Grey-Scale Analysis Improves the Ultrasonographic Evaluation of Thyroid Nodules

Abstract

Ultrasonography is the main imaging method for the workup of thyroid nodules. However, interobserver agreement reported for echogenicity and echotexture is quite low. The aim of this study was to perform quantitative measurements of the degree of echogenicity and heterogeneity of thyroid nodules, to develop an objective and reproducible method to stratify these features to predict malignancy.A retrospective study of patients undergoing ultrasonography-guided fine-needle aspiration was performed in an University hospital thyroid center. From January 2010 to October 2012, 839 consecutive patients (908 nodules) underwent US-guided fine-needle aspiration. In a single ultrasound image, 3 regions of interest (ROIs) were drawn: the first including the nodule; the second including a portion of the adjacent thyroid parenchyma; the third, the strap muscle. Histogram analysis was performed, expressing the median, mean, and SD of the gray levels of the pixels comprising each region. Echogenicity was expressed as a ratio: the nodule/parenchyma, the nodule/muscle, and parenchyma/muscle median gray ratios were calculated. The heterogeneity index (HI) was calculated as the coefficient of variation of gray histogram for each of the 3 ROIs. Cytology and histology reports were recorded.Nodule/parenchyma median gray ratio was significantly lower (more hypoechoic) in nodules found to be malignant (0.45 vs 0.61; P = 0.002) and can be used as a continuous measure of hypoechogenicity (odds ratio [OR] 0.12; 95% confidence interval [CI] 0.03–0.49). Using a cutoff derived from ROC curve analysis (<0.46), it showed a substantial inter-rater agreement (k = 0.74), sensitivity of 56.7% (95% CI 37.4–74.5%), specificity of 72.0% (67.8–75.9%), positive likelihood ratio (LR) of 2.023 (1.434–2.852), and negative LR of 0.602 (0.398–0.910) in predicting malignancy (diagnostic odds ratio 3.36; 1.59–7.10). Parenchymal HI was associated with anti-thyroperoxidase positivity (OR 19.69; 3.69–105.23). The nodule HI was significantly higher in malignant nodules (0.73 vs 0.63; P = 0.03) and, if above the 0.60 cutoff, showed sensitivity of 76.7% (57.7–90.1%), specificity of 46.8% (42.3–51.4%), positive LR of 1.442 (1.164–1.786), and negative LR of 0.498 (0.259–0.960).Evaluation of nodule echogenicity and echotexture according to a numerical estimate (nodule/parenchyma median gray ratio and nodule HI) allows for an objective stratification of nodule echogenicity and internal structure.

Radiomics signature for prediction of lateral lymph node metastasis in conventional papillary thyroid carcinoma

“…Although several previous studies have applied histogram and texture analysis in thyroid nodules, most have investigated only a small number of imaging parameters and were not based on high-dimensional data [23][24][25][26][27][28]. Furthermore, the majority of studies applying histogram or texture analysis have focused on differentiating malignant and benign thyroid nodules [23][24][25][26][27]29]. Therefore, in this study, we attempted to develop a radiomics signature for the prediction of lateral LNM in patients with cPTC based solely on the radiomic features of the primary thyroid tumor.…”

Deep convolutional neural network for the diagnosis of thyroid nodules on ultrasound

“…Although these technical operator‐dependent components and variations in acquired image quality cannot be addressed by quantitative analysis, quantitative analysis can overcome variation and lack of reproducibility in image interpretation, and a multiple quantitative analysis has been developed to validate the effectiveness for differentiating thyroid malignancy. Quantitative measurement of thyroid nodules through histogram analysis was suggested as an objective method to stratify US features, and numerical estimates could significantly differentiate benign nodules from malignant nodules . However, when the diagnostic performances of quantitative analysis and visual assessment by radiologists were compared in predicting thyroid malignancy, quantitative analysis showed relatively inferior diagnostic performance compared to traditional grayscale US .…”

“…Quantitative measurement of thyroid nodules through histogram analysis was suggested as an objective method to stratify US features, and numerical estimates could significantly differentiate benign nodules from malignant nodules. 12 However, when the diagnostic performances of quantitative analysis and visual assessment by radiologists were compared in predicting thyroid malignancy, quantitative analysis showed relatively inferior diagnostic performance compared to traditional grayscale US. 13,14 A number of machine learning algorithms using US images have been investigated for the differential diagnosis of thyroid nodules and have shown conflicting results when diagnostic performances were compared between these algorithms and visual assessment by radiologists.…”

Nodular Thyroid Disease in the Era of Precision Medicine

“…The combination of US patterns leads to a higher specificity, but it associates to a lower sensitivity (15). Finally, it is worth to mention that the evaluation of these US features is characterized by a high interobserver variability (16,17). In the last decade, to improve standardization of thyroid ultrasound reporting, the guidelines of the American Association of Clinical Endocrinologists/American College of Endocrinology/Associazione Medici Endocrinologi (AACE/ACE/AME) (3), the 2015 guidelines of the American Thyroid Association (ATA) (18), the guidelines of the European Thyroid Association (ETA; EU-TIRADS, European Thyroid Imaging Reporting and Data System) (8), the American College of Radiology (ACR) TIRADS (9), and the Korean Society of Thyroid Radiology's K-TIRADS system, have proposed risk stratification systems with the goal of detecting nodule at greatest risk for malignancy and then to recommend graduated size cut-offs for FNA cytology (19) ( Table 1).…”