Multimode ultrasonic technique is recommended for the differential diagnosis of thyroid cancer

Introduction

Thyroid nodules are a common finding in symptomatic and asymptomatic patients, and they have a malignancy risk rate of approximately 5–15% (Brito et al., 2014). The detection rate of thyroid nodules is dramatically rising due to increased sensitivity in diagnostic imaging tools (Brito, Morris & Montori, 2013; Hoang, Nguyen & Davies, 2015). The current thyroid management guidelines recommend fine-needle aspiration (FNA) for almost all thyroid nodules with suspected malignancies to acquire histopathological confirmation (Haugen et al., 2016; Gharib et al., 2016). However, FNA is invasive and yields nondiagnostic results in approximately 25% of samples (Durante et al., 2018). The high detection rate for thyroid nodules and the relatively low accuracy of diagnosis for thyroid cancer are leading to unnecessary FNA procedures and unwarranted surgeries, which results in economic burden and worsens quality of life. Thus, it is important to improve the accuracy of diagnosis for differentiating thyroid cancer.

B-mode ultrasound is one of the most commonly used imaging techniques for screening thyroid nodules. A diagnosis made by B-mode ultrasound depends on the presence of grayscale features of malignancy and benignity, mainly including size, shape, margin, composition, echogenicity and echogenic foci (Xie et al., 2016). By counting these grayscale features, some risk stratification guidelines have been generated. Of them, the thyroid imaging reporting and data system (TI-RADS) adopted by the American College of Radiology (ACR) is most widely accepted in routine clinical practice. The ACR TI-RADS was constructed from multicenter big data by evaluating the association between five B-mode features and malignancy. The scoring system classifies thyroid nodules into five classes, from TR1 to TR5. Patients with TR3, TR4 and TR5 thyroid nodules are recommended to undergo FNA (Tessler et al., 2017). Although the ACR TI-RADS is robust in terms of statistics, the judgment of the grayscale features of nodules in B-mode ultrasound remains subjective. The diagnostic accuracy heavily depends on the experience of ultrasonographers and the performance of imaging devices (Hoang et al., 2007).

In addition to traditional B-mode ultrasound, ultrasound elasticity and contrast-enhanced techniques have been reported to be valuable in differentiating thyroid cancers (Sui et al., 2016; McQueen & Bhatia, 2015). Shear wave elastography (SWE) is a noninvasive tool that quantitatively reflects the stiffness property of nodules by measuring five elasticity values based on Young’s modulus (Dietrich et al., 2017). In recent years, an increasing number of studies on SWE have shown its good performance in distinguishing malignant thyroid nodules from benign nodules. However, the stability of elastic measurements is easily affected by the swallowing and breathing of subjects, the thyroid location and size, and sonographers’ experience (Veyrieres et al., 2012; Hu, Liu & Qian, 2017; Swan et al., 2019). Contrast-enhanced ultrasound (CEUS) reflects the perfusion of contrast agents within a nodule. After a patient receives an intravenous injection of ultrasonic contrast agent, which consists of gas-filled microbubbles, the contrast agent travels with the blood flow and reaches a targeted thyroid nodule in a few seconds (Lindner, 2004). By switching to the contrast mode of ultrasonic devices, the contrast-enhanced perfusion of agents can be detected differently between benign and malignant thyroid nodules. Previous studies have reported that benign thyroid nodules tend to show homogeneous iso/hyperechogenic contrast enhancement, while malignant nodules tend to show heterogeneous hypoechogenic contrast enhancement (Jiang et al., 2015; Zhang et al., 2010; Ma et al., 2016). However, contrast agents remain within nodules for only a few seconds; thus, the observed results are operator-dependent and easily affected by patients’ swallowing.

In fact, no single ultrasonic technique is perfect. These techniques tend to be complementary to each other. Thus, it may be a promising approach to perform a multimode ultrasonic technique that includes the B mode, elastic mode, and contrast-enhanced mode to improve the accuracy of diagnosis for differentiating thyroid cancer and to decrease unnecessary FNA or unwarranted surgery for benign nodules.

Materials and Methods

SWE-mode examination

The elastic mode was turned on after a clear grayscale image of thyroid nodules was captured under the B mode. The elastic mode shows both the grayscale image and the SWE image simultaneously (Fig. 1). First, we adjusted the Q-box of interest of the region so that it could cover the whole nodule and the surrounding tissue. Second, we froze the elastic image when the color of the surrounding tissue in the box was almost blue (blue indicates normal soft thyroid tissue). The right-sided color bar in Fig. 1 indicates the stiffness of a nodule, which was set from 0 to 100 kPa. Finally, we measured the stiffness of the nodule tissue as well as its comparative surrounding normal tissue by moving a small circular region where five quantitative elastic Young’s modulus values were displayed on the screen and recorded accordingly (Fig. 1).

Statistical analysis

In this study, two types of datasets were analyzed for differentiating malignant from benign thyroid nodules. The qualitative datasets included features of nodules based on the grayscale B mode and types of contrast-enhanced perfusion of nodules and were analyzed with the Chi-square test or Fisher’s exact test (Table 2). The quantitative dataset consisted of elastic indices measured via Young’s modulus property of nodule stiffness based on SWE technology. These data included the E_max, E_min, E_mean, E_SD and E_ratio and were analyzed by Student’s t-test (Table S1). A P value less than 0.05 was considered statistically significant. In addition, the cutoff value of different elastic parameters, the corresponding sensitivity and specificity, and the area under the receiver operating characteristic (ROC) curve were calculated (Table S2).

Table 2:

Comparison of qualitative characteristics obtained from the conventional 2D US and CEUS between benign versus malignant thyroid nodules.

Characteristics and classification	Benignity	Malignancy	P value
Composition
Solid or Almost solid	106	80	0.049b
Mixed cystic and solid	9	1
Echogenicity
Hypo-echoic or very hypo-echoic	90	79	2.68 × 10−4b
Iso/Mix-echoic	25	2
Margin
Irregular or Extra-thyroidal extension	19	43	1.41 × 10−7a
Regular	96	38
Shape (A/T)
>1	14	42	3.76 × 10−9a
<1	101	39
Calcification
Micro-calcification	4	28	2.11 × 10−8a
Absence or Macro-/Peripheral calcification	111	53
TI-RADS
RT3	13	6	0.316b
RT4	93	64
(4,5, and 6 points)	(43, 32, 18)	(2, 8, 54)
RT5	9	11
CEUS
Heterogeneous hypo-enhance	6	71	1.52 × 10−30a
Homogeneous iso/hyper-enhance	109	10

DOI: 10.7717/peerj.9112/table-2

Notes:

a Chi-square test.

b Fisher exact test.

2D US, two dimensional ultrasound; CEUS, contrast enhance ultrasound; A/T, anteroposterior/transverse diameter; ACR TI-RADS, American college radiologist thyroid imaging report and data system.

In this study, multivariate regression analysis was used to predict the risk of malignant thyroid nodules by integrating grayscale features, perfusion types of nodules, and elasticity indices into a comprehensive logistic regression model. Each of five grayscale features, shape, margin, calcification, composition and echogenicity, were analyzed by simple logistic regression. The statistically significant features (P < 0.05) and the elasticity indices with the highest accuracy were selected to be entered into the comprehensive logistic regression models (Fig. 2; Tables S3–S7). Finally, the areas under the ROC curves were used for the comparison of diagnostic accuracy among the various comprehensive models (Fig. 3; Table 3). All the raw data for the statistical analyses are included in Table S8. All the statistical analyses were executed with R software version 3.6, and the relevant code is presented in Supplemental Note 1.

Table 3:

Comparison of ROC curve analysis of multivariate logistic regression models for the prediction of benign versus malignant thyroid nodules.

Combination model	Sensitivity (%)	Specificity (%)	AUC (95% CI)
2D_US	87.65	66.96	0.841 [0.786–0.896]
2D_US+CEUS	88.89	93.91	0.960 [0.935–0.986]
2D_US +E_mean	92.59	87.83	0.954 [0.925–0.982]
2D_US+CEUS+E_mean	96.30	90.04	0.985 [0.971–0.999]

DOI: 10.7717/peerj.9112/table-3

Note:

ROC, receiver operating characteristic; AUC, the area under ROC curve; 2D_US, two dimensional ultrasound; CEUS, contrast enhance ultrasound; SWE, shear wave elastography; 95% CI, 95% confidence interval.

Results

Discussion

Recently, due to progress in the resolution of ultrasound imaging devices, the identification rate of thyroid nodules has increased to more than 30%. However, only 5–15% of thyroid nodules are malignant (Brito et al., 2014; Frates et al., 2006). The overdiagnosis of “malignant” nodules in ultrasonic screening leads to unnecessary FNA or unwarranted surgery, which results in a poor quality of life for patients. The risk scoring system proposed by the ACR TI-RADS has been widely accepted in clinical routine, where FNA is suggested when TR3, TR4 and TR5 nodules are diagnosed (Tessler et al., 2017). However, the accuracy of the scoring system is easily affected by the experience of ultrasonographers/radiologists due to the overlap of grayscale features in B-mode ultrasound between malignant and benign nodules. This viewpoint is also suggested by our finding that no statistically significant difference in the risk stratification was detected among the suspicious thyroid nodules with grades TR3, TR4 and TR5 (Table 2). Therefore, improving the diagnostic accuracy of suspicious thyroid nodules using multiple noninvasive tools would be meaningful in the current context.

Contrast-enhanced ultrasound has been reported to be helpful in diagnosing thyroid nodules (Jiang et al., 2015; Zhang et al., 2010), although some researchers argue that CEUS is ineffective in differentiating benign from malignant thyroid nodules (Friedrich-Rust et al., 2010). Our previous study showed that CEUS could improve the diagnostic accuracy of differentiating thyroid cancers, suggesting that heterogeneous hypo-echo enhancement (limited agent perfusion within a nodule) may be a high-risk type of malignant thyroid nodule. Such a perfusion type was also adopted in the present study (Jiang et al., 2015). This evidence is opposite to the results of blood perfusion within a breast tumor in which the hyperecho enhancement (more agent perfusion within a nodule) suggests a malignancy (Zhao et al., 2017). This evidence shows that there is a richer blood supply in breast cancer but a much less rich blood supply in thyroid cancer. Additionally, our results showed that the diagnostic accuracy of the B mode combined with the CEUS mode (2D_US+CEUS) was superior to that of the B mode alone (Fig. 3; Table 3), which is consistent with previous reports, providing further evidence for the value of CEUS in the differential diagnosis of thyroid nodules. However, due to the limitation of the short duration of microbubble contrast agents within a thyroid nodule, the judgment of perfusion type is strongly operator-dependent. Additionally, involuntary swallowing in subjects may cause a transient escape of nodules from the scope, thus further increasing the instability of the assessment.

In recent years, the SWE technique, which directly measures the stiffness of nodule tissue, has been increasingly studied for the differential diagnosis of suspicious thyroid nodules. Some have reported that SWE combined with the B mode can help to quantify nodules’ elastic stiffness and increase the accuracy of differentiating thyroid cancer (Veyrieres et al., 2012; Hu, Liu & Qian, 2017; Swan et al., 2019). In our study, of five elasticity values measured by Young’s modulus of nodules, the E_mean showed the highest accuracy, with an AUC of 0.892 (Fig. 2; Table S2). This result suggests that more attention should be paid to the E_mean rather than other values in clinical practice. Some studies claim that the B mode combined with SWE indicates a higher accuracy than the B mode combined with the CEUS technique (Chen et al., 2016). In contrast, our study showed that the B mode combined with the CEUS technique (2D_US + CEUS) had a similar diagnostic capacity to the B mode combined with SWE (2D_US + E_mean) (Fig. 3; Table 3). The inconsistency may be related to the unstable measurements of nodule elasticity, which is easily affected by swallowing and breathing in subjects, thyroid location and size, and sonographers’ experience.

The diagnostic value of the B mode combined with SWE or CEUS in differentiating thyroid cancer has been extensively reported. However, the effects of combining all three modes are rarely studied. Our study indicated that the combination of the three techniques led to the highest accuracy, with an AUC of 0.985, a sensitivity of 98.77% and a specificity of 90.43% (Fig. 3; Table 3). The findings encourage patients with suspicious thyroid nodules diagnosed by the B-mode grayscale ultrasound technique to receive both SWE and CEUS for further confirmation of evidence, which may help to avoid unnecessary FNA and unwarranted surgery of benign nodules. However, considering the instability of the SWE and CEUS techniques in clinical practice, multicenter trials with larger sample sizes would be necessary in future studies.

Conclusions

No single ultrasonic technique is perfect for differentiating thyroid cancer, and all the techniques tend to be complementary to each other. Additionally, new techniques have been developed, such as the current artificial intelligence, which has been attending to intelligent detection and differential diagnosis of thyroid nodules throughout deep learning methods (Song et al., 2019; Buda et al., 2019) and aims to relieve the labor of sonographers and increase the accuracy of diagnosis. However, the clinical application is still at a very early stage. Since the features of thyroid nodules are very complex and the high quality of images is operator-dependent, machine learning methods are challenging (LeCun, Bengio & Hinton, 2015). Therefore, as knowledge of the benefits and limitations of emerging ultrasonic techniques is continuously expanding, the trend of integrating multiple techniques to diagnose suspicious thyroid nodules will become increasingly apparent. The findings in our study follow this trend and encourage further exploration of the value of multimode ultrasonic techniques in clinical practice.

Supplemental Information