Study subjects
The patients who underwent multimodal US examination of thyroid nodules in Beijing Tiantan Hospital, Capital Medical University, from January 2020 to September 2021 were analyzed retrospectively. The inclusion criteria are as follows: (1) Adults over 18 years old; (2) The patient was examined preoperatively multimodal US examination (if the patient has multiple nodules on one side, only one of the nodules with the highest TI-RADS level is included); (3) The patient underwent thyroidectomy and central compartment lymph node dissection (Area VI) in our hospital. After operation, the pathological diagnosis was well-differentiated PTC (pathological type was classic or follicular subtype PTCs) and whether the central lymph nodes were metastatic was determined. The exclusion criteria were as follows: (1) Lack of pivotal information; (2) Patients with other head or neck malignancies; (3) Patients had thyroidectomy before. The scope of thyroid surgery is mainly determined according to the 2015 management guideline of American Thyroid Association (ATA) [5]. This study was approved by the ethics committee of our hospital. Among the 852 thyroid nodules examined by multimodal US before surgery, 173 nodules underwent thyroidectomy in our hospital, 163 nodules were pathologically confirmed as well-differentiated thyroid papillary carcinoma, 10 were excluded due to pathologically confirmed as other types of thyroid cancer or benign nodules, and 15 were excluded due to higher grade nodules on the same side or lack of pivotal data. Therefore, 148 PTCs from 142 patients were brought into this retrospective study (Fig. 1). All cases had undergone pCND, and CLNM was confirmed by pathology.
Clinical Characteristics and Ultrasonographical Features
Baseline clinical characteristics, including age, gender, body weight, body mass index (BMI), thyrotropin (TSH), thyroglobulin (Tg), thyroglobulin antibody (TgAb), and thyroid peroxidase antibody (TPOAb) were collected from medical records within one month before surgery. The tumor number, location, ETE and Hashimoto's thyroiditis (HT) status were confirmed by pathological results.
Aplio i900 (Toshiba, Tokyo, Japan) equipped with a linear array probe i18LX5 probe for all US inspections. All patients were in a supine position with head flexion backward. US examination is performed by examiners with more than ten years of working experience. Conventional US is considered as grayscale US and Color Doppler Flow Imaging (CDFI). Three diameter lines (longitudinal, transverse, and thick) of thyroid nodules were measured on grayscale US, and the following characteristics were evaluated: shape, margin, echogenicity, type of calcification (absent, macrocalcification or microcalcification, microcalcification is defined as strong echo spots less than 1 mm in the nodule), microcalcification quantities (less than 5, greater than or equal to 5), peripheral halo, tall than wide (anteroposterior/transverse ratio > 1 or ≤ 1), contact capsule (whether the nodules contact the hyperechoic line of the capsule, shown in (Fig. 2) and diffused background.
CDFI grading was divided into three levels based on the Alder standard: (1) none, no blood flow inside or around the nodule; (2) moderate, there are less than five punctate blood flow or two strip blood flow in the nodule; (3) marked, there are five or more punctate blood flow or more than two main vessels in the nodule[19]. The characteristics of abnormal cervical lymph node on US (ACLN) include microcalcifications, cystic aspect, hyperechogenicity, round shape or peripheral vascularity, according to the 2015 ATA guidelines[5]. The color SMI (cSMI) and monochrome SMI (mSMI) were employed to assess the vascular distribution of the nodules on the section with the most abundant blood flow. The standard to estimate blood flow of SMI was similar to that of CDFI (Fig. 3). The microvascular density was obtained by calculating the percentage of color pixels / total pixels in the region of interest (SMI-VI).
All patients had signed informed consent before CEUS examination. CEUS was conducted by the same equipment as the conventional US. The mechanical index was set to 0.08–0.10. SonoVue (Bracco spa, Milan, Italy) was mixed well with 5 ml of saline as the contrast agent of CEUS. 1.6 to 1.8 ml of the suspension was injected into the patient’s vena-intermedia-cubiti rapidly, followed immediately by a 5.0-mL 0.9% saline. CEUS imaging lasted two minutes and was stored into the device digitally for analysis. Then, TIC (time-intensity curve) was obtained of nodules and surrounding parenchyma. Quantitative parameters were automatically calculated, including the peak intensity (PI), time to peak (TTP), mean transit time (MTT), and area under the curve (AUC). Meantime, the examiners evaluated the qualitative parameters, including the enhancement homogeneity (homogeneous or heterogeneous), enhancement direction (centripetal or non-centripetal), enhancement intensity, peripheral enhancement, and enhancement type (wash-in and wash-out order).
Shear wave elastic imaging was used to evaluate tumor stiffness. After the waveform was stable, three measurements of shear wave velocity were performed by placing a region of interest (ROI) inside the nodule and in the surrounding parenchyma. Soon afterward, the average shear wave velocity (SWV mean) of the nodules and the shear wave velocity ratio (SWV ratio, SWV of the nodules/ surrounding parenchyma) were calculated.
Statistical Analysis
Normally distributed data were expressed by mean ± standard deviation and analyzed with t-test, while nonnormally distributed data were shown as median with interquartile range (IQR) and compared by the Mann-Whitney U test. The χ 2 test or Fisher’s exact probability method was used to compare categorical variables, while Pearson’s chi square test was applied for stratified or matched categorical data. Chi-squared automatic interaction detector (CHAID) growth method was used to build the prediction model based on conventional US and multimodal US individually. The maximum depth was set to 3 and the minimum number of cases of the parent node and child node were 10 and 5 respectively. A ten-fold cross test was used to verify the misjudgment rate of the model. The ROC (receiver operating characteristic) curves and PR (precision-recall) curves were used to manifest the diagnostic capability between models. By calculating the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy (ACC), and the area under the curve (AUC) with 95% confidence intervals of different models, the diagnostic efficiency of the models was compared. All statistical tests were performed by SPSS software (version 26.0, IBM Corporation, USA). For all tests, p < 0.05 was considered to be statistically different.