Incidental abnormalities of the thyroid gland are commonly encountered by the radiologist, at a rate of 16% on cross-sectional imaging and 27% on sonography [1, 2]. Clinically detected thyroid nodules, including those detected on physical examination or because of patient symptoms, are benign in most patients (90-95%) [3]. The incidentally detected lesions of the thyroid found initially on cross-sectional imaging are a distinct and clinically silent group the significance of which is uncertain. The radiologist encountering one of these lesions may use several strategies, ranging from recommending further imaging and clinical evaluation for every lesion to ignoring every lesion. An appropriate management algorithm requires knowledge of the imaging and demographic features that may predict malignancy and the proportion of these lesions that may be malignant.

The purpose of this study was to determine the prevalence of malignancy in these incidental abnormalities of the thyroid gland detected on CT and to correlate CT with sonography and pathology results to determine if any CT appearance distinguishes malignant from benign lesions.

This study was approved by our institution's human subjects research committee and was conducted in a manner compliant with the Health Insurance Portability and Accountability Act of 1996 (HIPAA). We searched our department's computerized clinical database for all thoracic and cervical CT scans in which a new abnormality was incidentally identified in the thyroid gland from 1998 to 2001. A second list of all diagnostic thyroid sonographic examinations was then created, including examinations from 1998 to 2001 inclusive, plus the first 6 months of 2002. A list of patients whose first thyroid sonographic examination was scheduled after the CT was then extracted from the combination of both lists, yielding 358 patients. One hundred twenty-eight patients were excluded because the initial CT was prompted by a suspected abnormality of the thyroid gland or because the requisition or a detailed chart review revealed a history of thyroid disease, thyroid cancer, or surgery. This yielded a final study population of 230 patients (159 women and 71 men; mean age, 64.5 ± 14.1 years; range, 29-94 years).

All CT examinations were performed with HiSpeed or LightSpeed scanners (GE Healthcare). Scans were obtained at 2.5- to 5-mm collimation throughout the neck or thorax with or without contrast administration. In contrast-enhanced studies, 100 mL of nonionic contrast medium was administered IV using a power injector. Sonography of the thyroid gland was performed after CT, with a mean delay of 97 days (range, 0-1,167 days) between the two examinations. Sonography was performed with a Logiq sonography machine (GE Healthcare) using a high-frequency linear transducer.

Previous studies have shown a high incidence of focal incidental thyroid abnormalities among imaging studies performed for other purposes (16%) [2]. The radiologist encountering one of these lesions is faced with a dilemma about how best to manage this incidental finding. An appropriate management algorithm should balance sensitivity in detecting malignancy against the costs, which are considerable, to the patient and to society of an additional evaluation [2]. The purpose of our study was to determine the rate of malignancy among these incidental abnormalities and to determine whether imaging characteristics on CT could help define which patients with incidental thyroid lesions need further diagnostic evaluation with sonography or biopsy.

No distinguishing feature was seen on CT that could identify an incidental lesion as malignant. A few trends could be identified that may warrant suspicion, however, such as punctuate calcifications, larger size, or younger patients. For instance, we found a significant correlation between punctuate calcifications on CT and microcalcifications on sonography. Microcalcifications in a thyroid lesion constitute a sonographic pattern that has been correlated with malignancy [7-12]. Our results showed a similar trend toward this association. However, no significant correlation existed between those punctate calcifications on CT and malignant or potentially malignant histology (p = 0.72) in those patients who had a biopsy or resection. In counterdistinction, a previous study by Ishigaki et al. [13] revealed a relatively poor correlation between CT and sonography for the imaging finding of microcalcifications (κ = 0.32).

Our results also showed that benign nodules on CT were significantly smaller than malignant and potentially malignant nodules (p = 0.01). Nodules greater than 2.5 cm were significantly more likely to be malignant (p = 0.03). However, considerable size overlap exists between the two groups, limiting the usefulness of this difference when applied to the interpretation of a particular scan.

Patient demographics may be the single most important determinant of the significance of a lesion detected on CT. In this study, patients 35 years or younger were significantly more likely to harbor malignancy (p < 0.01). This finding is consistent with the known increasing prevalence of benign lesions with increasing age [3]. Such patient information can therefore guide the radiologist's recommendation after the detection of an incidental abnormality. Although the clinical literature suggests that a nodule in a man is more likely to be malignant than one in a woman [3], we did not find this correlation in our results.

In terms of correlation between CT and sonographic findings of thyroid lesions, we found that the CT appearance of incidental thyroid lesions did not correlate well with the sonographic appearance. This contrasts with an early study by Radecki and colleagues [14], which indicated that CT and sonography yielded comparable information in most cases (80%); however, that early study used an earlier generation sonography scanner and, compared with optimized CT evaluations, focused only on the thyroid gland. Our results concur with an early study by Stark et al. [15] that revealed increased sensitivity of sonography relative to CT in the detection of thyroid lesions correlated with intraoperative palpation. In addition, in our study there were no features on CT to differentiate simple cysts from complex cystic or solid nodules. A simple density threshold measurement on CT (in Hounsfield units) could not distinguish a cyst from a nodule. This may be partly due to artifacts on CT, including volume averaging, the small size of a lesion, or variations in contrast administration and timing.

Although abnormalities of the thyroid gland can be detected on CT, sonography provides important additional information that may be useful in guiding further clinical management. This additional information includes the presence of additional coexisting nodules, which were seen in 93/230 patients (40.4%). High-frequency sonography also provides greater spatial resolution than CT and affords improved evaluation of nodule architecture and consistency. Current management of multinodular glands, although still debated, emphasizes that the decision should be based on the sonographic features of each nodule [16]. Therefore, knowledge of additional nodules may influence the decision to biopsy. Our findings are concordant with previously published data that suggest that sonography is preferable for the evaluation of solid-component homogeneity [13]. In addition, published recommendations suggest that simple cystic lesions do not require biopsy [3, 16]. In a small minority of patients (10 patients), the subsequent sonographic examination showed normal findings; in those cases, the apparent CT abnormality was either incorrectly characterized as an abnormality on CT (because of an adjacent structure or misinterpretation of a contour irregularity as a nodule) or inaccessible to sonography (because of a substernal location).

Sonography also provides better accuracy than CT in measuring lesions in the thyroid gland. The overall trend of increasing concern for malignancy in larger lesions is echoed in the Society of Radiologists in Ultrasound (SRU) consensus statement on thyroid nodules [16], which uses a graded threshold for consideration of thyroid nodules based on size and sonographic characteristics. Using this scheme, the smallest nodules (≥ 1.0 cm) require a strong sonographic feature of malignancy (microcalcifications), whereas larger lesions (≥ 2.0 cm) can have less concerning features (mixed solid and cystic or predominantly cystic) to consider biopsy. The absence of a distinct size threshold to differentiate benign from malignant, as well the possibility of malignancy in lesions even smaller than 1.0 cm, is consistent with a study by Papini et al. [17] in which an equal frequency of malignancy was observed in nodules greater than and those less than 10 mm as measured on sonography.

In our calculation of the overall rate of malignant and potentially malignant lesions, we assumed that all nonbiopsied lesions were benign, and we used the total number of patients in the series as a denominator in the calculation of the prevalence of malignancy among incidental thyroid lesions. This method countered the potential bias of considering only lesions that were biopsied, which would have likely represented a highly selected group with a much greater expectation of malignancy. In fact, many of the patients who did not undergo biopsy may have been lost to follow-up or had an evaluation elsewhere; those patients were counted as having benign lesions, although a portion of them may have harbored malignant or potentially malignant lesions. This possibility suggests that our calculation may underestimate the true prevalence of malignant or potentially malignant lesions among the incidentally detected abnormalities of the thyroid gland detected on CT.

Despite the assumption made in our calculation, we identified a surprisingly high number of malignant or potentially malignant lesions among incidental abnormalities initially detected on CT. This relatively high fraction necessitates careful consideration of further imaging or evaluation in all patients in whom incidental lesions are detected. The presence of malignancy in two patients with apparent multifocal abnormalities of the thyroid gland on CT further indicates that even a multinodular gland necessitates further evaluation [17]. The rate of malignant and potentially malignant lesions stands in distinction to the complete absence of malignant or potentially malignant lesions in the series described by Yousem et al. [2], although their series included pathologic correlation in only four of 36 patients. In contrast, previous studies of “incidental” lesions of the thyroid on sonography have actually been based on selected referral populations initially referred for a finding on sonography; those studies have revealed malignancy rates of 12% [18] and 28.8% [12].

Our study has several limitations based primarily on its retrospective design. It is impossible to draw conclusions about patients who may have had incidental lesions seen on CT who did not (for any reason) receive further evaluation with sonography, because these patients were not included in the study population. The inclusion of only patients who underwent subsequent sonography may have introduced a bias because the clinical factors that led to the referral for subsequent sonography have not been included in the evaluation. However, it has been routine practice at our institution to refer all patients with an incidental lesion of the thyroid gland for a dedicated sonographic examination of the thyroid gland, thereby reducing the magnitude of this selection bias in our study population.

A second source of potential bias is the dependence on the initial CT interpretation to detect an abnormality of the thyroid gland; certainly, a number of incidental lesions were probably not detected during the initial interpretation and therefore were not included in our analysis; one might hypothesize that these missed lesions tended to be smaller, located at the periphery of the gland, or isodense relative to the thyroid gland. It is difficult to say how the exclusion of these missed lesions might affect the final result. In addition, the study population was older and more likely to have concurrent health problems than the general population because all were being imaged for other reasons. However, our study group probably represents a population that might be encountered in any large metropolitan setting.

Definitive surgical excision was not available in all cases, either because patients were lost to follow-up, refused surgery, or pursued subsequent treatment at a different institution. So as not to bias our results by excluding these patients entirely, the category of “potentially malignant” was applied to microfollicular or predominantly microfollicular lesions without definite surgical excision. Although the diagnosis and management of these lesions are debated in the clinical literature [4-6], the current standard is to recommend resection of hypofunctioning or eufunctioning microfollicular lesions because the possibility of follicular carcinoma can be excluded only through the evaluation of architectural changes such as vascular or capsular invasion.

The incidental nodules of the thyroid were detected on CT performed for a variety of reasons. This entailed the use of different scanning parameters, including differing fields of view and slice thicknesses. These parameters may have altered the ability to accurately measure thyroid nodules. Contrast enhancement was also variable: Some studies were unenhanced and others used variable delays between contrast injection and imaging. Although previous studies have shown improved contrast between thyroid lesions and normal gland parenchyma on unenhanced or latephase contrast-enhanced studies (relative to early-phase contrast enhancement) [13], it remains to be determined whether the enhancement characteristics of a lesion (which were not assessed in our study) may prove useful in the differentiation of incidental thyroid lesions.

In conclusion, a relatively high number of thyroid abnormalities are incidentally detected on CT that are malignant or potentially malignant. The clinical literature indicates that most palpable thyroid nodules are benign (90-95%) [3, 19]. However, in our study, we found a 3.9% prevalence of malignancy (95% CI: 1.8-7.3%) and a 7.4% prevalence of malignant potential (95% CI: 4.4-11.6%) among incidental lesions of the thyroid gland. Considered as a group, these findings yield an overall rate of 11.3% for malignant or potentially malignant lesions among incidentally detected thyroid lesions on CT.

Unfortunately, we did not find any distinct CT feature that invariably distinguishes benign from malignant lesions when correlated to sonographic appearance or histopathology. Patient demographics may be the most useful tool for the radiologist because patients 35 years or younger were significantly more likely to harbor malignancy in an incidental thyroid gland abnormality. CT tends to underestimate the true number of nodules and cannot reliably distinguish cystic from solid nodules; the adjunctive information afforded by the sonography examination suggests that evaluation with sonography should be performed after detection of an incidental abnormality of the thyroid on CT. Although sampling of these incidental lesions may be necessary on the basis of the results of the subsequently performed sonographic or clinical evaluation, referring a patient for sampling based on the CT findings alone would be imprudent and premature, particularly when a noninvasive examination (sonography) is readily available that can provide useful adjunctive information that will inform the decision to pursue biopsy. Our results suggest that every incidental abnormality of the thyroid detected on CT deserves additional clinical or imaging evaluation to exclude the possibility of malignancy. The optimal form of this subsequent evaluation, from the perspective of both patient care and cost-effectiveness, should be the subject of future study.

Address correspondence to S. K. Shetty ([email protected]).