless, since US is likely to be employed in a multi-modality imaging approach to OPC, the rate of detection rate observed here is likely to reflect what is attainable in clinical practice. The assessment of oropharyngeal tumor extent with standard imaging modalities remains challenging. Anatomic subsite concordance by diagnostic biopsy, CT, and MRI was fair to moderate (κ = 0.29–0.60) [19]. The low tumor detection rate by MRI likely contributed to the weak concordance of biopsy and MRI (κ = 0.29). Comparison of US vs SI demonstrated marginally lower subsite concordance rates compared to CT vs MRI (κ = 0.47 vs. 0.55). Given that efficacy of US depends on operator skill, and our finding that performance improved in the latter group of patients, assessing anatomic subsite involvement by OPC may improve with further experience and protocol optimization. To compare accuracy of the different imaging modalities, tumor size

in BOT tumors (Fig. 4C). As expected given the sonographers’ experience in identifying and measuring abnormal neck lymph nodes, analysis of lymph nodes did not demonstrate improvement in correlation strength or significance from the training (Rho = 0.74, p = 0.005, n = 13) to the test set (Rho = 0.59, p = 0.004, n = 22) (Fig. 4D).

Discussion The data presented here support previous conclusions that no single imaging modality is sufficient to comprehensively evaluate oropharyngeal tumors [12]. Primary tumor detection was highest for US, followed by CT and MRI. Tumor dimensions varied by imaging modality, but remained significantly correlated across modalities. The primary tumor blinded detection rate of CT (83%) in our study is comparable with other reports [22,23]. Given that MRI is generally considered to have superior tissue resolution to CT [23], its relatively low tumor detection rate (71%) may be a result of a high proportion (94%) of T1 and T2 stage tumors in our study. In addition, gadoliniumbased MRI contrast can diminish the contrast between a primary lesion of the oropharynx and its surrounding normal tissue [24,25], obscuring the tumor boundaries. At least one other study has demonstrated superior identification of oropharyngeal tumors by CT compared to MRI [26]. This study also demonstrated an MRI sensitivity (69%) similar to 183

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

Conclusion

was compared in the greatest orthogonal dimension observed on imaging to tumor diameter on pathology. Correlation of greatest tumor dimension on pathological analysis versus CT and MRI were significant, while pathology versus US was not. The median time difference of 8 days between US imaging and surgical resection was unlikely to explain these discrepancies in tumor dimension. Tumor dimensions on pathological analysis were estimated during gross examination by several different pathologists. Notably, tumor measurements were often rounded to the nearest half centimeter, which is generally appropriate for pathological staging. A dedicated neuroradiologist read all images and measured with millimeter-precision. These differences in data acquisition limited this set of comparisons. Although the limitations of our approach impede confident interpretation of this data, future prospective studies could be designed to achieve uniformity in measurements across all modalities. Direct and careful comparison of tumor extent on pathology and imaging would address reports that tumor extent on surgery may differ from that on preoperative imaging [27]. Correlations in primary tumor size estimated by CT and MRI were statistically significant. However, the general strength of monotonic correlations tended to be moderate, indicating that tumor extent is not equivalent across CT and MRI. In comparing CT and MRI, the slope of the regression line tended to equal less than 1, suggesting a greater variance in tumor size as measured by CT than MRI. Whether the increased variance indicates enhanced dynamic range or greater error is not ascertainable by our study design. Analyses of the largest suspicious lymph node were included as a control, since pathologic lymph nodes in HPV-OPC tend to be superficial and exhibit high contrast from surrounding tissues. Observations that lymph nodes consistently demonstrated stronger correlations than primary tumor highlight the challenge of imaging oropharyngeal tumor with currently available modalities. Despite a median 16.5-day delay in US relative to standard imaging, unidimensional comparisons demonstrated comparable correlations in CC and AP dimensions of the primary tumor on US vs SI and CT vs MRI. In addition, lymph node size estimates were weaker in all dimensions on US vs SI than CT vs MRI, a somewhat surprising finding given that cervical lymph node ultrasonography is well-established [28]. We thus investigated if weaker correlation strengths in US vs SI reflected differences in orthogonal plane measurement between US and the other modalities by calculating volumes from measured dimensions. Estimating ellipsoid volumes resulted in improvements in US vs SI correlation strength for both tumor and lymph nodes, suggesting that the differences in measurement of orthogonal dimensions may contribute to weak correlations in US vs SI. Consistent improvements in correlation strength indicate that volume estimates compensate for unidimensional inaccuracies and stabilize correlations independent of imaging modality. The operator-dependency and learning curve of US has been extensively described. Our findings support this notion with analyses comparing the training set to the test set evaluated in this study demonstrated significant improvements in correlation strength. A greater number of patients in Set 2 may have inflated correlation strengths. However, despite the differences in sample sizes, correlation strengths for lymph node dimensions – a structure with which our sonographers had extensive experience – did not change from the training to the test set. In view of the limited experience with US to visualize OPCs, these data support the continued development of this image modality. While its properties appear to be comparable to standard imaging in this analysis, with broadening and increasing experience, there may be opportunity for improvement. The ability to characterize known OPC tumors, could serve as the foundation for potential future directions where non-invasive methods are needed, such as screening scenarios.

Our data collectively suggest that transcervical ultrasound is a sensitive and relatively accurate adjunct to standard imaging for the evaluation of oropharyngeal tumors. Its relative cost, portability, and potential for in-clinic and serial imaging render US an attractive modality to further develop for imaging oropharyngeal tumors. Financial disclosures The authors have no financial disclosures. Conflicts of interest None declared. Acknowledgements We would like to thank Beth Smith for assisting with ultrasonography and all staff in the Johns Hopkins Departments of Otolaryngology and Radiology Division of Ultrasound for supporting this study. The Oral Cancer Foundation. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.oraloncology.2018.01. 016. References [1] Chaturvedi AK, Anderson WF, Lortet-Tieulent J, Curado MP, Ferlay J, Franceschi S, et al. Worldwide trends in incidence rates for oral cavity and oropharyngeal cancers. J Clin Oncol 2013;31:4550–9. [2] Gillison ML, Chaturvedi AK, Anderson WF, Fakhry C. Epidemiology of human papillomavirus-positive head and neck squamous cell carcinoma. J Clin Oncol 2015;33:3235–42. [3] Viens LJ, Henley SJ, Watson M, Markowitz LE, Thomas CC, Thompson TD, et al. Human papillomavirus-associated cancers – United States, 2008–2012. MMWR Morb Mortal Wkly Rep 2016;65:661–6. [4] Westra WH. The morphologic profile of HPV-related head and neck squamous carcinoma: implications for diagnosis, prognosis, and clinical management. Head Neck Pathol 2012;6(Suppl. 1):S48–54. [5] Chaturvedi AK, Engels EA, Anderson WF, Gillison ML. Incidence trends for human papillomavirus-related and -unrelated oral squamous cell carcinomas in the United States. J Clin Oncol 2008;26:612–9. [6] Ang KK, Harris J, Wheeler R, Weber R, Rosenthal DI, Nguyen-Tan PF, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010;363:24–35. [7] Fakhry C, Westra WH, Li S, Cmelak A, Ridge JA, Pinto H, et al. Improved survival of patients with human papillomavirus-positive head and neck squamous cell carcinoma in a prospective clinical trial. J Natl Cancer Inst 2008;100:261–9. [8] Koch WM. Clinical features of HPV-related head and neck squamous cell carcinoma: presentation and work-up. Otolaryngol Clin North Am 2012;45:779–93. [9] Deschler DG, Richmon JD, Khariwala SS, Ferris RL, Wang MB. The “new” head and neck cancer patient-young, nonsmoker, nondrinker, and HPV positive: evaluation. Otolaryngol Head Neck Surg 2014;151:375–80. [10] Corey AS, Hudgins PA. Radiographic imaging of human papillomavirus related carcinomas of the oropharynx. Head Neck Pathol 2012;6(Suppl. 1):S25–40. [11] Bhargava P, Rahman S, Wendt J. Atlas of confounding factors in head and neck PET/CT imaging. Clin Nucl Med 2011;36:e20–9. [12] Vergez S, Moriniere S, Dubrulle F, Salaun PY, De Mones E, Bertolus C, et al. Initial staging of squamous cell carcinoma of the oral cavity, larynx and pharynx (excluding nasopharynx). Part I: locoregional extension assessment: 2012 SFORL guidelines. Eur Ann Otorhinolaryngol Head Neck Dis 2013;130:39–45. [13] Dammann F, Horger M, Mueller-Berg M, Schlemmer H, Claussen CD, Hoffman J, et al. Rational diagnosis of squamous cell carcinoma of the head and neck region: comparative evaluation of CT, MRI, and 18FDG PET. AJR Am J Roentgenol 2005;184:1326–31. [14] Mydlarz WK, Liu J, Blanco R, Fakhry C. Transcervical ultrasound identifies primary tumor site of unknown primary head and neck squamous cell carcinoma. Otolaryngol Head Neck Surg 2014;151:1090–2. [15] Fakhry C, Agrawal N, Califano J, Messing B, Liu J, Saunders J, et al. The use of ultrasound in the search for the primary site of unknown primary head and neck squamous cell cancers. Oral Oncol 2014;50:640–5. [16] Blanco RG, Califano J, Messing B, Richmon J, Liu J, Quon H, et al. Transcervical

184

Oral Oncology 78 (2018) 177–185

F. Faraji et al.

[17]

[18] [19] [20]

[21]

[22]

J Comput Assist Tomogr 1999;23:43–9. [23] Bootz F, Lenz M, Skalej M, Bongers H. Computed tomography (CT) and magnetic resonance imaging (MRI) in T-stage evaluation of oral and oropharyngeal carcinomas. Clin Otolaryngol Allied Sci 1992;17:421–9. [24] Phillips CD, Gay SB, Newton RL, Levine PA. Gadolinium-enhanced MRI of tumors of the head and neck. Head Neck 1990;12:308–15. [25] Heissler E, Steinkamp HJ, Heim T, Zwicker C, Felix R, Bier J. Value of magnetic resonance imaging in staging carcinomas of the oral cavity and oropharynx. Int J Oral Maxillofac Surg 1994;23:22–7. [26] Niederhagen B, Leipner N, Berge S, von Lindern JJ, Appel T. Value of computerized tomography and magnetic resonance tomography in diagnosis of malignancies of the mouth cavity and oropharynx. Mund Kiefer Gesichtschir 2000;4:217–21. [27] Liu WP, Reaugamornrat S, Sorger JM, Siewerdsen JH, Taylor RH, Richmon JD. Intraoperative image-guided transoral robotic surgery: pre-clinical studies. Int J Med Robot 2015;11:256–67. [28] Ahuja AT, Ying M. Sonographic evaluation of cervical lymph nodes. AJR Am J Roentgenol 2005;184:1691–9.

ultrasonography is feasible to visualize and evaluate base of tongue cancers. PLoS One 2014;9:e87565. Coquia SF, Hamper UM, Holman ME, DeJong MR, Subramaniam RM, Aygun N, et al. Visualization of the oropharynx with transcervical ultrasound. AJR Am J Roentgenol 2015;205:1288–94. Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Measur 1960;20:37–46. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–74. Sorensen AG, Patel S, Harmath C, Bridges S, Synnott J, Sievers A, et al. Comparison of diameter and perimeter methods for tumor volume calculation. J Clin Oncol 2001;19:551–7. Dejaco D, Url C, Schartinger VH, Haug AK, Fischer N, Riedl D, et al. Approximation of head and neck cancer volumes in contrast enhanced CT. Cancer Imaging 2015;15:16. Leslie A, Fyfe E, Guest P, Goddard P, Kabala JE. Staging of squamous cell carcinoma of the oral cavity and oropharynx: a comparison of MRI and CT in T- and N-staging.

185