One of the consequences of increased use of imaging has been the discovery of incidentalomas, or pseudodiseases, that are common in the general population but have no or minor clinical significance. Once such incidentaloma, the thyroid nodule, is extremely common, found in some autopsy series in as much as 50% of the general population [1, 2]. Most of these nodules are benign; the incidence of malignancy is quite low, 3-7% [3-5]. In the late 1990s, articles began to appear questioning the reliability of radiotracer uptake as a predictor of benignity, occasioning a rapid transition from nuclear medicine to ultrasound for evaluation of the thyroid [6-8]. The superior resolution of ultrasound images has resulted in discovery of a large number of thyroid nodules that heretofore had been obscured [9].

Since the late 1990s, several studies have been conducted to analyze the relation between specific sonographic features of thyroid nodules and malignancy [2, 10-16]. Although guidelines have been established, such as those of the Society of Radiologists in Ultrasound, the American Thyroid Association, and the European Thyroid Association [2, 17-22], they are commonly confusing and at times ignored in everyday practice, largely because of lack of familiarity with and trust in their validity. Common in the studies is a persistent limitation of specificity and sensitivity of specific ultrasound features in the prediction of malignancy. Some authors [23, 24] advocate a changed approach of recognition of specific patterns rather than individual ultrasound features in separation of nodules that require biopsy from those that do not. The purpose of our study was to evaluate the accuracy of such a morphologic feature-oriented approach to the identification of benign thyroid nodules.

The individual ultrasound features of each nodule analyzed were size, number, texture (Fig. 1A), margination (Fig. 1B), presence of internal densities or calcifications (Figs. 1C, and 1D), edge refraction, and vascularity relative to the rest of the gland [13, 25, 26] (Fig. 1E). Analysis of the presence or absence of individual sonographic features revealed no feature with consistently high sensitivity or specificity for malignancy (Table 1). In our study, sensitivity for the presence or absence of specific features was 35-100% and specificity, 8.9-97.8%. There was no correlation between diagnosis and nodule size, which was categorized as less than 1 cm (n = 7), 1-2 cm (n = 288), and larger than 2 cm (n = 206) (Table 2). However, several features were found to have a statistically significant negative predictive value. These individual features, the absence of which was common in benign disease, included calcification, halo, hypoechogenicity, isoechogenicity, and ring or peripheral hypervascularity.

Each nodule was assigned to one of 10 discrete morphologic groupings. These patterns, which were based on a previous report [23] and expanded according to our experience, were as follows: 1, spongiform without hypervascularity (Fig. 2A); 2, cyst with avascular colloid plug (Fig. 2B); 3, giraffe pattern (Fig. 2C) with blocks of hyperechogenicity, or white, separated by bands of hypoechogenicity, or black; 4, uniform hyperechogenicity (“white knight”) (Fig. 2D); 5, intense hypervascularity (“red light”) (Fig. 2E); 6, hypoechogenicity (Fig. 2F); 7, isoechogenicity without halo (Fig. 2G); 8, isoechogenicity with halo (Fig. 2H); 9, “ring of fire,” or nodules with intense peripheral vascularity (Fig. 2I); and 10, other (Fig. 2J), or a mixed pattern or pattern that did not fit the other categories (Table 3). A distinct pattern emerged in which it became evident that there were specific morphologic groupings or patterns that were accurate predictors of benign disease. Specifically, there were no malignant nodules in the 303 patients (61%) with patterns 1-4 (Table 4). Spongiform nonhypervascular masses were the most common type of nodule seen, 210 of 210 being found benign at FNA biopsy. All 53 of the cysts with internal colloid clot, all 23 giraffe pattern nodules, and all 17 hyperechoic nodules were benign. The results in patterns 5-10 were unpredictable, ranging from 35 of 37 isoechoic nodules without halo biopsied being benign to only 31 of 45 hypoechoic nodules being benign.

A thyroid nodule is a discrete lesion, sonographically distinct from the surrounding thyroid parenchyma [27]. Rather than a single disease, nodules are manifestations of a gamut of thyroid diseases [28]. Although some thyroid nodules may be discovered at physical examination, many are incidental findings of other imaging studies, such as CT and MRI of the neck or chest and carotid ultrasound imaging. FNA of thyroid nodules has replaced blind surgical excision as the procedure of choice in the diagnosis of thyroid nodules. Use of FNA has led to a considerable decrease in the number of surgical excisions and to a twofold increase in the diagnosis of carcinoma [4, 5, 29]. The relative ease of FNA compared with surgery and the increased frequency and refinement of imaging studies has resulted in what some authors have referred to as an epidemic of thyroid nodules [3, 30].

In view of their ubiquity, it is not feasible to biopsy every thyroid nodule discovered with ultrasound. Reasons for limiting thyroid biopsy, which is relatively painless and safe, include the small percentage of malignant lesions, the small number of cases of thyroid cancer in which early diagnosis may actually have an influence, the economic and societal costs, the strain on radiology resources, and the patient uncertainty and anxiety incumbent on a potentially malignant diagnosis. Hence, reliable guidelines for nodules that may not require biopsy have become essential.

Not surprisingly in view of the experience of other authors [31], we concluded that no individual sonographic feature had both high sensitivity and high specificity in the detection of malignancy. Nonetheless, many of these previously described high-risk features, such as calcification, hypoechogenicity, poor definition, and hypervascularity, were found to be absent over and over again in nodules that did not require biopsy.

The persistent combination of some of these common individual ultrasound characteristics, or, more properly, their absence, led us to consider a more pattern-oriented approach, such as that advocated by Reading et al. [23] as an alternative to the analysis of individual features. Those authors described eight typical appearances of commonly encountered benign and malignant nodules, allowing them to separate more than one half of thyroid nodules into those that could be observed versus those requiring biopsy. According to their results, the following four classic patterns necessitate biopsy: 1, a hypoechoic nodule with microcalcifications; 2, coarse calcifications in a hypoechoic nodule; 3, well-marginated, ovoid, solid nodules with a thin hypoechoic halo; and 4, a solid mass with refractive shadowing from the edges, which is believed to occur as a result of fibrosis. The four classic patterns of nodules that did not require biopsy in that series were the following: 1, small (< 1 cm) colloid-filled cystic nodules; 2, a nodule with a honeycomb appearance consisting of internal cystic spaces with thin echogenic walls; 3, a large predominantly cystic nodule; and 4, diffuse multiple small hypoechoic nodules with intervening echogenic bands, which are indicative of Hashimoto's thyroiditis.

Like Reading et al. [23], we found that use of a pattern approach to thyroid nodules is highly sensitive and specific for the presence of benignity. Our patterns differed somewhat from those proposed previously, yet there are definite similarities. Analysis of our data revealed four patterns that were invariably benign at FNA biopsy (Table 5). The most common overall pattern is a nodule with diffuse internal linear cysts, described as spongiform or honeycomb, our type 1 pattern. In our cases, this finding was commonly described as a “puff pastry” pattern similar to the ultrathin layers of flaky pastry in desserts such as napoleons. This pattern was characteristic of colloid nodules or goiter. The only spongiform nodule not classically benign was a single nodule that also was intensely hypervascular. Our type 1 or spongiform nodule consequently is defined as avascular or, occasionally, isovascular in relation to the rest of the gland.

The second pattern (type 2) was a cystic nodule containing a central plug of avascular colloid, similar to the previously described small or large cyst patterns [23]. In our initial analysis of individual features, size of cyst was deemed insignificant. Important, however, was the characterization of the plug as avascular and puff pastry. All of these nodules were also colloid nodules. If the cystic portion of the lesion is subtracted visually, a type 1 spongiform nodule remains. The third pattern (type 3), or giraffe pattern, was characterized by globular areas of hyperechogenicity surrounded by linear thin areas of hypoechogenicity, similar to the two-tone blocklike coloring of a giraffe. This pattern was quite characteristic of Hashimoto's thyroiditis. A variation of this pattern is our type 4 “white knight,” or hyperechoic, nodule, which was found commonly to be a regenerative nodule of Hashimoto's thyroiditis.

Analysis of our other patterns revealed more variability in final cytologic findings (Table 6). Such nodules included both insignificant and significant lesions with such variability that prediction before biopsy was not reliable. These nodules had the four biopsy-recommendation patterns described earlier, such as isoechoic nodule with a surrounding halo or refractive edges, which came to be simplified in our series as isoechoic nodules with or without a halo (types 7 and 8). A hypoechoic nodule with or without central microcalcification or with central macrocalcification in other series [25, 26, 32], for which biopsy was recommended, was the most worrisome pattern (type 6) in our study.

We identified other common patterns, including the type 5 “red light” pattern, or an intensely hypervascular lesion that on Doppler images glowed like a red stoplight. This pattern was commonly seen in lesions with abundant cellularity, including, commonly, follicular neoplasms and, less commonly, hyperplastic nodules and carcinoma. Other nodule types included type 9 ring-of-fire nodules with intense peripheral vascularity and nodules described as other (type 10), which did not fit any of the classic patterns. Calcification, although commonly seen in nodules requiring biopsy, was never seen as an isolated finding. The likelihood of benignity of these nodules (type 5-10) ranged from 60% (type 9, ring of fire) to 91% (type 10, other). Because of this lack of predictability, we believed that these nodules should be considered for FNA biopsy.

The limitations of our study are related to the fact that most of the diagnoses were based on cytologic rather than histologic findings, the retrospective nature of the study, and the fact that nodule characterization was dependent on only two observers. The readers were blinded to the cytologic results at the time of nodule characterization. The period 2005-2007 was chosen to minimize the potential for recall bias. To answer our concerns with respect to these limitations, we are preparing a study in which we train radiologists with varying degrees of experience in this pattern approach. A series of consecutive thyroid biopsies will be chosen prospectively in the weeks before their performance, and the images will be shown to these readers, who will decide whether biopsy should be performed. Analysis of interobserver variability for assigning nodules to a specific pattern will be analyzed, as will the characterizations with final cytologic result.

We conclude that biopsy of a large number of thyroid nodules (in our study, 61%) can be avoided when a pattern approach to nodule characterization is used. Specific morphologic patterns are highly predictive of benignity. Specifically, a nodule that has a uniform nonhypervascular spongiform appearance, is a cystic lesion with a colloid clot, has a giraffelike pattern, or is diffusely hyperechoic can be observed rather than biopsied. If, conversely, a nodule does not correspond to one of these four patterns, according to our data biopsy should be performed regardless of the individual features or pattern of the nodule.

Address correspondence to J. Bonavita ([email protected]).