Validation of Diagnostic Performance and Interobserver Agreement of DTD-TIRADS for Diffuse Thyroid Disease on Ultrasound: A Single-Center Study

Abstract

OBJECTIVE. This retrospective study aimed to investigate the capability of the already-proposed thyroid imaging reporting and data system for detecting diffuse thyroid disease (DTD-TIRADS) on ultrasound (US) by assessing interobserver agreement and diagnostic performance. MATERIALS AND METHODS. A total of 180 patients who underwent thyroid US before thyroid surgery were included. Three radiologists blinded to the pathologic and serologic data independently categorized the US features according to a four-category DTD-TIRADS classification system. On the basis of the pathologic results of thyroid parenchyma, diagnostic performance values were calculated using ROC curve analyses. Interobserver agreements of each US feature and DTD-TIRADS category among the three radiologists were also assessed. RESULTS. Of the 180 patients, 143 (79.4%) had normal thyroid parenchyma and 37 (20.6%) had diffuse thyroid disease (DTD). The areas under the ROC curve for DTD were not significantly different among the three radiologists: 0.876 (95% CI, 0.819-0.920) for radiologist 1, 0.883 (95% CI, 0.827-0.926) for radiologist 2, and 0.861 (95% CI, 0.8010.908) for radiologist 3 (p >.05). The cutoff for the diagnosis of DTD was category III DTD-TIRADS. The sensitivity, specificity, and accuracy of DTD-TIRADS for detecting DTD were 86.5%, 81.1%, and 82.2% for radiologist 1; 86.5%, 83.2%, and 83.9% for radiologist 2; and 83.8%, 82.5%, and 82.8% for radiologist 3, respectively. Interobserver agreement of DTD-TIRADS categorization was almost perfect (K = 0.81). CONCLUSION. DTD-TIRADS has high diagnostic performance and almost-perfect interobserver agreement. Thus, DTD-TIRADS can be considered to be an effective classification system for diagnosing DTD.