This prospective study was approved by the ethics committee of our institutional review board, and written informed consent was obtained from each patient. We calculated the sample size (21, 22) and prospectively recruited 110 patients with suspicious PTC for preoperative DECT evaluation from December 2020 to October 2021. The patients were routinely followed up through serology and imaging examinations (1). Inclusion criteria were as follows: 1) final diagnosis of PTC was made based on postoperative pathology; 2) cervical LN dissection was conducted and pathologically analyzed; 3) no history of other cancers; and 4) image quality was adequate for subsequent analysis. Finally, 54 patients (13 men, 41 women; mean age, 33 years; age range, 25–61 years) with PTC were enrolled in the study.

Taking the American Joint Committee on Cancer cervical regional lymph node-level system as a reference, cervical LNs were divided into 7 levels (23). According to the pathological results, cervical LNs were classified as non-metastatic or metastatic LNs. The specific process of cervical LN labeling and grouping was as follows: If the harvested LNs at one cervical level were pathologically negative, all the identified LNs on CT images at this level were grouped as non-metastatic. If the harvested LNs at one cervical level were pathologically positive, all the identified LNs on CT images at this level were grouped as metastatic. If one cervical level contained both non-metastatic and metastatic LNs, all the LNs on CT images at this level were excluded from subsequent analysis (4, 19). To reduce the risk of false “all positive” or “all negative” cervical levels, if only one LN within one specific cervical level was harvested and reported in the pathological result, this level would also be discarded. To avoid the partial volume effect, only the LNs with a maximal diameter in the short axis of more than 5 mm were enrolled in this study. Finally, 157 cervical LNs (81 non-metastatic and 76 metastatic) were analyzed.

All patients underwent neck CT examinations on a dual-layer spectral detector CT (IQon spectral CT, Philips Healthcare). Detailed acquisition parameters were as follows: tube voltage, 120 kVp; tube current, modulated by automated radiation dose control; collimation, 64 × 0.625 mm; rotation time, 0.5 s; pitch factor, 1.1. Triphasic acquisition protocol included unenhanced, early arterial phases, and equilibrium phases at 3 mm slice thickness using the spectral mode. For contrast-enhanced scanning, 75 ml of contrast agent (iopromide; Bayer HealthCare) at a flow rate of 3.5 ml/s was intravenously delivered through the right elbow vein, followed by a bolus injection of 25 ml of saline at the same flow rate. The delay times for the early arterial and equilibrium phases were fixed at 25 and 180 s after the unenhanced scan. All the original data were reconstructed into contiguous axial images with a slice thickness of 1 mm for further analyses. Post-processing workstation (IntelliSpace Protal, Version 10, Philips Healthcare) was used for image reconstruction. The 120-kVp equivalent blended images were reconstructed through an iterative algorithm (iDose⁴ [level 3], Philips Healthcare), while iodine maps were reconstructed through a spectral reconstruction algorithm (Spectral [level 3], Philips Healthcare).

To calculate radiation exposure, CT dose index (CTDIvol) and dose-length product (DLP) were recorded for each patient. The mean volume of CTDIvol and DLP using unenhanced CT images was 5.3 ± 1.1 mGy (range, 4.3–7.7 mGy) and 141.8 ± 30.9 mGy × cm (range, 117.9–199.7 mGy × cm), respectively. The mean volume CTDIvol and DLP of using dual-phase enhanced CT images was 10.4 ± 2.9 mGy (range, 8.4–14.6 mGy) and 291.5 ± 64.0 mGy × cm (range, 224.5–403.0 mGy × cm), respectively. The total mean volume of CTDIvol and DLP was 15.7 ± 3.8 mGy (range, 12.7–22.3 mGy) and 433.3 ± 94.7 mGy × cm (range, 342.4–602.7 mGy × cm), respectively.

Qualitative image analysis was conducted by two radiologists (with 6 and 5 years of experience in head and neck radiology, respectively) who were blinded to the study design, based on the 120-kVp equivalent blended CT images in both arterial and equilibrium phase. If disagreements occurred, another radiologist (with 26 years of experience in head and neck imaging) would make the final decision. All these readers were blinded to the final clinicopathologic results. Positive findings for cervical LN metastasis were as follows: 1) a maximal diameter in short axis of ≥10 mm; 2) strong or heterogeneous enhancement was abnormal enhancement; 3) calcification; 4) cystic change; and 5) extra nodal extension, which was defined as obscure boundary or invasion into adjacent structures (4, 6). Cervical LNs affected by cervical root artifact on the contrast-enhanced CT images in the arterial phase were also recorded.

Attenuation values on unenhanced and equilibrium phase 120-kVp equivalent blended images were measured for ECV_S. Iodine density (ID) on equilibrium phase iodine maps was measured for ECV_D. Regions of interest (ROIs) were manually drawn on the slices encompassing the whole LN. Calcification and cystic regions were excluded with reference to the unenhanced CT images. Another circular ROI (about 15 ± 5 mm²) was drawn on the ipsilateral common carotid artery (CCA) for normalization. ECV_S and ECV_D were calculated using the following formula: 1) ECV_S (%) = (1 − hematocrit) × (ΔHU_LN/ΔHU_CCA) ×100. ΔHU_LN and ΔHU_CCA were Hounsfield units in the equilibrium phase minus in the unenhanced phase of LN and CCA, respectively; 2) ECV_D (%) = (1 − hematocrit) × (ID_LN/ID_CCA) × 100. ID_LN and ID_CCA had iodine densities in the equilibrium phase of LN and CCA, respectively (15) ( Figure 1 ). All the quantitative measurements were conducted by two radiologists (with 5 and 3 years of experience in head and neck radiology, respectively), blinded to the study design and pathological results. The average values of these two radiologists were employed for further statistical analyses.

Statistical analyses were performed using SPSS (version 23.0, SPSS, Chicago, IL, USA) and MedCalc (version 15.0, MedCalc, Mariakerke, Belgium) software. Sample size calculation used the method proposed by Li and Fine (21, 22). The intraclass correlation coefficient (ICC) was used to assess the inter-reader reproducibility. The agreement was interpreted as excellent (ICC >0.90), good (ICC = 0.75–0.90), moderate (ICC = 0.50–0.75), or poor (ICC <0.50) (24). Pearson correlation coefficient and Bland–Altman analysis were performed to evaluate the correlations between ECV_D and ECV_S (25). Mann–Whitney U test was conducted to compare ECV_D and ECV_S non-metastatic and metastatic cervical LNs. Receiver operating characteristic (ROC) curves analysis and the Delong method were performed to assess and compare the diagnostic efficiency of conventional qualitative LNs features and two ECV measurements (26). Optimal cutoff values were determined by maximizing the Youden index (sensitivity + specificity − 1). A two-sided p-value less than 0.05 was considered statistically significant.

One hundred and fifty-seven cervical LNs (81 non-metastatic and 76 metastatic) were analyzed. The distribution of all 81 non-metastatic LNs was as follows: 1) II level with 30 LNs; 2) III level with 17 LNs; 3) IV level with 9 LNs; and 4) VI level with 25 LNs. The distribution of all 76 metastatic LNs was as follows: 1) II level with 5 LNs; 2) III level with 7 LNs; 3) IV level with 20 LNs; and 4) V level with 1 LN; 5) VI level with 43 LNs. The mean size of all 81 non-metastatic and 76 metastatic LNs was 8.4 ± 1.3 mm and 14.2 ± 2.7 mm, respectively ( Table 1 ).

Fifty-nine cervical LNs (27 non-metastatic and 32 metastatic) were affected by cervical root artifact on arterial phase CT. The distribution of all 27 non-metastatic LNs was as follows: 1) IV level with 2 LNs; and 2) VI level with 25 LNs. The distribution of all 32 metastatic LNs was as follows: 1) IV level with 4 LNs; and 2) VI level with 28 LNs. The mean size of all 27 non-metastatic and 32 metastatic LNs was 7.1 ± 0.8 mm and 11.3 ± 1.2 mm, respectively ( Table 1 ).

Positive findings of size >10 mm, abnormal enhancement, calcification, cystic change, or extra nodal extension were 69.7% (53 of 76) in all metastatic cervical LNs, while 46.9% (15 of 32) were affected by cervical root artifacts on arterial phase CT. Detailed conventional qualitative features for diagnosing metastatic cervical LNs are summarized in Table 1 . The diagnostic performance of these conventional qualitative LN features is summarized in Table 2 .

The inter-reader agreement in the measurement of ECV_S and ECV_D was excellent with an ICC of 0.929 [95% confidence interval (CI): 0.893, 0.978] and 0.902 (95% CI: 0.878, 0.946), respectively. A strong positive correlation was found between ECV_D and ECV_S [r = 0.925 (95% CI: 0.899, 0.945); p <0.001). A scatter plot between ECV_D and ECV_S is shown in Figure 2 . The Bland–Altman plot for ECV_D and ECV_S showed a small bias (−0.6) with the lower and upper limits of agreement of −14.4 and 13.3 ( Figure 3 ).

In the non-metastatic LN group, ECV_S and ECV_D were 25.45% [interquartile range (IQR): 18.15%, 31.17%] and 22.53% (IQR: 16.97%, 29.09%), respectively. In the metastatic LN group, ECV_S and ECV_D were 39.18% (IQR: 30.09%, 50.44%) and 42.41% (IQR: 29.58%, 54.90%), respectively. Metastatic LNs showed significantly higher ECV_D (42.41% vs 22.53%, p <0.001) and ECV_S (39.18% vs 25.45%, p <0.001) than non-metastatic LNs. Box plots of two ECV fractions between the non-metastatic and metastatic LNs groups are shown in Figures 4A, C . Setting an ECV_D of 36.45% as the cut-off value, optimal diagnostic performance could be achieved with an AUC of 0.813 (95% CI: 0.743, 0.871), which was comparable with that of ECV_S [cut-off value = 34.99%; AUC = 0.793 (95% CI: 0.721, 0.853); p = 0.265]. The two ECV measurements both performed significantly better than conventional qualitative LN features (p all >0.05). The detailed diagnostic performance of ECV_S and ECV_D is summarized in Table 3 and shown in Figure 5A .

For cervical LNs affected by cervical root artifact on the arterial phase contrast-enhanced CT images, a significantly higher ECV_D (41.19% vs 20.27%, p <0.001) and ECV_S (35.96% vs 25.23%, p <0.001) of metastatic LNs than non-metastatic were also found ( Figures 4B, D ). ECV_D also showed good efficiency with an AUC of 0.756 (95% CI: 0.620, 0.862), which was also comparable with that of ECV_S [AUC = 0.716 (95% CI: 0.577, 0.830); p = 0.244]. Detailed diagnostic efficiency of ECV_S and ECV_D is summarized in Table 3 and shown in Figure 5B . A representative case of using ECV_S and ECV_D for diagnosing cervical LNs affected by cervical root artifact is shown in Figure 6 .