Background
Thyroid nodules are commonly detected in the adult population, and can be diagnosed as noncancerous or cancerous. Diffusion-weighted imaging (DWI) is an emerging technique for evaluating head and neck tumors, representing a non-invasive method for measuring the diffusion of molecular water, which has the potential to distinguish tissue properties and physiological features. Recently, DWI with derived apparent diffusion coefficient (ADC) has been applied to differentiate noncancerous thyroid nodules from cancerous ones quantitatively [
1‐
20]. Because of the heterogeneity of tumor tissues, ADC measurements may depend on region of interest (ROI) selection. In DWI, ROIs obtained by three main ROI techniques, i.e. whole-volume, single-slice, and small solid-sample, have been applied for obtaining the ADC values of tumors [
21,
22]. However, it is rarely assessed for thyroid nodules. The current study aimed to evaluate the effect of ROI selection on ADC measurements and interobserver variability in thyroid nodules.
Methods
Patients
Forty-five patients with thyroid nodules were recruited between September 2013 and December 2014, with signed informed consent obtained from all participants. The inclusion criteria comprised: diameter of thyroid nodules larger than 6 mm; no motion artifacts; no contraindications for MRI; no history of thyroidectomy or radiotherapy. Eight cases were excluded for thyroid nodules smaller than 6 mm, which makes it hard to identify the border using ADC measurements. Four additional patients with obvious movement artefacts on DWI images were also excluded. Therefore, 33 patients (7 males and 26 females; mean age, 52.2 ± 10.3 years; age range, 25–71 years) with 45 (30 benign and 15 malignant) thyroid nodules confirmed by histopathological findings were finally enrolled in the current study. The histopathological types of the thyroid nodules are listed in Table
1.
Table 1
Histopathological types of thyroid nodules in the evaluated patients
Papillary carcinoma | 14 | 12 | 37~64 | 51.5 |
Follicular carcinoma | 1 | 1 | 28~28 | 28.0 |
Adenoma | 20 | 11 | 25~63 | 51.6 |
Nodular goiter | 15 | 9 | 43~71 | 56.4 |
MRI
Imaging was carried out on a 1.5 T MR (Signa Excite HD Twinspeed, GE Healthcare, USA) using a four-channel array coil, by conventional MRI sequences and transverse single shot echo-planar DWI (b values, 0 and 400 s/mm
2). The main parameters of MRI sequences are presented in Table
2.
Table 2
Main scan parameters and sequences for thyroid gland evaluation
Axial T1WI | 520 | 14 | 200 × 200 | 320 × 192 | 4 | 0.5 | 3 | 12~14 |
Axial T2WI | 3500 | 102 | 200 × 200 | 320 × 224 | 4 | 0.5 | 4 | 12~14 |
Coronal T1WI | 550 | 11.1 | 240 × 240 | 320 × 224 | 3 | 0.5 | 4 | 12~14 |
Coronal T2WI | 3000 | 102 | 240 × 240 | 320 × 224 | 3 | 0.5 | 4 | 12~14 |
Axial DWI (b = 0, 400 s/mm2) | 6000 | 60.1 | 200 × 200 | 128 × 128 | 4 | 0.5 | 4 | 12~14 |
Data analysis
Inclusion and exclusion of patients were carried out by the same senior radiologist (10-year experience in head and neck radiology). The reader was not aware of histopathological findings. DWI images were analysed based on the lesion’s signal intensity relative to adjacent noncancerous thyroid tissues on axial T1WI and T2WI.
DWI data and ADC maps were analysed with the AW 4.3 software (GE Healthcare). Two senior radiologists (10- and 8-year experience in head and neck radiology, respectively) independently measured the lesions’ ADCs according to three ROI methods: placement of three circular ROIs (small solid-sample method); placement of a freehand ROI outlining the tumor on a single slice (single-slice method); and placement of freehand ROIs outlining the tumor on each slice containing the tumor (whole-volume method).
In the small solid-sample technique, average ADC was obtained from 3 circular ROIs within the tumor regions, with the highest cellular activity on DWI of b400 (median ROI area, 22 mm2; range, 18–26 mm2). In the single-slice technique, the ROI was defined by tracing a line along the perceived tumor margins on DWI of b400 (median ROI area, 217 mm2; range, 33–970 mm2). In the whole-volume technique, ROIs were made along the perceived tumor borders on DWI of b400, covering the whole tumor region on every tumor-containing slice (median ROI area, 896 mm2; range, 51–5201 mm2), and ADC values in all sections were averaged for further analysis.
Statistical analysis
Statistical analyses were carried out with the Medcalc software (Version 13.0.0.0, MedCalc software). Interobserver variability of tumor ADCs in all ROI methods was assessed by determining interclass correlation coefficients (ICCs), with the values of 0–0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80 and 0.81–1.00 reflecting poor, fair, moderate, good and excellent correlations, respectively [
23] as previously proposed [
24]. Average ADCs in various ROI selection techniques were compared by the Friedman test, with post-hoc assessment by the Wilcoxon signed-rank test [
25]. Statistical significance thresholds for the Friedman and post-hoc tests were set at
P < 0.05 and
P < 0.017 (0.05/3), respectively [
21].
Discussion
Our results demonstrated that the reproducibility of ADC measurements of both noncancerous and cancerous thyroid nodules in the whole-volume and single-slice methods was acceptable, with average interobserver bias not exceeding 0.1 × 10
− 3 mm
2/s and LOAs below 0.3 × 10
− 3 mm
2/s [
21]. The reproducibility of average ADC measurements of both noncancerous and cancerous thyroid nodules in the solid-sample ROI selection method was inadequate because of the scattered mean ADCs in Bland-Altman test results, with the limits of agreement over ±0.30 × 10–3 mm2/s for the small solid-sample group. Mean ADC measurements have been employed to assess the ADC diagnostic threshold in distinguishing noncancerous thyroid nodules from cancerous ones in previous studies [
1‐
20]. However, the reliability of ADC data obtained with the above ROI selection techniques in thyroid nodules has been rarely evaluated. Previously, the effects of the three ROI selection techniques on tumor ADC and interobserver variability in patients with pancreatic [
21] and advanced colorectal [
22] cancers have been assessed. The latter reports demonstrated that ROI size and placement considerably impact the tumor’s mean ADC as well as interobserver variability [
21,
22], with the whole-volume technique showing the highest reproducibility [
21,
22]. A consensus report by Padhani et al. advocated that basic standards for tissue diffusion coefficient assessment and reporting are necessary, and strongly warned against the use of small delineations in ADC measurements [
26]. Therefore, a feasible method should be determined to standardize ADC measurements in thyroid nodules. Additionally, we found that the whole-volume method for ROI selection yielded the highest diagnostic value for differentiating noncancerous and cancerous thyroid nodules.
The reproducibility of average ADC measurements for both noncancerous and cancerous thyroid nodules in the solid-sample ROI selection method was inadequate because of the scattered mean ADCs in the Bland-Altman test. In the current study, the solid-sample technique was limited to the size and position in the tumor area, significantly varying from one patient to another, which could be due to pathological and structural differences. In clinic, the solid-sample technique is commonly employed to obtain the ADCs of thyroid nodules [
1‐
20]. Here, three circular ROIs were placed within solid tumor areas showing high signals on DWI. Two investigators independently measured tumor ADCs using the solid-sample method, without consensual selection of the same slice and position for the ROI in each case. Tumors usually show heterogeneity, especially in cases with thyroid goitre and adenoma, and ROIs should be placed away from cystic areas. In this study, the solid-sample method yielded the worst interobserver variability for mean ADC measurements. Therefore, the solid-sample method derived ROI is not optimal for ADC measurements of thyroid nodules.
Regarding benign thyroid nodules, obvious differences were observed in mean ADCs among the assessed ROI selection techniques. In ADC measurement of malignant thyroid nodules by the small solid-sample technique derived ROI, the most viable solid tumor portions were ncluded in the ROI, while blood vessels as well as the tumor border and cystic parts were excluded. Blood vessels or the tumor boundary could elevate the ADCs due to the tissue having high blood perfusion along the vessel or boundary, which accelerates the diffusional movement of water molecules. In the whole-volume and single-slice methods, blood vessels and the tumor border were easily included in the ROI, which may lead to relatively elevated ADCs. Furthermore, previous studies reported markedly decreased ADCs in cancerous nodules but not in noncancerous and normal tissues [
8‐
12], corroborating the current work. The solid portions of malignant thyroid nodules contain areas with hypercellularity and high nucleocytoplasmic ratio, resulting in reduced extracellular space and limited cellular diffusion, which may lead to low ADC values [
15]. Additionally, field strength, b-value selection and the post-processing approach employed potentially contribute to the final ADC [
27]. As stated above, DWI was carried out on the same 1.5 T scanner in one hospital with the single-shot echo-planar imaging sequence to avoid instrument bias [
1‐
12]. Some studies have used b values of 300, 400 and 500 s/mm
2, respectively, for DWI of thyroid nodules, which can reduce the effects of blood perfusion to reflect the actual diffusion in the tissue [
15‐
17]. Other studies have used higher b values (over 600 s/mm
2) for DWI of thyroid nodules, which may increase susceptibility artefacts in DWI [
2,
12]. The b value of 400 s/mm
2 used in this study was suitable for the reproducibility of ADC measurements, and selected according to the signal-to-noise balance and DWI image quality [
28]. In addition, a meta-analysis of DWI value in distinguishing cancerous thyroid nodules from noncancerous ones advocated for higher b values to increase diagnostic accuracy, although no notable differences in AUC values were found between the low and high b value groups [
29].
The limitations of this work should be mentioned. Firstly, the b value for DWI in this study was low. DWI examinations were carried out with a b value of 400 s/mm
2 to minimize susceptibility artefacts and ameliorate the SNR of the thyroid gland, while greater b values might show higher sensitivity and reflect the actual diffusion [
29‐
31]. Additional b values for DWI should be assessed in further research, identifying the optimal b value for the detection of thyroid gland lesions. Secondly, the effects of ROI selection techniques on ADC measurement in distinct field strengths or b-values were not evaluated, which deserves further investigation.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.