Print
facebook
twitter
Linkedin
Introduction
Cervical lymph node (LN) metastasis is common in thyroid cancer, with a reported prevalence of up to 80% [1–4]. LN metastasis significantly increases the risk of local recurrence and often necessitates repeat surgery [5–9]. Therefore, accurate preoperative detection of metastatic LNs is critical for determining the appropriate extent of neck dissection and minimizing the risk of persistent disease after surgery [10].
Current guidelines from the National Comprehensive Cancer Network and the American Thyroid Association recommend ultrasound (US) as the primary imaging modality for cervical LN evaluation in thyroid cancer [10,11]. On US, suspicious LNs are identified by imaging features including cystic changes, echogenic foci (EF), cortical hyperechogenicity, and abnormal vascularity [2,10,12,13]. Biopsy is indicated when any of these features are present and the LN exceeds the size threshold [10,12,13]. Although US demonstrates high specificity, its sensitivity remains relatively low [14–16]. The limitations of US, including operator dependency and reduced efficacy in visualizing deep or artifact-prone anatomical structures, highlight the need for complementary imaging techniques [17–20].
Computed tomography (CT) is often employed as an adjunct to US in advanced thyroid cancer [10,11]. CT identifies suspicious LNs based on features analogous to those observed on US, including strong or heterogeneous enhancement, cystic changes, and calcifications [2,12,21,22]. CT is especially useful in challenging areas, such as the retropharyngeal and superior mediastinal regions [18].
The combination of CT and US increases sensitivity with the trade-off of decreased specificity [2,18,22,23]. Although the current consensus considers an LN suspicious on either US or CT to be malignant, this approach may overlook the diagnostic value of evaluating concordant versus discordant findings [2,18,23]. To date, few studies have performed node-by-node comparisons of US and CT imaging features to clarify their complementary roles. Therefore, this study aimed to analyze the malignancy rate of LNs according to the concordance of US and CT imaging features in patients with thyroid cancer.
Materials and Methods
Compliance with Ethical Standards
This study was approved by the relevant institutional review board of Seoul National University Hospital (IRB No. 1506-107-682). Because the study was retrospective, the requirement for informed consent was waived.
Study Participants
This study consecutively enrolled patients who underwent US-guided fine-needle aspiration (FNA) or core-needle biopsy (CNB) at a single tertiary referral institution from December 2006 to June 2015 (n=55,276). The exclusion criteria were biopsy performed for non-LN neck lesions (n=47,427), non-thyroid malignancy (n=6,768), previous thyroidectomy (n=793), or absence of corresponding CT (n=53). Ultimately, 277 LNs in 228 patients were included in the final analysis (Fig. 1). Data from these participants were previously used to validate US- and CT-based risk stratification systems and biopsy criteria for cervical LNs in preoperative patients with thyroid cancer [24].
US and CT Image Acquisition
US was performed at the time of biopsy using linear transducers (7.5–15 MHz) by board-certified head and neck radiologists or radiology residents under faculty supervision. For each biopsied LN, detailed anatomical localization was recorded, including the cervical LN level (II–VI) and the relationship of the LN to major vessels (e.g., the common carotid artery and internal jugular vein) and neck muscles (including the sternocleidomastoid and strap muscles). Longitudinal and transverse images were obtained to enable three-dimensional assessment of LN position and morphology.
Axial CT images were obtained using 16- to 128-channel multidetector CT scanners with a 2.5–3 mm slice thickness. Post-contrast images were acquired 40 seconds after injection of 90 mL of an iodinated nonionic contrast agent (300–350 mgI/mL) at 3 mL/s using an automated injector. Pre-contrast CT images were also obtained to better depict calcifications. The median time interval between the US and CT examinations was 19 days (interquartile range, 9 to 40 days).
Image Analysis and LN Classification
To ensure accurate matching of each LN on US and CT, a radiologist with 10 years of experience in thyroid imaging identified the corresponding LN on both modalities based on cervical LN level, its relationship to adjacent anatomic structures, and LN size and shape. Only LNs that were confidently identified on both US and CT were included in the analysis. After a 6-month washout period, two radiologists with 10 and 21 years of experience in thyroid imaging, respectively, independently analyzed imaging features on each modality while blinded to the findings of the other modality, as well as to the clinical data and pathological results. Discrepancies in imaging feature assessment were resolved by consensus.
Based on the Korean Society of Thyroid Radiology risk stratification system for cervical LNs, suspicious features on US were defined as cystic changes (US-cystic), echogenic foci (US-EF), and cortical hyperechogenicity or abnormal vascularity (US-echo-vascularity) [12]. On CT, suspicious features included cystic changes (CT-cystic), calcifications (CT-calcification), and strong or heterogeneous enhancement (CT-enhancement) [12] (Fig. 2). A dichotomized LN classification was applied for the US- and CT-based categories: an LN was considered suspicious if any of the aforementioned features were present and non-suspicious if none were present [2,10,12,13].
To analyze concordance between US and CT features, suspicious features were matched between modalities and categorized as positively concordant if the matched features were present on both US and CT, negatively concordant if they were absent on both US and CT, and discordant if they were present on one modality but absent on the other (Fig. 1). Features were matched as follows: US-echo-vascularity with CT-enhancement, US-cystic with CT-cystic, and US-EF with CT-calcification. For US-EF, findings were classified as punctate (≤1 mm) or large (>1 mm), while CT-calcification findings were classified as microcalcification (≤1 mm) or macrocalcification (>1 mm).
Cytopathological Diagnosis for Cervical LNs
Board-certified radiologists performed US-guided biopsies after obtaining grayscale and Doppler images of the target LNs. Biopsies were performed for US-suspicious or indeterminate LNs (lacking an echogenic hilum and hilar vascularity) at the radiologist’s discretion, as well as for benign-appearing LNs (exhibiting an echogenic hilum and hilar vascularity) [12] at the clinician’s request. FNA with thyroglobulin measurement was performed using a 23-gauge needle up to three times. In selected cases, CNB was performed using an 18-gauge double-action, spring-activated biopsy needle with a 1.1 cm excursion length (TSK Acecut, Create Medic, Yokohama, Japan). LNs were considered malignant if malignancy was confirmed by cytopathology or if the thyroglobulin concentration in the aspiration sample exceeded the cutoff of 8.3 ng/mL, which served as the reference standard [21,25].
Statistical Analysis
LN characteristics were summarized using medians with standard deviations for continuous variables and percentages for categorical variables. Inter-reader agreement for LN suspicion category and imaging features on US and CT was assessed using the Cohen κ. Values <0.20 indicated slight agreement, 0.21–0.40 fair agreement, 0.41–0.60 moderate agreement, 0.61–0.80 substantial agreement, and 0.81–1.00 near-perfect agreement.
To determine concordance rates between US and CT imaging features, the proportion of LNs with concordant CT findings among those with corresponding US findings was calculated. Malignancy rates based on US-CT concordance were calculated as the number of pathologically confirmed malignant LNs divided by the total number of LNs with corresponding imaging features. To account for clustering due to multiple LNs per patient, generalized estimating equations with a binomial distribution were used to compare US-CT concordance rates and malignancy rates across concordance categories. Raw data showing frequencies for specific imaging features are provided in Supplementary Tables 1 and 2. Additionally, the frequencies of malignant and benign LNs according to specific US or CT imaging features were compared using the Fisher exact test.
For cases with discordant US-echo-vascularity and CT-enhancement findings, diagnostic performance metrics and odds ratios were calculated for each modality and compared using the McNemar test.
All statistical analyses were performed using SPSS version 23.0 (IBM Corp., Armonk, NY, USA). All tests were two-sided, and P-values of less than 0.05 were considered to indicate statistical significance.
Results
Clinical Characteristics of Cervical LNs
Table 1 summarizes the clinical characteristics of the study participants. A total of 277 cervical LNs from 228 patients with thyroid cancer were analyzed (median age, 47.4 years; 73.3% female). The primary cancer type was predominantly papillary thyroid carcinoma (218, 95.6%), with other subtypes including follicular (n=2, 0.9%), medullary (n=5, 2.2%), anaplastic (n=2, 0.9%), and undifferentiated (n=1, 0.4%) thyroid carcinomas.
Biopsy of the LNs was performed based on US imaging features or at the clinician’s request: FNA with thyroglobulin measurement was performed in 227 LNs, CNB in 36 LNs, and both procedures in 14 LNs. Final cytopathological results confirmed malignancy in 147 (53.1%) LNs and benign findings in 130 (46.9%). Among 227 thyroglobulin washout samples, 113 exceeded the 8.3 ng/mL cutoff, and 17 LNs were diagnosed as malignant based on thyroglobulin level alone.
Malignancy Rate by Concordance of Suspicion Categories on US and CT
The two radiologists demonstrated excellent inter-reader agreement for LN categorization, with Cohen κ values of 0.80 for US and 0.88 for CT. Based on the suspicion criteria for US and CT, 136 LNs (49.1%) were categorized as positively concordant (suspicious on both modalities), 40 (14.4%) as discordant (suspicious on one modality only), and 101 (36.5%) as negatively concordant (non-suspicious on both modalities).
A clear hierarchy was observed in malignancy rates across the concordance groups. Positively concordant LNs had a significantly higher malignancy rate (91.2% [124/136]) than discordant LNs (42.5% [17/40], P<0.001). Discordant LNs, in turn, had a significantly higher malignancy rate than negatively concordant LNs (5.9% [6/101], P<0.001). No significant differences in age, sex, or LN distribution were identified among the positively concordant, discordant, and negatively concordant groups.
Concordance and Discordance of Suspicious US and CT Features
The two radiologists demonstrated substantial to near-perfect inter-reader agreement for suspicious imaging features on US and CT. US-echo-vascularity and CT-enhancement showed substantial interobserver agreement (κ=0.74 and κ=0.78, respectively). US-cystic and CT-cystic showed near-perfect interobserver agreement (κ=0.98 for both). US-EF and US-punctate EF also showed near-perfect interobserver agreement (κ=0.90 and κ=0.88, respectively), as did US-large EF, CT-calcification, CT-microcalcification, and CT-macrocalcification (κ=1.00 for all).
Table 2 presents concordance rates between US and CT suspicious features for cervical LNs. The unadjusted results are provided in Supplementary Table 1. US-echo-vascularity and CT-enhancement displayed 78.8% (219/277; 95% confidence interval [CI], 73.4% to 83.3%) concordance. US-cystic and CT-cystic features exhibited 90.4% (250/277; 95% CI, 86.1% to 93.4%) concordance, and US-EF and CT-calcification showed 72.3% (199/277; 95% CI, 66.4% to 77.5%) concordance. Fig. 2 illustrates representative cases with concordant US and CT features.
Concordance between US-EF and CT-calcification varied significantly by size. Only 6.5% (6/92; 95% CI, 3.0% to 13.6%) of US-punctate EF corresponded to CT-microcalcification, indicating low positive concordance for submillimeter lesions. In contrast, US-large EF (>1 mm) showed excellent agreement with CT-macrocalcification, with a positive concordance rate of 87.0% (20/23; 95% CI, 66.4% to 95.7%) and a negative concordance rate of 99.2% (252/254; 95% CI, 96.9% to 99.8%).
Among all LNs, 97 (35.0%) exhibited more than one suspicious feature on US, while 64 (23.1%) showed more than one suspicious feature on CT.
Malignancy Rate by Concordance of Imaging Features on US and CT
Table 3 presents the malignancy rate of cervical LNs based on concordance of suspicious features between US and CT. The unadjusted rates are provided in Supplementary Table 2. Furthermore, the frequency of malignant and benign LNs according to each US or CT feature is provided in Supplementary Table 3. When analyzed individually, the presence of any suspicious feature on US or CT was associated with a higher malignancy rate than in the negative group.
Malignancy rates in positively concordant cases were significantly higher than those in negatively concordant cases across all imaging features (US-echo-vascularity/CT-enhancement: 93.0% [101/108; 95% CI, 87.1% to 96.4%] vs. 8.5% [9/111; 95% CI, 4.5% to 15.5%], P<0.001; US-cystic/CT-cystic: 90.7% [34/36; 95% CI, 76.8% to 96.6%] vs. 42.7% [87/214; 95% CI, 35.9% to 49.9%], P<0.001; and US-EF/CT-calcification: 83.8% [22/26; 95% CI, 65.2% to 93.4%] vs. 34.7% [58/173; 95% CI, 27.7% to 42.5%], P<0.001).
For discordant cases involving US-cystic/CT-cystic features and US-EF/CT-calcification, malignancy rates were comparable to those in positively concordant cases (US-cystic/CT-cystic: 91.6% [26/27; 95% CI, 77.3% to 97.2%] vs. 90.7% [34/36; 95% CI, 76.8% to 96.6%], P=0.735; and US-EF/CT-calcification: 83.3% [67/78; 95% CI, 74.3% to 89.6%] vs. 83.8% [22/26; 95% CI, 65.2% to 93.4%], P=0.873), suggesting that a positive finding on either modality was sufficient to indicate malignancy. In contrast, discordant cases of US-echo-vascularity and CT-enhancement showed an intermediate malignancy rate (65.0% [37/58; 95% CI, 51.9% to 76.1%]), which was significantly lower than that in positively concordant cases (93.0% [101/108; 95% CI, 87.1% to 96.4%], P<0.001) but significantly higher than that in negatively concordant cases (8.5% [9/111; 95% CI, 4.5% to 15.5%], P<0.001). Fig. 3 illustrates representative cases with discordant US and CT features.
Further analyses were performed to assess diagnostic performance in cases with discordant US-echo-vascularity and CT-enhancement findings (Table 4, Fig. 3A–D). In this group, CT-enhancement demonstrated superior diagnostic performance compared to US-echo-vascularity, with a higher odds ratio (3.896 [95% CI, 1.188 to 12.775] vs. 0.257 [95% CI, 0.078 to 0.842], P=0.002), sensitivity (81.1% [30/37; 95% CI, 65.3% to 90.7%] vs. 18.9% [7/37; 95% CI, 9.3% to 34.7%], P<0.001), and accuracy (70.2% [40/58; 95% CI, 57.2% to 80.6%] vs. 29.8% [18/58; 95% CI, 19.4% to 42.8%], P<0.001). Specificity was comparable between US and CT (47.4% [10/21; 95% CI, 27.8% to 67.9%] vs. 52.6% [11/21; 95% CI, 32.1% to 72.2%], P=0.608). In this discordant group, CT demonstrated a higher diagnostic odds ratio than US (OR = 3.896 vs 1, P=0.020).
Discussion
This study analyzed concordance between suspicious cervical LN features on US and CT and their association with malignancy rate. Overall agreement between US and CT imaging features was good. Notably, among LNs with discordant US-echo-vascularity/CT-enhancement findings, LNs with CT-enhancement showed a higher malignancy rate than those with US-echo-vascularity.
US-echo-vascularity and CT-enhancement were grouped because intranodal hyperechogenicity on US, particularly when eccentrically located in the cortex, is known to indicate tumor cell infiltration from the periphery of the LN, with displacement of normal nodal structures [15,17]. Tumor proliferation is often accompanied by angiogenesis, which can manifest as abnormal vascular patterns on color Doppler US or contrast-enhanced US, as well as strong or heterogeneous enhancement on CT [26–29].
Cystic changes, which appear as anechoic areas on US and low-attenuation foci on CT, typically indicate necrosis within metastatic LNs [17,30]. Evaluation of cystic change can be complementary, with US providing higher resolution and CT enabling objective measurement of Hounsfield units.
EF on US are often attributable to psammoma bodies, dystrophic calcifications, or eosinophilic colloids [31–33]. This study showed that most punctate EF observed on US were not detected on CT, despite the high attenuation of calcifications readily detectable on CT. This discrepancy is likely related to the lower spatial resolution of CT with 2.5–3 mm slice thickness in the institutional thyroid CT protocol, compared with the 0.4–0.7 mm in-plane pixel size of US. These findings suggest that EF on US should still be considered suspicious even when calcification is not visible on CT. In contrast, EF >1 mm showed excellent concordance with CT calcification, suggesting that CT can provide an objective and easily interpretable representation of large EF observed on US.
Previous studies that classified suspicious US and CT imaging features as indicative of LN metastasis have suggested that these features are highly specific [2,12]. However, several studies have revealed that specificity varies by feature, with US hyperechogenicity and abnormal vascularity being less specific than punctate EF or cystic changes [34,35]. The relatively lower diagnostic accuracy of hyperechogenicity on ultrasonography may be attributable to operator competency, US system performance, artifacts, and shadowing. These factors can interfere with the accurate assessment of LN echogenicity relative to adjacent neck muscles, which often exhibit inhomogeneous echogenicity, and may result in missed small or micrometastatic foci, as well as misinterpretation between irregularly shaped normal hilum and hyperechogenic tumorous portions [36]. For abnormal vascularity, the accuracy of Doppler examination is likely influenced by machine settings. In contrast, on CT, strong or heterogeneous enhancement is known to be highly specific [37]. These factors may help explain why CT outperformed US in discordant cases of US-echo-vascularity and CT-enhancement.
This is the first study to directly match specific suspicious features on US and CT for cervical LNs on a node-by-node basis and quantify the associated malignancy rates. This approach revealed feature-specific concordance patterns, including poor correlation between punctate EF on US and calcifications on CT. The findings also showed that cystic changes or EF/calcifications detected on a single modality were sufficient to suggest malignancy, which supports prior work describing the complementary roles of US and CT [18,22,23,38]. Notably, CT-enhancement outperformed US-echo-vascularity in discordant cases. These findings may help guide surgical decision-making when US and CT findings conflict.
This study had several limitations. The retrospective, single-center design and relatively small sample size may have introduced selection bias. Generalizability to other institutions may be limited by technical variations in US and CT acquisition protocols, which can affect diagnostic performance [39]. Furthermore, the fixed axial and coronal orientations of CT necessitate cautious comparison with the adjustable imaging planes of US. In addition, the analysis focused exclusively on suspicious findings on US and CT, and the indeterminate category was excluded for simplicity. In the dataset, indeterminate nodes were operationally regarded as non-suspicious, which may have overestimated the malignancy rate in the concordant non-suspicious group compared with what might be expected if separate benign concordant categories were analyzed. Additionally, the research did not account for potential interactions among suspicious imaging features. The observed malignancy rates may have been influenced by the presence of other coexisting suspicious features. Finally, given the retrospective design, the possibility cannot be excluded that not all suspicious findings were captured or comprehensively documented during ultrasonography examinations, although radiologists were specifically focused on identifying abnormal LNs. Future prospective multicenter studies with larger cohorts and standardized imaging protocols are needed to better establish the clinical utility of combining US and CT for LN evaluation.
In conclusion, positive concordance of all US-CT features, as well as discordance involving cystic changes or US-EF/CT-calcification, suggested LN metastasis in patients with thyroid cancer. CT-enhancement was particularly reliable when discordant with US-echo-vascularity.
Download Citation
PDF Links
PubReader
XML Download
Full text via DOI
Full text via PMC
