Dual-Source Dual-Energy Thin-Section CT Combined with Small Field of View Technique for Diagnosing Small Lymph Node in Thyroid Cancer

Background To evaluate the diagnostic performance of quantitative special CT parameters derived from dual-source dual-energy CT at small eld of view (FOV) for small lymph node metastasis in thyroid cancer. Methods This was a retrospective study. From 2016 to 2019, 280 patients with thyroid disease underwent thin-section dual-source dual-energy thyroid CT and thyroid surgery. The data of patients with lymph nodes having a short diameter of 2-6 mm were analyzed. The targeted lymph nodes were sketched, their quantitative dual-energy CT parameters were measured, and all parameters between metastatic and nonmetastatic lymph nodes were compared. These parameters were then tted to univariable and multivariable binary logistic regression models. The diagnostic role of spectral parameters was analyzed by receiver operating characteristic curves and compared with the McNemar test. Small FOV CT images and a mathematical model were used to judge the lymph nodes status respectively and then compared with pathological results. Results Of the 216 lymph nodes investigated in this study, 52.3% and 23.6% had a short diameter of 2-3 mm and 4 mm, respectively. Multiple quantitative CT parameters were found to be signicantly different between benign and malignant lymph nodes and binary regression analysis was performed. The mathematical model was: p=e /(1+ ), electron cloud phase normalized iodine concentration+2.156×arterial phase normalized effective atomic number -0.540×venous phase slope of the spectral Hounseld unit curve +1.676×venous phase iodine concentration. This parameter model has an AUC of 92%, with good discrimination and consistency, and the diagnostic accuracy was 90.3%. The diagnostic accuracy of CT image model was 43.1%, and for lymph nodes with short-diameter 2-3 mm, the diagnostic accuracy was 22.1%. Conclusions Parameter model show higher diagnostic accuracy than CT image model for diagnosing small lymph node metastasis in thyroid cancer, and quantitative dual-energy CT parameters were very useful for small lymph nodes that were dicult to be diagnosed only on


Background
Thyroid cancer is one of the most common malignant tumors with increasing prevalence in recent years.
Lymph node (LN) metastasis is also very common and its extent determines surgical planning. Based on the American Thyroid Association guidelines, ultrasound is the recommended imaging modality to evaluate cervical lymph nodes status while computed tomography (CT) is suggested as an adjunct [1] . However, ultrasound has limitations for detecting LN metastasis in tricky areas due to gas, bony structure obstructions or deep locations. Further, since macrometastasis in lymph nodes originally develop from micrometastasis and all large metastatic lymph nodes from small metastatic lymph nodes, therefore, micrometastatic LNs or metastasis in small LNs can be considered as the early stage of lymph node metastasis [2] . To nd out lymph node metastasis at an early stage, it is necessary to investigate small lymph node metastasis. However, the detection and diagnosis of very small lymph nodes or micrometastatic lymph nodes is quite cumbersome using CT or ultrasound. Small lymph nodes were di cult to be diagnosed only on conventional CT images, because there were few metastatic features on CT images.
Recently, dual-source dual-energy thyroid CT has shown promising ability for detecting cervical lymph node status of thyroid cancer [3,4] . Its CT images at small eld of view (FOV) has a smaller range of scatter artifacts and can detect LN boundary more sharply, with improved spatial and contrast resolution, which can be visualized in geometrically enlarged images, highlighting the details of small LNs that are often di cult to be shown on CT images at normal FOV. Further, thin-section CT images at small FOV is additionally useful for accurately delineating and diagnosing the LNs. Some dual-energy parameters such as dual energy index are seldomly reported, and are they useful for diagnosing lymph node status? Will the above techniques help to diagnose small lymph nodes that were di cult to be diagnosed only on conventional CT images?
Till present, there has been no study investigating the diagnostic accuracy of dual-energy CT for small LN metastasis in patients with thyroid cancer. In this study, we aimed to investigate the diagnostic performance of thin-section dual-source dual-energy thyroid CT at small FOV for detecting small cervical LNs in thyroid cancer and investigate its radiologic-pathologic correlation.

Methods
This was a retrospective study based on prospective design. From December 2016 to October 2019, the data of 280 patients who underwent thin-section dual-source dual-energy thyroid CT and thyroid surgery with lymphadenectomy at Sun Yat-sen University Cancer Center were retrieved. A total of 216 lymph nodes in 74 patients were enrolled, and the study enrollment process is illustrated in Figure 1.
Cases were selected based on the following inclusion criteria: (1) the maximal short diameter of all selected cervical lymph nodes should be within the rang 2 -6 mm. (2) the pathological results of all lymph nodes at the same level or sub-level should be the same and their status were con rmed by pathologic examination. (3) the lymph nodes from level VI were further divided into multiple subgroups during surgery, including the anterior laryngeal region, anterior tracheal region, and paratracheal region (left and right). Lymph nodes from each subgroup of level VI were all metastatic or normal lymph nodes, so as to facilitate lymph node selection and radiologic-pathologic correlation. (4) preoperative ultrasound showed no suspicious cervical lymph nodes in level Vb. A small number of lymph nodes were located at speci c anatomical locations, which were easy to match and locate pathological results on CT images, and they were also included in this study.
All the LNs were labeled after resection according to their cervical compartments and those from cervical level VI were further divided into 4 subregions. Patients in our study were with thyroid cancer or nodular goiter, with or without Hashimoto's thyroiditis.
Quantitative parameters of dual-source dual-energy thin-section CT Dual-energy scan mode was used with Siemens dual-source CT (SOMATOM Force CT, SIEMENS, German), and the scan range was from the level of the skull base to the upper edge of the aortic arch. All patients underwent thin-section dual-energy thyroid CT in the precontrast phase, arterial phase and venous phase. Parameters of dual-energy scanning mode were as following: A and B two tubes were used, with tube voltage /tube current 90kV/250mA and Sn150kV/125mA, respectively. Turn on CARE Dose 4D, and the collimation was 2×192×0.6mm. CT images were obtained with a gantry rotation of 0.5 s, pitch factor 1. A, B, M groups of CT images were automatically obtained. Images of group A and B were obtained from low and high tube voltages, respectively. M group of images were mixed energy images for diagnosis, from linear fusion, with dual-energy fusion coe cient 0.6. The CT images with slice thickness 0.75 mm and layer spacing 0, were reconstructed at a 512 × 512 matrix size. Thin-section CT images at small FOV was performed in the precontrast phase, arterial and venous phase. The FOV sizes were 140mm × 140mm or 160mm × 160mm, according to neck size. The patients were all in the supine position during the examination. High pressure syringe was used. The contrast agent used was iopromide 1.0ml/Kg (370mgI/ml) and was injected through the elbow vein at a ow rate of 4.0ml/s. The arterial phase scan was automatically triggered. The trigger point was the aortic arch, and the trigger threshold was 100HU. The venous phase was delayed by 50s scanning. After scanning, conventional mixed energy images and two sets of single-energy images were obtained. All images were transferred to the PACS terminal and workstation. The basic signs of lesion morphology evaluation (conventional CT images) and multi-parameter analysis of energy spectrum were performed using the software Syngo.via. Multi-planar reconstructions were performed for axial, coronal, and sagittal planes, with slice thickness 0.75 mm. The dualsource dual-energy CT data sets were reconstructed into monochromatic image sets, iodine-based materialdecomposition images, and effective atomic number images.
To obtain the slope of the spectral Houns eld unit curve, the region of interest (ROI) was drawn to be as large as possible to include the entire lymph node. Internal fat, necrosis and calci cations were not excluded. Further, attention was taken to avoid the peripheral fat tissues. To gain Rho and Z, we drew the ROI according to the boundary of the selected LN, that is, the entire lymph node was included. Every characteristic part of a lymph node, such as obvious enhancement, cystic changes and calci cation, was also included. Small LNs were conformally sketched to accurately outline their contour and we could sketch lymph nodes with short diameter of 2-3 mm. All measurements were performed for twice on 2 consecutive maximal slices in the selected lymph nodes and their average values were calculated.
Dual-source dual-energy thyroid CT was used for 280 patients with gemstone spectral imaging mode and 74 patients were enrolled according to the inclusion criteria. GSI quantitative parameters, including iodine concentration (IC), normalized iodine concentrations (NIC), slope of the spectral Houns eld unit curve (λ HU ), normalized effective atomic number (Z), electron cloud density (Rho), and double energy index (DEI) in the precontrast, arterial and venous phase, were measured. Iodine concentration and effective atomic number of the LNs were respectively divided by the iodine concentration and effective atomic number of the aorta to obtain the normalized iodine concentration and normalized effective atomic number. The slope of the spectral Houns eld unit curve, which was de ned as the difference between the CT value at 40 keV and that at 70 keV divided by the energy difference (30 keV), was calculated as using the following formula: λ HU = (HU40keV-HU70keV)/30 keV, where HU40keV represents the CT value measured on 40-keV images and HU70keV stands for the CT value measured on 70-keV images. CT values of the target LN in the precontrast, arterial and venous phase were measured on 40-keV images and 70-keV images, respectively. Iodine concentration of the LN and the aorta were measured on iodine-based material-decomposition images.

Lymph node selection
In order to ensure that the lymph nodes selected on CT images corresponded to the pathological results, we have adopted the following methods: 1. Some benign lymph nodes originate from the cervical lymph nodes of benign thyroid disease, such as nodular goiter or chronic lymphocytic thyroiditis, and these patients have no malignant tumor; 2. Other benign lymph nodes, which originate from the benign cervical lymph nodes in patients with thyroid cancer, were con rmed by the pathological results. For example, the pathological results reported that all level IV lymph nodes at the right neck were benign, and some of these LNs were selected as target ones to measure quantitative parameters. The cervical lymph nodes in some patients with thyroid cancer, are all non-metastatic lymph nodes, con rmed by postoperative pathological results. We selected some of these lymph nodes for CT-pathological comparison study. 3.Metastatic lymph nodes were from cervical lymph nodes in patients with thyroid cancer, and all lymph nodes in one level or compartment were reported as metastasis postoperatively. For example, the pathological report showed that all the anterior tracheal lymph nodes are metastatic, and then we selected some of them for CT-pathological comparison. 4.Some small lymph nodes were located at speci c anatomical sites and their locations were labeled in pathological reports. These lymph nodes were also included in our study.
We further divided level VI into four sub-regions for more screening areas and lymph nodes to assess their radiologic-pathologic correlation; i.e. all lymph nodes in this sub-region were metastatic lymph nodes or nonmetastatic ones. If the entire area of LNs were benign or malignant according to postoperative pathological results, these LNs were selected as target ones for measuring spectral CT parameters. For example, if level IV lymph nodes at the right side were metastatic, or one lymph node was at special anatomical location, we measured the multiple quantitative parameters for these lymph nodes.
For lymph nodes selected for multi-parameter measurement, two radiologists (10 years' experience in thyroid imaging), used plain and dual phases enhanced images to diagnose whether these lymph nodes were metastatic or not, without prior knowledge of the patients' pathological results. Then, dual-energy CT parameters were used to diagnose these LNs again. At last, we compared the diagnostic performance of different models: the image model was based on thin-section CT images at small FOV, and the parameter model was based on multiple dual-energy CT parameters.
De nition of diagnostic di culty for LNs on thin-section CT images at 140 mm or 160 mm FOV in our study were as follows. Lymph nodes that were usually too small (with short diameter of 2-3 mm), without signi cant signs of metastasis, and they were usually located at level VI and on the same side as the thyroid cancer. In this situation, it was di cult for radiologist to judge if one lymph node was malignant or benign on CT images. Or other lymph nodes (with short diameter of 4-6 mm) lacked obvious metastatic signs on CT images, so the radiologist had di culty in diagnosing them. Suspicious radiologic features of malignant LNs on CT images included cystic changes, calci cations, rounded shape, short diameter more than 5 mm, obvious enhancement and nodular enhancement.

Statistical analysis
The SPSS software (version 22.0) was used to analyzed. The quantitative parameters between benign and malignant LNs were compared. The receiver operating characteristic (ROC) curves were drawn to establish the optimal threshold values and diagnostic ability for small lymph nodes. P< 0.05 was considered as having signi cant difference for the parameters. The quantitative parameters were tted to univariable and multivariable binary logistic regression models. The diagnostic role of quantitative parameters was analyzed using the ROC curves and compared using the McNemar test. The optimal threshold was determined by using the Youden index (sensitivity + speci city -1). The sensitivity, speci city, PPV, NPV and accuracy of the quantitative dual-energy CT parameters were also calculated. The diagnostic e ciency of the parameter model was calculated using the area under the curve (AUC) values from ROC curves.

Results
Lymph nodes from the benign group were distributed diffusely in levels II, III, IV and VI, whereas the lymph nodes in the malignant group were in levels IV and VI. A total of 216 lymph nodes, including 92 (42.6 %) metastatic LNs from 43 patients with PTC and 124 (57.4%) benign LNs from 31 patients with thyroid disease, were found on the preoperative CT images. Lymph nodes with short diameter of 2-4 mm accounted for 75.92% of all LNs and the average short diameter of these lymph nodes was 3.6 mm. 13 benign lymph nodes were in level II and III, and 9 LNs were in level IV, and 194 LNs were in level VI.
The diagnostic error rate and accuracy rate was 9.7% (21 LNs) and 90.3% for the parameter model based on dual-energy CT parameters. This model had false positive results in 11 LNs and false negative results in 10 LNs. The diagnostic performances of parameter model and image model are shown in Table 1.
Of the 113 small lymph nodes with short diameter of 2-3 mm, the radiologists believed that only 30 of them could be diagnosed using thin-section CT images at small FOV, of which 25 LNs were correctly diagnosed and 5 misdiagnosed; while the remaining 83 lymph nodes did not show visible signs of metastasis on CT images. Some of them were near the thyroid cancer in level VI and their metastatic status could not still be judged. Eight lymph nodes with short-diameter 4 mm and 5 lymph nodes with short-diameter 5 mm did not show any metastatic sign on CT images, making it di cult to diagnose the LN status. Among all these lymph nodes, the diagnostic accuracy of the image model was 43.1% and the diagnostic error rate was 12.5%. However, the diagnostic di culty rate was 44.4%. In small lymph nodes with short diameter of 2-3 mm, the probability of correct diagnosis was only 22.1% and the diagnosis di culty rate was 73.5%.

Univariate analysis
After calculating the diagnostic value of each cut-off point for multiple parameters, Table 2 shows the diagnostic value of the parameter model in diagnosing small lymph nodes. Quantitative dual-energy CT parameters in discriminating between the benign lymph nodes and malignant lymph nodes are shown in Table   3. Representative contrast enhanced CT images showed that IC, spectral curve, and effective atomic number were different in benign and metastatic lymph nodes (Figure 2). Independent sample t-test showed that the P value of multiple spectral parameters were <0.05 between benign and malignant LNs (Table 3), and they were included for the binary logistic regression analysis. Only four parameters showed no statistical difference. ROC analysis of quantitative dual-energy CT parameters for differential diagnosis of metastatic and nonmetastatic lymph nodes and their diagnostic performances are shown in Table 4. The arterial phase IC had the highest diagnostic accuracy (85.65%) on per-lymph node basis. The arterial phase NIC had a higher AUC value than other parameters.

Binary logistic regression model
Several independent variables demonstrated statistical signi cance between benign and malignant LNs in the univariate analysis. According to the binary logistic regression analysis, we built the following model with dualenergy CT parameters for predicting metastatic lymph nodes using the following formula: Probability of malignant LNs=e y /(1+ e y ) , where y= -23.119+0.033×precontrast phase Rho +0.076×arterial phase NIC +2.156×arterial phase Z -0.540×venous phase K +1.676×venous phase IC.
The above parameters were included in the binary logistic regression analysis to calculate the diagnostic value of each cut-off point for multiple parameters. We found that pre-contrast electron cloud density, normalized IC in the arterial phase, normalized effective atomic number in the arterial phase, λ HU in the venous phase, and IC in the venous phase, had the best diagnostic value. The AUC was 92% for the diagnostic model ( Figure 3). Figure 4 shows the assessment of this parameter model, which indicated that the predicted value was close to the observed value in both the benign LN group and malignant LN group for this diagnostic model. Logistic regression analysis of metastatic status results are shown in Table 5.

Discussion
There have been dual-energy CT studies investigating lymph node metastasis [3,4] , and several studies have found that CT has additional bene ts for lymph node metastasis in thyroid cancer [7] . However, there are still diagnosis dilemma in the diagnosis of small LNs.The smaller the lymph nodes are, the more di cult it may become to diagnose. Especially when one lymph node is ≤ 3 mm in short diameter, it become more di cult to determine the metastatic status of small LNs. Still, about half of the 216 small lymph nodes, are di cult to diagnose only on thin-section CT images at small FOV. While most of the 216 small LNs can be diagnosed correctly with our parameter model. We found that the parameter model had a high accuracy and diagnostic values in diagnosing small lymph nodes of thyroid cancer, which was better than the diagnostic results of image model based on thin-section CT images at small FOV, especially for lymph nodes of diameter ≤ 3 mm.
Our diagnostic accuracy of LN status is also better than that in previous studies [7,9,14−16] . Compared with previous dual-energy CT studies, the dual-energy CT scanner we used belong to third generation, which improves the energy resolution and the ability to distinguish substances; and we also investigated some new parameters, including electron cloud density, normalized effective atomic number and DEI in three phases, to develop a parameter model. We performed dual-source dual-energy CT at small FOV with multiple quantitative parameters to facilitate accurate detection and diagnosis of small lymph nodes.This study uses thin-section images at small FOV, and it is helpful for the identi cation of small lymph nodes and the delineation of conformal ROI, and its boundary can be identi ed more accurately, which is bene cial to multi-parameter measurement and diagnosis of small lymph nodes with dual-energy CT. Small FOV, thin-section and multiplanar reconstruction can help observe cervical LNs from different planes, detect more small LNs and discover more features in one LN, which help to detect suspicious metastatic LN [14] . After using small FOV, the improved resolution is helpful to display internal features in these small lymph nodes. CARE Dose 4D or iterative reconstruction algorithm can reduce the radiation dose so that the radiation dose of CT at small FOV is not higher than that of CT at conventional FOV; dual-source dual-energy CT with energy spectrum puri cation technique, reduces the radiation dose [5,6] , leading to a 30% radiation dose reduction.
Macrometastatic LNs must be treated, because it leads to worse prognosis. While benign cervical lymph nodes of thyroid cancer do not require surgical resection. Some scholars believed that after removal of micrometastatic lymph node [2] , the prognosis is similar to that of benign LNs [1] . For patients with thyroid cancer and micrometastatic lymph nodes, if these LNs are not resected, is the prognosis really the same as that of thyroid cancer patients without LN metastasis? is the overall survival between benign LN and micrometastatic LN of thyroid similar? No studies have ever provided the answer. And micrometastatic LNs may develop into macrometastatic lymph nodes or metastasis in other sites. So, the diagnosis of micrometastatic LNs and very small LN metastasis should be investigated., and its management should be different from that of benign or macrometastatic LNs. Should active surveillance be adopted for micrometastatic LNs? The methods we used in our study may be helpful and inspiring for the diagnosis of small LN or micrometastatic LN in other cancers.
Cervical levels VI and VII are believed to be the rst stage of lymph node metastasis in thyroid cancer [8,9] . Most of the selected cervical lymph nodes were located in level VI. Small lymph nodes account for the majority in level VI. For small FOV CT images, the signal-to-noise ratio and spatial resolution has improved [10][11][12][13] .
ROI was often drawn in the enhancement area in previous studies, and this area does not necessarily represent the entire lymph node. We usually encompass the entire lymph node as ROI, not just a small part of it. Our ROI delineation of lymph nodes was different from that of previous studies. When we measure and calculate slope of the spectral Houns eld unit curve, our ROI at least included most part of the LN; while measuring other quantitative parameters of dual-energy CT, we tried to make the ROI include the entire lymph node in the conformal ROI, rather than delineating typical part or the central part of one lymph node. And we believed that this method of ROI delineation is more comprehensive, accurate and representative. In many studies, the ROI of lymph node excluded cystic, necrotic and calci cation areas, while our delineation included all areas in target lymph node. We believe that in most cases, the calci cation and cystic changes in one lymph node represent the characteristics of metastatic foci, and these special areas should be included in the region of interest, so that the spectral parameters can fully re ect the true characteristics of lymph node metastasis in thyroid cancer. Further, our delineation area included the marginal part of the lymph node, and the marginal part of lymph node is often the rst or most common site for tumor deposits. Pathological examination shows that about 57% of tumor deposits in LN are located at the marginal sinus [17] , which is usually at the marginal part of one lymph node. Our ROI including the commonest site of tumor deposits in lymph node may better re ect whether the lymph node is metastatic.
The method we selected LN made the pathological results of the chosen LN reliable, and we could precisely compare the predicted results of multiple parameters model with pathological results, so we can investigate the diagnostic value of dual-energy CT at small FOV. Because the pathological results of all LNs at the same level or sublevel were the same, we can select any of these small lymph nodes for multiple quantitative parameters measurement, so as to ensure the radiologic-pathologic correlation reliable. The method reduces the tedious procedures of the one node-by-one node correlation between CT and pathology, and it lessens the probability of possible errors [4] , improving the quality of the study. All selected lymph nodes have de nite pathological results in our study. Thyroid surgeon in our team routinely divides bilateral cervical level VI further into four sublevels for patients with thyroid disease preoperatively: pre-larynx, pre-tracheal, left paratracheal and right paratracheal sublevels [18] .And then these cervical nodes were resected and labeled, we measured multiple quantitative parameters of dual-energy CT and performed diagnostic model. More regions with LNs could be screened and enrolled in this study. This is a retrospective study based on prospective design, and there may be selection bias. Prospective research is required, and we have registered a prospective study (Registration number: ChiCTR2000035195; http://www.chictr.org.cn). Random selection of targeted lymph nodes for one-by-one correspondence between imaging and pathology can help eliminate selection bias.

Conclusions
We believe that the by using dual-source dual-energy CT at small FOV, the parameter model can accurately predict the metastasis of small lymph node in thyroid cancer, which is superior to the image model and previous studies, and it has increased diagnostic accuracy. Quantitative dual-energy CT parameters were very useful for small lymph nodes that were di cult to be diagnosed only on conventional CT images. The parameter model has better discrimination and allows more small LNs to be identi ed. Accurate assessment of cervical LN status does good to cancer staging and treatment. This study has been approved by the institutional ethics committee of Sun Yat-sen University Cancer Center, and informed consent was waived.

Consent for publication
Not applicable

Availability of data and materials
The data that support the ndings of this study will be available from https://www.researchdata.org.cn/ but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of the corresponding author.

Competing interests
The authors declare that they have no competing interests.

No Funding
Authors' contributions: Li Sheng design this study, analyzed the data and wrote this paper. Zhuo Shuiqing and Sun Jiayuan contributed equally. Zhuo Shuiqing also performed CT scan and spectral parameter measurement. Sun Jiayuan also collected the data. Liulongzhong collected the pathologic data. Chang Jinyong and Li Sheng read CT images and revised the paper.
Corresponding author: Liulongzhong, Li Sheng, e-mail: lisheng@sysucc.org.cn Among the lymph nodes diagnosed with dual-energy CT parameters model, the diagnostic values of 4 false negative lymph nodes and 2 false positive lymph nodes were very close to the diagnostic threshold.