Comparison of computed tomography features between follicular neoplasm and nodular hyperplasia

This retrospective study was approved by the Institutional Review Board, and informed consent of the patients was waived. From March 2010 to June 2014, 83 patients (male:female = 11:72, mean age: 49.7 years, age range: 28–78 years) underwent thyroid surgery in our institution for the treatment of FN. Of the 83 patients with FNs, 26 were excluded because of the absence of a preoperative neck CT scan (n = 11), mismatch between CT and histopathological findings (n = 2), small nodule size < 1 cm at the largest diameter (n = 8), and poor image quality of CT (n = 5). Finally, 59 FNs in 57 patients (male:female = 12:45, mean age: 50.4 years, range: 28–78) were included in this study.

From August 2013 to June 2014, 289 patients (male:female = 32:257, mean age: 52.2 years, age range: 22–79 years) underwent thyroid surgery for the treatment of thyroid malignancy or other thyroid lesions, and 159 had one or more NHs on histopathological analysis. Of the 159 patients, 94 were excluded because of the absence of a preoperative neck CT scan (n = 52), mismatch between CT and histopathological findings (n = 5), small nodule size < 1 cm at the largest diameter (n = 25), and poor image quality of CT (n = 12). Finally, 65 NHs in 65 patients (male:female = 5:60, mean age: 55.7 years, age range: 22–61 years) were included.

Neck CT examinations were performed with multi-detector CT scanners (Discovery CT750 HD, GE Healthcare, Milwaukee, USA; Aquilion One, Toshiba Medical System Corporation, Tokyo, Japan). The scout view was scanned through the lateral projection. Both non-enhanced and contrast-enhanced images of the neck, skull base, and superior mediastinum were obtained. The acquisition parameters for Aquilion One were 120 kV; 180 mA; rotation time, 0.5 s; pitch, 0.625; field of view, 35 cm; matrix size, 512 × 512 pixel; detector collimation, 64 mm × 0.5 mm; slice thickness, 1 mm; slice increment, 1 mm. The acquisition parameters for Discovery CT750 HD were filed of view, 30 cm; slice thickness, 1.25; and slice increment, 1.25 mm. The other parameters were approximately the same as those of Aquilion One. A contrast-enhanced image was obtained after the intravenous injection of contrast media (300 mg of non-ionized iodine per milliliter, Scanlux; Sanochemia Pharmazeutika AG, Leitha, Austria) at 3 mL/s, delayed until 40 s after the initiation of contrast media.

A single radiologist with 3 years of experience in head and neck imaging performed an image analysis using a picture archiving and communication system under blinded conditions for the histopathological results. The following CT features of thyroid nodules were retrospectively investigated: the degree and pattern of attenuation, configuration, margin, shape, pattern of calcification, degree and pattern of enhancement, and presence of a CT halo sign (defined as a low-attenuated rim ≥ 1 mm in thickness and surrounding ≥ 75 % of the peripheral margin of the thyroid nodule on either non-enhanced or contrast-enhanced CT image). To determine the degrees of attenuation and enhancement of thyroid nodules on non-enhanced and contrast-enhanced images, respectively, the adjacent thyroid parenchyma was used as a reference. The patterns of attenuation and enhancement of thyroid nodules on CT were classified as homogeneous or inhomogeneous. The configuration of the thyroid nodules was divided into intraglandular (the presence of thyroid parenchyma between the nodule and adjacent thyroid capsule), expansile (with a blunt angle between the lesion and adjacent thyroid capsule), or exophytic (with an acute angle between the lesion and adjacent thyroid capsule). The margin of the thyroid nodules was classified as smooth, irregular, lobulated, or poorly defined. The shape was classified based on the ratio of anteroposterior to transverse diameters in axial images as follows: (1) anteroposterior and transverse diameters of the nodule should be parallel with the anteroposterior and transverse axes of the thyroid gland, respectively. (2) the shape of the thyroid nodules was divided into ovoid (anteroposterior diameter/transverse diameter ratio ≤ 0.8), round (0.8 < anteroposterior diameter/transverse diameter ratio ≤ 1), or taller-than-wide (anteroposterior diameter/transverse diameter ratio > 1). Based on the degree of calcification, the thyroid nodule was classified as non-calcified, rim, nodular, punctate (tiny dot-like calcification), complete, or mixed (nodular plus punctate). The degree of nodular enhancement was classified as no/scant, decreased, iso-, and increased, based on a comparison with the adjacent thyroid parenchyma. CT halo sign was classified as absent or present.

To avoid investigation bias, nodule selection and image analysis were independently performed in two stages: (1) First, the selection, localization, and measurement of a thyroid nodule in all patients were performed on the basis of the CT and histopathological findings by a single radiologist. (2) After 4 weeks, image analysis of the thyroid nodules was performed by the same radiologist blinded to the histopathological findings.

The data were tested for normal distribution by using a Kolmogorov–Smirnov test. Age at the time of diagnosis and nodular size were expressed as the mean ± standard deviation. Mean differences in age and nodule size between FN and NH were compared using the independent t-test. Group comparisons of categorical variables were performed using the χ2 test. A univariate logistic regression analysis was first used to evaluate the predictive power of each variable. CT features with a high predictive power (p < 0.20, Wald test) were selected and analyzed by multivariate logistic regression analysis to determine an optimal logistic regression model for distinguishing NH from FN. A receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of CT features, and the area under the ROC curve (AUC) was compared between the two groups. All statistical analyses were performed with a statistical software (SPSS, version 17.0, SPSS, Chicago, IL, USA), and a p value less than 0.05 was considered statistically significant.